티스토리(Tistory) 개설 · 애드센스 부착 · 원클릭 발행 스튜디오

\n This note is a personal analysis record and is not investment advice or a solicitation to buy or sell any security. As of the writing date, I do not hold shares or positions in any of the companies discussed.\n

\n

\n Korea & Asia Semiconductors · DART 1H26 · TWSE 2330 · SEC 10-Q · 2026-11-01\n

\n

\n Executive Thesis: Auditing four balance-sheet and cash-flow leading indicators across SK Hynix (DART 1H report & 3Q IR deck), TSMC (3Q Management Report & FS), and Tesla (3Q 10-Q)—plus the exact IR breakdown of SK Hynix's -32.6T KRW 3Q26 non-operating loss (11.63T FX loss and 22.92T investment asset loss).\n

\n \n \n

Among the verified 3Q26 filings, Tesla was the only company where depreciation grew faster than revenue, while SK Hynix's 3Q IR deck explains the -32.6T KRW non-operating swing behind its net profit drop.

\n

As of 2026.10.30 (Source: SK Hynix DART 1H26 Report & 3Q26 IR Deck, TSMC 3Q26 FS/Management Report & Tesla 3Q26 10-Q) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

In my earlier filing-footnote note, I wrote that footnotes and cash flow statements matter more than headline operating income. Here I apply that lens to the semiconductor and AI hardware capital cycle across four specific accounting lines: depreciation & amortization (D&A), construction-in-progress (CIP), contract liabilities (customer advances/deferred revenue), and inventory days across TSMC, SK Hynix, and Tesla [1][4][10][12].

\n\n

Not all four metrics are disclosed through September 30 for every company: as of October 30, SK Hynix's full 3Q26 quarterly report (10-Q equivalent on DART) has not yet been filed, so I read 3Q26 D&A and inventory from its 3Q26 IR presentation (through September 30) and read CIP and contract liabilities from its DART Semi-Annual Report (as of June 30, 2026) [10][12]. TSMC's 3Q26 consolidated financial statements and management report disclose D&A and inventory days, but do not break out CIP or contract liabilities as separate lines [1][2].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Company & Filing	Verified CapEx-Cycle & Below-OP Figures
TSMC (2330.TW, 3Q26)	D&A +10.9% QoQ vs. revenue +17.6% QoQ (-6.7%p gap, signal off); Inventory days fell from 87 days to 82 days [1][2]
SK Hynix (000660.KS, 3Q26 & 1H26 DART)	3Q D&A +21.8% YoY vs. revenue +327.9% YoY (+7.6% vs. +31.9% QoQ, signal off); 1H26 CIP 20.83T KRW (-4.2% vs. year-end); 1H26 Contract Liabilities ~523.5B KRW (+10.4%) [10][12]
SK Hynix Below-OP Line (3Q26 IR Deck)	3Q non-operating loss ~-32.6T KRW (vs. ~+62.2T in 2Q26): driven by 11.63T KRW FX loss + 22.92T KRW investment-asset loss (34.55T KRW combined) [9][12]
Tesla (TSLA, 3Q26 10-Q)	D&A/impairment +14.4% QoQ vs. revenue -0.7% QoQ (Note 5 D&A $1.58B, +17.0% YoY vs. revenue -0.2% YoY, signal ON); CIP $11.211B (+31.4% QoQ); Deferred revenue $4.661B (+36.0% QoQ); Ex-tariff OPM ~0.95% (vs. 2.56% rep.) [4][5]

\n\n

1. Hypothesis 1: Does D&A Outpacing Revenue Signal a Margin Peak?

\n\n

Depreciation begins only when a completed facility or tool is placed into service. My first hypothesis is that the quarter in which D&A growth overtakes revenue growth serves as an early peak-margin signal (falsified if operating margins continue rising in subsequent quarters):

\n

\n
• TSMC (Signal Off): In 3Q26, D&A rose 10.9% QoQ while reported revenue rose 17.6% QoQ (-6.7%p gap) [1][2]. Even with the signal off, gross margin dipped from 67.7% to 66.4% on initial 2nm ramp dilution—showing that single-node ramp dilution can lower margins before aggregate D&A outpaces revenue.
\n
• SK Hynix (Signal Off): In the 3Q26 IR presentation cash flow table, D&A (including intangible and idle-asset amortization) rose 21.8% YoY while revenue surged 327.9% YoY (and QoQ D&A rose 7.6% vs. 31.9% revenue growth; in 2Q26 YoY D&A rose 17.0% vs. 256.8% revenue growth), while OPM climbed from 76% to 80% [8][10][11][12].
\n
• Tesla (Signal On): Total D&A and impairment rose 14.4% QoQ while revenue dipped 0.7% QoQ (+15.1%p gap), and pure depreciation in Note 5 ($1.58B) rose 17.0% YoY while revenue dipped 0.2% YoY (after being off in 2Q26 at +19.1% D&A vs. +25.5% revenue YoY) [4][5][6]. While reported GAAP OPM rose from 1.41% in 2Q26 to 2.56% in 3Q26 (vs. 5.66% in 4Q25, 4.20% in 1Q26, and 5.78% in 3Q25), stripping out the Note 11 IEEPA tariff refunds ($786M total COGS reduction across $526M Energy and $260M Auto, minus $339M Energy customer refund) lowers 3Q26 operating margin to ~0.95%—below 2Q26's 1.41% [4][5].
\n

\n\n

2. Hypothesis 2: Construction-in-Progress (CIP), Contract Liabilities & Inventory Days

\n\n

\n
• SK Hynix DART 1H26 CIP & Utilization Nuance: As of June 30, 2026, SK Hynix's CIP stood at 20.83 trillion KRW (-4.2% vs. year-end) while 2Q D&A rose YoY [10]. In the 3Q26 IR deck, DRAM bit shipments rose ~10% QoQ while NAND bit shipments edged down slightly [12]. Note for foreign readers of Korean DART reports: the \"100% average utilization rate\" printed in SK Hynix's 1H26 report measures operating hours divided by available hours (and production capacity of 24.15T KRW is printed equal to actual cost-basis output across all reported periods), so it cannot be used as a physical wafer-volume utilization metric [10].
\n
• Tesla CIP vs. Capacity Expansion: Tesla's CIP jumped 31.4% QoQ to $11.211 billion at September 30, 2026, even as Texas Model Y installed capacity stepped up from >250,000 to >375,000 units/year and Texas Megapack moved from commissioning to production [4][5][7]—showing that when new projects enter CIP faster than completed lines exit, waiting for net CIP to fall can miss capacity additions.
\n
• Contract Liabilities vs. LTAs: Tesla's current deferred revenue rose 36.0% QoQ to $4.661 billion [5]. At SK Hynix, 1H26 contract liabilities rose 10.4% from year-end to ~523.5 billion KRW—which is actual cash prepayments received, distinct from the multi-year LTA volume commitments (>50% of 2027–2028 capacity) discussed on the call [8][10].
\n
• Inventory Days: While raw inventory balances rose sequentially at all three companies, TSMC's inventory turnover days actually fell from 87 days in 2Q26 to 82 days in 3Q26 on higher wafer shipments [1].
\n

\n\n

3. Below the Operating Line: Explaining SK Hynix's -32.6T KRW Non-Operating Swing

\n\n

In my earlier SK Hynix note, I flagged that 3Q26 net profit (39.78T KRW) fell 58% QoQ even as operating profit rose 38% QoQ to 83.33T KRW (80% OPM), and noted that the press release alone did not explain the gap. Checking the DART preliminary earnings table alongside page 6 of SK Hynix's FY2026 Q3 IR presentation resolves the question [9][12]:

\n

\n
• Subtracting operating profit from pre-tax income shows that net non-operating income swung from ~+62.2 trillion KRW in 2Q26 to ~-32.6 trillion KRW in 3Q26 (Calc. from DART table) [9].
\n
• In the 3Q26 IR presentation, management disclosed the two items driving that non-operating loss: an 11.63 trillion KRW FX-related loss from exchange-rate decline and a 22.92 trillion KRW investment-asset valuation loss (totaling 34.55 trillion KRW, meaning remaining net non-operating items were slightly positive) [12]. I will check the specific investment assets behind the 22.92T KRW loss once the full 3Q26 quarterly report footnotes are filed on DART [3].
\n

\n\n

Sources

\n

\n
1. TSMC 3Q26 Management Report (PDF, pp. 1–2, 4), 2026.10.15.
\n
2. TSMC 3Q26 Consolidated Financial Statements (FS.pdf, pp. 2, 4), 2026.10.15.
\n
3. TSMC SEC EDGAR Form 6-K Filing Index, accessed 2026.10.30.
\n
4. Tesla Q3 2026 Form 10-Q (Notes 1, 5 & 11), 2026.10.22.
\n
5. Tesla Q3 2026 Update (Form 8-K Exhibit 99.1), 2026.10.21.
\n
6. Tesla Q2 2026 Form 10-Q (Note 5), 2026.07.23.
\n
7. Tesla Q2 2026 Update (Form 8-K Exhibit 99.1), 2026.07.22.
\n
8. SK Hynix Newsroom: 3Q 2026 Business Results, 2026.10.28.
\n
9. DART SK Hynix Preliminary 3Q26 Earnings Fair Disclosure, 2026.10.28.
\n
10. DART SK Hynix 2026 Semi-Annual Report (1H26 CIP, Utilization & Contract Liabilities), 2026.08.14.
\n
11. SK Hynix Newsroom: 2Q 2026 Business Results, 2026.07.29.
\n
12. SK Hynix FY2026 Q3 Earnings IR Presentation (PDF, pp. 4, 6, 16, 19–20), 2026.10.28.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/capex-cycle-4-signals-skhynix-tsmc-tesla-3q26/\n

\n

\n

\n KOSPI & KOSDAQ Filing Cross-Check · KOSPI: 298040·353200·009150 | KOSDAQ: 064760·232680·396470·348350 · 2026-11-01\n

\n

\n Executive Thesis: Foreign investors in Seoul often concentrate exclusively in Samsung Electronics and SK Hynix. Auditing seven October 29–30 DART filings across KOSPI power transformers and package substrates (Hyosung Heavy 298040.KS, Daeduck 353200.KS, Samsung Electro-Mechanics 009150.KS) and KOSDAQ fab consumables and wafer robotics (TCK 064760.KQ, Raon Robotics 232680.KQ, Wort 396470.KQ, Withtech 348350.KQ) maps where physical AI CapEx is translating into 14%–27% operating margins.\n

\n \n \n

While foreign institutional flows in Korea crowd into the two HBM mega-caps, October 29–30 DART filings show operating profit surging +259.4% YoY at KOSPI substrate maker Daeduck Electronics and +596.1% QoQ at KOSDAQ wafer-transfer robot specialist Raon Robotics.

\n

As of 2026.10.31 (Source: Korean FSS DART K-IFRS Preliminary 3Q26 Earnings & Single-Sales Contract Filings filed 2026.10.29–2026.10.30) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

When global investors look at the Korea Exchange (KRX), they frequently treat KOSPI as a two-stock memory proxy (Samsung Electronics 005930.KS and SK Hynix 000660.KS) and dismiss KOSDAQ as a retail-driven biotech and small-cap index. That framing misses how South Korea's industrial and semiconductor ecosystem is actually structured: KOSPI houses the heavy electrical grid transformers, multilayer ceramic capacitors (MLCCs), and flip-chip ball grid array (FC-BGA) substrates required to power AI clusters, while KOSDAQ houses the specialized cleanroom equipment, silicon carbide (SiC) etch consumables, and vacuum wafer-transfer robotics that go inside Samsung's and SK Hynix's fabs.

\n\n

To test whether upstream memory CapEx and global data-center buildouts are flowing down into the broader Korean market, I audited seven primary K-IFRS preliminary earnings releases (영업(잠정)실적 공정공시) and mandatory single-sales supply contracts (단일판매ㆍ공급계약체결) filed on DART between October 29 and October 30, 2026 across three KOSPI issuers and four KOSDAQ issuers [1][2][3][4][5][6][7].

\n\n\n

Left: 3Q26 operating margin comparison across KOSPI (Hyosung Heavy 298040.KS, Daeduck Electronics 353200.KS) and KOSDAQ (TCK 064760.KQ, Raon Robotics 232680.KQ). Right: 3Q26 YoY operating profit growth from official DART preliminary filings [1][2][4][5].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Issuer (Market & Ticker)	Verified Oct 29–30 DART Filing Figures (3Q26 / Contract)
Hyosung Heavy Industries (KOSPI: 298040.KS)	3Q26 Rev 1,740,635M KRW (+3.18% QoQ, +7.17% YoY; 9M26 4,785,822M KRW, +13.26% YoY); OP 243,455M KRW (-7.88% QoQ, +10.76% YoY; 9M26 660,064M KRW, +35.69% YoY) → 13.99% 3Q26 OPM (Calc.); Pre-tax income 280,237M KRW (+18.31% QoQ, +43.25% YoY); Net profit 226,303M KRW (+31.77% QoQ, +49.11% YoY) [1]
Daeduck Electronics (KOSPI: 353200.KS)	3Q26 Rev 418,115M KRW (+4.3% QoQ, +46.1% YoY; 9M26 1,165,404M KRW, +55.9% YoY); OP 87,871M KRW (+25.0% QoQ, +259.4% YoY; 9M26 209,446M KRW, +941.4% YoY) → 21.02% 3Q26 OPM (Calc. vs. 8.54% in 3Q25); Net profit 54,376M KRW (+133.6% YoY) [2]
Samsung Electro-Mechanics (KOSPI: 009150.KS)	MLCC supply contract for 879,285.4M KRW (US$658.0M at 1,336.30 KRW/USD) = 7.8% of FY25 consolidated revenue (11,314,459.2M KRW); period 2026.12.21–2027.12.21; zero advance payment; counterparty withheld until 2027.12.21 [3]
TCK (Tokai Carbon Korea) (KOSDAQ: 064760.KQ)	3Q26 standalone Rev 110,102M KRW (+16.05% QoQ, +38.95% YoY; 9M26 300,366M KRW, +30.44% YoY); OP 30,219M KRW (+12.55% QoQ, +35.04% YoY; 9M26 85,622M KRW, +31.02% YoY) → 27.45% 3Q26 OPM (Calc.; 9M26 OPM 28.51% Calc.) [4]
Raon Robotics (KOSDAQ: 232680.KQ)	3Q26 standalone Rev 26,535M KRW (+45.9% QoQ, +104.7% YoY; 9M26 58,863M KRW, +51.3% YoY); OP 3,933M KRW (+596.1% QoQ vs. 565M KRW, +259.3% YoY vs. 1,095M KRW) → 14.82% 3Q26 OPM (Calc. vs. 3.11% in 2Q26) [5]
Wort & Withtech (KOSDAQ: 396470.KQ / 348350.KQ)	Wort: 1,638.0M KRW THC contract with Samsung Electronics (10.76% of FY25 rev 15,220.9M KRW; Hwaseong/Pyeongtaek fabs through 2027.05.31) [6]. Withtech: 13,044.0M KRW TMS O&M contract with Samsung Electronics (22.56% of FY25 rev 57,807.0M KRW; 2026.11.01–2027.12.31) [7]

\n\n

1. KOSPI's Second Wave: High-Voltage Transformers, Package Substrates & High-Cap MLCCs

\n\n

On the KOSPI board, three October 30 filings show how AI infrastructure spending extends beyond memory wafers into power delivery and board-level interconnects:

\n

\n
• Hyosung Heavy Industries (298040.KS) — Sustained Double-Digit Grid Margins: Hyosung Heavy is one of Korea's premier ultra-high-voltage transformer and switchgear exporters serving U.S. and global grid upgrades. In 3Q26, consolidated revenue rose 7.17% YoY (3.18% QoQ) to 1,740,635 million KRW, while 9-month cumulative operating profit rose 35.69% YoY to 660,064 million KRW [1]. Even with a slight sequential dip in operating profit (-7.88% QoQ to 243,455 million KRW, or 13.99% OPM Calc. vs. 15.67% in 2Q26), pre-tax continuing income jumped 18.31% QoQ (+43.25% YoY) to 280,237 million KRW and net profit climbed 31.77% QoQ (+49.11% YoY) to 226,303 million KRW [1].
\n
• Daeduck Electronics (353200.KS) — Operating Leverage in Semiconductor Substrates: Semiconductor package substrate makers suffered a multi-year utilization trough in 2024–2025, which makes their operating leverage during a recovery extreme. Daeduck's 3Q26 revenue expanded 46.1% YoY (+4.3% QoQ) to 418,115 million KRW, but operating profit surged 259.4% YoY (+25.0% QoQ) from 24,448 million KRW to 87,871 million KRW—lifting its operating margin by +12.48 percentage points YoY from 8.54% to 21.02% (Calc.) [2]. Across the first nine months of 2026, operating profit is up 941.4% YoY (209,446 million KRW vs. 20,112 million KRW in 9M25) [2].
\n
• Samsung Electro-Mechanics (009150.KS) — A $658M One-Year MLCC Supply Commitment: On October 30, Samsung Electro-Mechanics disclosed a single one-year MLCC supply contract valued at 879,285.4 million KRW (US$658.0 million converted at the October 29 base rate of 1,336.30 KRW/USD), equivalent to 7.8% of its entire FY25 consolidated revenue (11,314,459.2 million KRW) [3]. For foreign investors reading Korean contract disclosures, note two standard KRX mechanics in Section 6 and Section 8 of the filing: there is no upfront deposit (계약금ㆍ선급금 유무: 무), and the global counterparty's name is legally withheld under trade-secret protection until December 21, 2027 [3].
\n

\n\n

2. KOSDAQ's Fab-Floor Specialists: Etch Consumables, Vacuum Robotics & Pyeongtaek Orders

\n\n

Whereas KOSPI's hardware suppliers sell into global data centers and board assemblers, KOSDAQ's semiconductor tier captures the physical utilization and cleanroom expansion of Korean memory fabs:

\n

\n
• TCK (064760.KQ) — 27.45% Operating Margin on Etch-Chamber Consumables: TCK manufactures solid silicon carbide (SiC) focus rings used inside plasma etch chambers for 3D NAND and DRAM. Because SiC rings erode during plasma etching, TCK's revenue tracks actual fab wafer runs rather than greenfield announcements. In its October 29 DART filing, 3Q26 revenue rose 16.05% QoQ (+38.95% YoY) to 110,102 million KRW and operating profit reached 30,219 million KRW (+12.55% QoQ, +35.04% YoY)—a 27.45% operating margin (Calc.), with 9M26 operating margin at 28.51% (Calc.) on 300,366 million KRW of revenue [4].
\n
• Raon Robotics (232680.KQ) — 6x Sequential Operating Profit Inflection: Raon Robotics builds vacuum wafer-transfer robots and automation modules for semiconductor manufacturing equipment. On October 30, it reported 3Q26 standalone revenue of 26,535 million KRW (+45.9% QoQ, +104.7% YoY) and operating profit of 3,933 million KRW (+596.1% QoQ from 565M KRW in 2Q26, and +259.3% YoY), driving standalone OPM from 3.11% in 2Q26 to 14.82% in 3Q26 (Calc.) [5].
\n
• Wort (396470.KQ) & Withtech (348350.KQ) — Direct Contract Visibility Into Samsung's Fabs: Earlier, in our Samsung Electronics 3Q26 note, we observed that Samsung's 9M26 cash PP&E CapEx stood at 56.9T KRW against an 84.4T KRW annual plan (leaving ~27.5T KRW for 4Q26). On October 30, two KOSDAQ fab-equipment specialists disclosed direct contracts with Samsung Electronics: Wort signed a 1,638.0 million KRW contract (10.76% of its FY25 revenue of 15,220.9M KRW) to supply ultra-precision temperature and humidity control (THC) units to Samsung's Hwaseong and Pyeongtaek campuses through May 31, 2027 (settled 100% in cash within 30 days of delivery) [6], while Withtech signed a 13,044.0 million KRW contract (22.56% of its FY25 revenue of 57,807.0M KRW) for industrial emission monitoring (TMS) maintenance at Samsung Electronics from November 1, 2026 through December 31, 2027 [7].
\n

\n\n

3. Falsifiable Hypothesis for 4Q26 KRX Filings

\n\n

My working hypothesis for foreign investors positioning across KRX is that 4Q26 backend fab-equipment and consumable revenue on KOSDAQ (TCK, Raon Robotics, Wort) will accelerate sequentially as Samsung Electronics executes its implied ~27.5T KRW 4Q26 facility investment balance and SK Hynix ramps M15X. This hypothesis will be falsified if TCK's and Raon Robotics' 4Q26 revenue growth turns negative QoQ or if Wort's Hwaseong/Pyeongtaek delivery schedule (currently May 31, 2027) is pushed back via a DART contract amendment (기재정정) [4][5][6].

\n\n

Sources

\n

\n
1. DART Hyosung Heavy Industries (298040.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
2. DART Daeduck Electronics (353200.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
3. DART Samsung Electro-Mechanics (009150.KS) Single-Sales & Supply Contract Disclosure (MLCC), 2026.10.30.
\n
4. DART TCK (064760.KQ) Preliminary 3Q26 Earnings Fair Disclosure, 2026.10.29.
\n
5. DART Raon Robotics (232680.KQ) Preliminary 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
6. DART Wort (396470.KQ) Single-Sales & Supply Contract Disclosure (Samsung Electronics THC), 2026.10.30.
\n
7. DART Withtech (348350.KQ) Single-Sales & Supply Contract Disclosure (Samsung Electronics TMS), 2026.10.30.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/korea-ai-second-derivative-kospi-kosdaq-dart-audit-3q26/\n

\n

\n

\n Korea Corporate Governance & Value-Up · KOSDAQ: 054950·126600 | KOSPI: 000270·012330·008930 · 2026-11-01\n

\n

\n Executive Thesis: Under Korean K-IFRS and Commercial Act Article 343(1), holding treasury shares is not equivalent to retiring them. On October 30, 2026, two KOSDAQ industrials—JVM (054950.KQ, -3.20% of issued shares) and BGF Eco Materials (126600.KQ, -6.00% of issued shares)—filed permanent share cancellations, while on the exact same day KOSPI bellwethers Kia (000270.KS, 190.01B KRW) and Hyundai Mobis (012330.KS, 33.94B KRW) filed to re-issue treasury shares back into circulation for employee wage settlements.\n

\n \n \n

On October 30, 2026, four DART filings on the same afternoon exposed the single most important governance distinction in Korea's Value-Up program: permanent share cancellation under Commercial Act Article 343(1) versus treasury share disposal back into the public float.

\n

As of 2026.10.31 (Source: Korean FSS DART Share Cancellation (주식소각결정) and Treasury Share Disposal (자기주식처분결정) Filings filed 2026.10.30) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

For foreign investors evaluating South Korea's Corporate Value-Up reforms, reading the headline of a \"Treasury Stock (자기주식)\" announcement without checking the exact DART form type is a recurring trap. In the United States, repurchased shares are almost always retired or netted against diluted share count. In South Korea, however, treasury shares sit on the balance sheet as a dormant pool that management can later re-issue into the market (자기주식처분결정)—expanding the active free float and diluting per-share claims—unless the Board passes a formal Share Cancellation Resolution (주식소각결정) under Article 343(1) of the Korean Commercial Act.

\n\n

On Friday, October 30, 2026, two KOSDAQ mid-caps and two KOSPI automotive bellwethers filed treasury-stock disclosures within hours of one another. Putting those eight related DART filings side by side shows why governance-focused capital must audit KOSDAQ and KOSPI filings line by line rather than relying on index membership [1][4][6][8].

\n\n\n

Left: October 30, 2026 KOSDAQ permanent share cancellations as a percentage of total issued shares (BGF Eco Materials 126600.KQ -6.00%, JVM 054950.KQ -3.20%). Right: October 30, 2026 KOSPI treasury share disposals re-entering circulation on November 30, 2026 to settle employee wage agreements (Kia 000270.KS 190.01B KRW, Hyundai Mobis 012330.KS 33.94B KRW) [1][4][6][8].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Issuer & Filing Type (2026.10.30)	Share Count, KRW Value & Float Impact
JVM (054950.KQ, KOSDAQ) Share Cancellation (주식소각결정)	Permanently canceling 387,032 common shares out of 12,089,892 issued shares (-3.20% of issued shares, Calc.); book value 15,902,384,907 KRW (15.90B KRW); cancellation date 2026.11.06; filed alongside 3Q26 OP +30.4% YoY (24.43% OPM Calc.) [1][2][3]
BGF Eco Materials (126600.KQ, KOSDAQ) Share Cancellation (주식소각결정)	Permanently canceling 3,764,752 common shares out of 62,766,899 issued shares (-6.00% of issued shares, Calc.); cost 14,494,295,200 KRW (14.49B KRW, 3,850 KRW/sh Calc.) acquired via Oct 8–27 tender offer; cancellation date 2026.11.10 [4][5]
Kia Corporation (000270.KS, KOSPI) Treasury Disposal (자기주식처분결정)	Re-issuing 1,754,496 treasury shares at 108,300 KRW/sh = 190,011,916,800 KRW (190.01B KRW) on 2026.11.30 to 30,916 employees under the 2026 wage agreement (disposing of 99.86% Calc. of its 1,756,955 held treasury shares) [6][7]
Hyundai Mobis (012330.KS, KOSPI) Treasury Disposal (자기주식처분결정)	Re-issuing 96,136 treasury shares at 353,000 KRW/sh = 33,936,008,000 KRW (33.94B KRW) on 2026.11.30 to company employees under the 2026 collective bargaining agreement (6.19% Calc. of its 1,552,000 held treasury shares) [8]

\n\n

1. Genuine Per-Share Accretion on KOSDAQ: JVM (054950.KQ) & BGF Eco Materials (126600.KQ)

\n\n

Under the proviso to Article 343(1) of the Korean Commercial Act, a company can cancel treasury shares held within distributable profit via a simple Board resolution—reducing the total issued share count without reducing paid-in capital. Two KOSDAQ industrial leaders executed textbook cancellations on October 30:

\n

\n
• JVM (054950.KQ, Subsidiary of KOSPI-listed Hanmi Science 008930.KS): JVM is South Korea's dominant hospital and pharmacy medication-packaging automation exporter. On October 30, JVM filed two simultaneous disclosures: (1) 3Q26 preliminary consolidated earnings showing revenue of 40,545 million KRW (+2.1% YoY, -17.8% QoQ), operating profit of 9,907 million KRW (+30.4% YoY, 24.43% OPM Calc. vs. 19.13% in 3Q25; 9M26 OP 32,722M KRW, +19.6% YoY, 24.18% OPM Calc.), and net income of 9,590 million KRW (+49.9% YoY) [3]; and (2) an unanimous Board resolution (4 of 4 independent directors present) to permanently cancel 387,032 existing treasury shares (3.20% of its 12,089,892 total issued shares, Calc.) carrying a book value of 15,902,384,907 KRW (15.90B KRW) on November 6, 2026 [1][2].
\n
• BGF Eco Materials (126600.KQ, Subsidiary of KOSPI-listed BGF 027410.KS): Even more aggressive in percentage terms, specialty semiconductor gas and engineering-plastics producer BGF Eco Materials resolved on October 30 to permanently cancel 3,764,752 common shares—exactly 6.00% of its 62,766,899 total issued shares (Calc.)—valued at 14,494,295,200 KRW (14.49B KRW) on November 10, 2026 [4][5]. Crucially, the footnote in Item 11 confirms that these 3,764,752 shares were acquired through a public tender offer run from October 8 to October 27, 2026 (settled October 29) for the explicit purpose of immediate retirement [4].
\n

\n\n

2. The Other Side of Korean Treasury Stock: 223.95B KRW of Wage Disposals at Kia & Hyundai Mobis

\n\n

Contrast those KOSDAQ cancellations with two Major Event Reports (주요사항보고서: 자기주식처분결정) filed on the same day by KOSPI heavyweights Kia Corporation (000270.KS) and Hyundai Mobis (012330.KS) [6][8]:

\n

\n
• Kia Corporation (000270.KS): Prior to October 30, Kia held 1,756,955 common shares in treasury (0.45% of shares). In its October 30 filing, Kia's Board approved the disposal of 1,754,496 treasury shares (99.86% of its held treasury stock, Calc.) at the October 29 closing price of 108,300 KRW per share—totaling 190,011,916,800 KRW (190.01B KRW)—to be transferred directly from the company's treasury account to the personal brokerage accounts of 30,916 employees on November 30, 2026 pursuant to the 2026 collective wage agreement [6]. This was filed alongside Kia's 3Q26 preliminary earnings (`31.60T KRW` revenue, `+10.1% YoY`; `1.82T KRW` operating profit, `-30.8% QoQ`, `+24.3% YoY`, `5.75% OPM Calc.`) [7].
\n
• Hyundai Mobis (012330.KS): In an identical structure, Hyundai Mobis resolved on October 30 to dispose of 96,136 treasury shares at 353,000 KRW per share—totaling 33,936,008,000 KRW (33.94B KRW)—on November 30, 2026 to pay employee performance grants agreed in its 2026 collective bargaining settlement [8].
\n

\n\n

3. Analytical Takeaway & Checklist for Foreign Investors

\n\n

When a Korean issuer buys back stock without canceling it, and later distributes 223.95 billion KRW (Calc.) of those treasury shares to employees as wage compensation, the cash outlay at the time of the original buyback was effectively a pre-funded labor expense—not a permanent reduction in share count. Recall from our earlier SK Hynix 3Q26 audit that SK Hynix explicitly excluded the 10 trillion KRW employee-compensation portion of its 20 trillion KRW buyback from its shareholder-return calculation for this exact reason. Foreign investors screening Korea's Value-Up universe should treat DART Form 주식소각결정 (Cancellation)—such as JVM's 3.20% and BGF Eco Materials' 6.00% float reductions on KOSDAQ—as the only verified denominator reduction [1][4].

\n\n

Sources

\n

\n
1. DART JVM (054950.KQ) Share Cancellation Decision (387,032 Shares), 2026.10.30.
\n
2. DART Hanmi Science (008930.KS) Subsidiary Material Management Report (JVM Share Cancellation), 2026.10.30.
\n
3. DART JVM (054950.KQ) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
4. DART BGF Eco Materials (126600.KQ) Share Cancellation Decision (3,764,752 Shares), 2026.10.30.
\n
5. DART BGF (027410.KS) Subsidiary Material Management Report (BGF Eco Materials Share Cancellation), 2026.10.30.
\n
6. DART Kia Corporation (000270.KS) Major Event Report: Treasury Share Disposal Decision (190.01B KRW), 2026.10.30.
\n
7. DART Kia Corporation (000270.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
8. DART Hyundai Mobis (012330.KS) Major Event Report: Treasury Share Disposal Decision (33.94B KRW), 2026.10.30.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/korea-value-up-treasury-share-cancellation-vs-disposal-dart-audit/\n

\n

\n

\n KOSPI & KOSDAQ Earnings & Footnote Audit · KOSPI: 000270·307950·204320 | KOSDAQ: 067160·236200·099320·174900 · 2026-11-01\n

\n

\n Executive Thesis: English wire summaries of Korean 3Q26 earnings typically stop at operating profit (영업이익). Auditing seven October 29–30 DART disclosures across KOSPI's mobility stack (Kia 000270.KS, Hyundai AutoEver 307950.KS, HL Mando 204320.KS) and KOSDAQ's platform, AI security, space, and biotech issuers (SOOP 067160.KQ, Suprema 236200.KQ, Satrec Initiative 099320.KQ, AbClon 174900.KQ) exposes sharp below-the-line swings into pre-tax losses at HL Mando (-41.20B KRW) and Suprema (-17.85B KRW) alongside a +502.7% QoQ operating inflection at Satrec Initiative.\n

\n \n \n

In Korean preliminary earnings disclosures, stopping at operating profit misses the -92.74B KRW non-operating swing that pushed KOSPI's HL Mando into a 3Q26 pre-tax loss and the -23.52B KRW below-the-line swing at KOSDAQ's Suprema—while KOSDAQ satellite maker Satrec Initiative delivered a +502.7% QoQ operating profit breakout.

\n

As of 2026.10.31 (Source: Korean FSS DART K-IFRS Preliminary 3Q26 Earnings & Material Management Filings filed 2026.10.29–2026.10.30) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

When Korean listed companies file their preliminary quarterly numbers (연결재무제표기준 영업(잠정)실적 공정공시) on DART, the standardized table prints four income-statement lines: Revenue (매출액), Operating Profit (영업이익), Continuing Income Before Tax (법인세비용차감전계속사업이익), and Net Income (당기순이익). Because automated English news feeds almost exclusively scrape the first two lines, foreign investors are routinely blindsided when a Korean stock with solid operating margins sells off on a massive gap between Operating Profit and Pre-Tax Continuing Income—or conversely, when a multi-year R&D contract overhang is quietly extinguished via a DART material management filing.

\n\n

To illustrate how to read the full four-line DART table across sectors, I compared seven October 29–30 disclosures spanning KOSPI's automotive hardware/software stack (Kia 000270.KS, Hyundai AutoEver 307950.KS, HL Mando 204320.KS) and KOSDAQ's non-semiconductor tech, aerospace, and biotech cohort (SOOP 067160.KQ, Suprema 236200.KQ, Satrec Initiative 099320.KQ, AbClon 174900.KQ) [1][2][3][4][5][6][7].

\n\n\n

Left: 3Q26 Operating Profit vs. Pre-Tax Continuing Income at KOSPI's Hyundai AutoEver (307950.KS) and HL Mando (204320.KS, showing a -92.74B KRW non-operating swing). Right: 3Q26 Operating Profit vs. Pre-Tax Continuing Income across KOSDAQ's SOOP (067160.KQ), Suprema (236200.KQ, -23.52B KRW non-operating swing), and Satrec Initiative (099320.KQ) [2][3][4][5][6].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Issuer (Market & Ticker)	3Q26 Operating Profit vs. Pre-Tax Continuing Income (DART)
Kia Corporation (KOSPI: 000270.KS)	Rev 31.60T KRW (-4.4% QoQ, +10.1% YoY); OP 1.82T KRW (-30.8% QoQ, +24.3% YoY; 5.75% OPM Calc. vs. 7.96% in 2Q26); Pre-tax 1.82T KRW (-40.6% QoQ, -3.4% YoY); Net profit 1.29T KRW (-44.8% QoQ, -9.6% YoY) [1]
Hyundai AutoEver (KOSPI: 307950.KS)	Rev 1,224.70B KRW (-2.1% QoQ, +16.2% YoY); OP 68.91B KRW (-23.9% QoQ, -2.6% YoY; 5.63% OPM Calc.); Pre-tax 60.18B KRW (-34.3% QoQ, -14.7% YoY); Net profit 44.93B KRW (-35.4% QoQ, -14.9% YoY) [2]
HL Mando (KOSPI: 204320.KS)	Rev 2,320.53B KRW (-7.0% QoQ, 0.0% YoY); OP +51.54B KRW (-51.8% QoQ, -45.3% YoY; 2.22% OPM Calc.), but Pre-Tax Continuing Income swung to -41.20B KRW loss (-92.74B KRW non-op gap, Calc.); Net loss -42.51B KRW [3]
Suprema (KOSDAQ: 236200.KQ)	Rev 32.84B KRW (-18.59% QoQ, +8.20% YoY); OP +5.67B KRW (17.26% OPM Calc.; 9M26 OPM 20.34% Calc.), but Pre-Tax Continuing Income swung to -17.85B KRW loss (-23.52B KRW non-op gap, Calc.); Net loss -12.64B KRW [4]
SOOP (KOSDAQ: 067160.KQ)	Rev 105.69B KRW (+1.8% QoQ, -16.0% YoY); OP 14.52B KRW (+15.0% QoQ, -59.9% YoY; 13.74% OPM Calc. vs. 28.81% in 3Q25); Pre-tax 13.19B KRW (-19.3% QoQ, -65.1% YoY); Net profit 10.58B KRW (+21.0% QoQ, -64.0% YoY) [5]
Satrec Initiative (KOSDAQ: 099320.KQ)	Rev 56.19B KRW (+9.9% QoQ, +11.4% YoY); OP 8.24B KRW (+502.7% QoQ vs. 1.37B KRW, +235.6% YoY; 14.66% OPM Calc. vs. 2.67% in 2Q26); Pre-tax 3.67B KRW (+698.5% QoQ); Net profit turned positive at 3.39B KRW [6]
AbClon (KOSDAQ: 174900.KQ)	Final settlement & full IP assignment agreement with Yuhan Corporation on 2016 TIGIT antibody joint R&D contract; settlement receipt 1.00B KRW (≥10% of prior-year rev); zero remaining obligations or contingent liabilities [7]

\n\n

1. KOSPI's Automotive Stack: Sequential Margin Compression & HL Mando's -92.74B KRW Non-Operating Gap

\n\n

Across the Hyundai-Kia automotive value chain on KOSPI, October 30 preliminary 3Q26 filings showed sequential margin compression from the 2Q26 peak, compounded by below-the-line headwinds:

\n

\n
• Kia Corporation (000270.KS) & Hyundai AutoEver (307950.KS): Kia's 3Q26 revenue dipped 4.4% QoQ (+10.1% YoY) to 31,597,447 million KRW while operating profit fell 30.8% QoQ (+24.3% YoY) to 1,818,286 million KRW, compressing its operating margin from 7.96% in 2Q26 to 5.75% in 3Q26 (Calc.) [1]. Notice that while 3Q26 operating profit rose +24.3% YoY, pre-tax income fell -3.4% YoY (1,822,205M KRW) and net income fell -9.6% YoY (1,285,538M KRW) [1]. Similarly, captive automotive software and cloud integrator Hyundai AutoEver grew 3Q26 revenue 16.2% YoY (-2.1% QoQ) to 1,224,703 million KRW, but saw operating profit slip 23.9% QoQ (-2.6% YoY) to 68,906 million KRW (5.63% OPM Calc.) and pre-tax income decline 34.3% QoQ (-14.7% YoY) to 60,178 million KRW [2].
\n
• HL Mando (204320.KS) — Positive Operating Profit, Negative Pre-Tax Income: Tier-1 chassis and ADAS supplier HL Mando reported 3Q26 revenue of 2,320,530 million KRW (0.0% YoY, -7.0% QoQ) and positive operating profit of 51,538 million KRW (-51.8% QoQ, -45.3% YoY; 2.22% OPM Calc.) [3]. Yet on the very next line of the DART table, continuing income before tax plunged to -41,197 million KRW (a -92,735 million KRW net non-operating deduction, Calc., compared with +92,695M KRW of pre-tax profit in 2Q26), resulting in a controlling-interest net loss of -47,057 million KRW [3]. Because preliminary fair disclosures do not include full financial-statement footnotes, foreign investors should check Note 26/27 (Finance Income/Costs and Other Non-Operating Items) when HL Mando files its statutory Q3 report in mid-November.
\n

\n\n

2. KOSDAQ's Tech Divergence: Suprema's Below-the-Line Swing, SOOP's Margin Reset & Satrec's 6x Operating Breakout

\n\n

On the KOSDAQ board, October 29–30 filings across four distinct technology and biotech niches demonstrated why stock-by-stock filing audits beat top-down index generalizations:

\n

\n
• Suprema (236200.KQ) — 17.26% Core OPM Masked by a -23.52B KRW Non-Operating Swing: AI biometric security leader Suprema generated 32,839 million KRW in 3Q26 revenue (+8.20% YoY, -18.59% QoQ) and 5,667 million KRW in operating profit—a solid 17.26% 3Q26 operating margin (Calc.), with 9-month cumulative operating profit up 7.24% YoY to 21,214 million KRW on 104,301 million KRW of revenue (20.34% 9M26 OPM Calc.) [4]. However, pre-tax continuing income swung from +22,445 million KRW in 2Q26 to -17,851 million KRW in 3Q26 (a -23,518 million KRW non-operating deduction, Calc.), pulling 3Q26 net income to -12,637 million KRW [4]. For cash-rich KOSDAQ exporters holding large USD cash/financial-asset balances, quarter-end FX and fair-value marks frequently dominate quarterly net profit without impairing core operating cash flow.
\n
• SOOP (067160.KQ, formerly AfreecaTV) — Sequential Stabilization After YoY Margin Reset: Live-streaming platform SOOP posted 3Q26 revenue of 105,687 million KRW (+1.8% QoQ, -16.0% YoY) and operating profit of 14,522 million KRW (+15.0% QoQ, -59.9% YoY), representing a 13.74% operating margin (Calc., up from 12.16% in 2Q26 but down from 28.81% in 3Q25) [5].
\n
• Satrec Initiative (099320.KQ) — +502.7% QoQ Operating Leverage in Earth-Observation Satellites: Satellite systems manufacturer Satrec Initiative (a consolidated subsidiary of Hanwha Aerospace) delivered one of the sharpest operating inflections on KOSDAQ on October 29: 3Q26 revenue rose 9.9% QoQ (+11.4% YoY) to 56,188 million KRW, while operating profit jumped 502.7% QoQ (+235.6% YoY) from 1,367 million KRW to 8,239 million KRW, expanding OPM from 2.67% in 2Q26 to 14.66% in 3Q26 (Calc.) and flipping quarterly net income back into the black at 3,387 million KRW (3,921 million KRW controlling interest, +795.2% QoQ) [6].
\n
• AbClon (174900.KQ) — How to Read a Clean Pre-IPO R&D Contract Closeout: On October 30, KOSDAQ antibody developer AbClon filed an Investment-Judgment Material Management Report (투자판단관련주요경영사항) disclosing an amended agreement with KOSPI pharmaceutical leader Yuhan Corporation (유한양행) [7]. Under the agreement, AbClon completed and transferred all remaining rights to the TIGIT-targeted antibody candidate originated under its 2016 pre-IPO joint R&D contract for a final settlement receipt of 1,000,000,000 KRW (1.00B KRW, exceeding the 10% prior-year revenue mandatory disclosure threshold), while explicitly confirming in Item 5(4) that zero further performance obligations or contingent clawback liabilities remain (추가적인 이행 의무나 우발부채는 존재하지 않습니다) [7].
\n

\n\n

Sources

\n

\n
1. DART Kia Corporation (000270.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
2. DART Hyundai AutoEver (307950.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
3. DART HL Mando (204320.KS) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
4. DART Suprema (236200.KQ) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.29.
\n
5. DART SOOP (067160.KQ) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.30.
\n
6. DART Satrec Initiative (099320.KQ) Preliminary Consolidated 3Q26 Earnings Fair Disclosure, 2026.10.29.
\n
7. DART AbClon (174900.KQ) Material Management Report: TIGIT Antibody Joint R&D & Technology Transfer Settlement With Yuhan Corp, 2026.10.30.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/kospi-auto-stack-vs-kosdaq-tech-3q26-op-pretax-divergence/\n

\n

\n

\n Semiconductors & HBM · TWSE: 2330 / NYSE: TSM · 3Q26 · 2026-10-29\n

\n

\n Executive Thesis: Management raised full-year 2026 USD revenue growth guidance to ~44%, saying AI demand still outstrips supply. A filing-by-filing review of 3Q26 revenue ($46.72B), gross margin absorption during the initial 2nm ramp (66.4%), HPC share (68%), and $60B-$64B CapEx discipline.\n

\n \n \n

Management raised full-year USD revenue growth guidance to roughly 44%, noting that AI demand still outstrips supply.

\n

As of 2026.10.28 (Source: TSMC 3Q26 earnings materials released 2026.10.15) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

TSMC has released its 3Q26 results and 4Q26 guidance. I put these notes together because I wanted to look at two things side by side: how fast top-line revenue is growing, and how much initial 2-nanometer (N2) mass-production costs pulled down gross margin during the quarter.

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Item	Disclosed & Guided Figures
3Q26 Revenue	US$46.72B (+16.2% QoQ, +41.2% YoY; above the prior guidance top end of US$45.8B)
3Q26 Profitability	Gross margin 66.4% (-1.3%p QoQ, +6.9%p YoY), operating margin 58.7%
4Q26 Guidance	Revenue US$52.3B–$54.3B (midpoint US$53.3B), gross margin 65%–67% (midpoint 66.0%)
Full-Year 2026 Outlook	USD revenue growth of ~44% (raised from prior guidance of above 40%)
Node & Platform Mix	≤7nm wafer revenue share 79% (3nm 33%, 5nm 28%, 2nm 8%), HPC share 68%
CapEx & Cash Flow	2026 CapEx budget kept at US$60B–$64B (3Q26 spend US$13.84B), free cash flow NT$476.56B

\n\n

Revenue Cleared the Guidance Ceiling, and Gross Margin Stayed Inside the Band

\n\n\n

Left: TSMC quarterly gross and operating margin progression through 4Q26 guidance midpoint. Right: 3Q26 wafer revenue mix by node (≤7nm = 79%). Source: TSMC 3Q26 releases [1][2][3].

\n\n

Third-quarter revenue came in at US$46.72 billion, clearing the US$45.8 billion upper bound of guidance management had given a quarter earlier. That is up 16.2% sequentially and 41.2% from a year ago [1][2].

\n\n

Gross margin was 66.4%, down 1.3 percentage points from 67.7% in 2Q26. Management attributed the sequential dip to initial dilution from 2nm mass production. Even so, gross margin landed comfortably inside the guided 65%–67% band and stood 6.9 percentage points higher than a year earlier (59.5%). Operating margin was 58.7%, down from 60.3% in 2Q26 but well above 50.6% in 3Q25 [2].

\n\n

For the fourth quarter, TSMC guided revenue to US$52.3B–$54.3B. At the US$53.3 billion midpoint, revenue would grow another 14% sequentially. The 4Q26 gross margin guidance midpoint of 66.0% (65%–67% range) is just 40 basis points below the third quarter [1][4].

\n\n

Management also lifted its full-year outlook, raising expected 2026 revenue growth in USD terms from \"above 40%\" previously to roughly 44% (approaching the mid-40% range) [3].

\n\n

Overseas fabs will continue to weigh on gross margin going forward. On the conference call, management explained that margin dilution from overseas fab ramps is expected to be around 2 percentage points in 2026 and widen to 3–4 percentage points per year over the next several years [4].

\n\n

Based on current customer demand visibility, management said AI demand still exceeds supply, and noted that the 2024–2029 AI accelerator revenue CAGR—previously guided in the mid-to-high 50% range—will likely come in higher in practice. CoWoS, TSMC's advanced packaging technology that integrates multiple logic dies and high-bandwidth memory on a single interposer, remains customers' first choice, and the company is preparing larger-substrate packaging at 9x and 14x reticle sizes [4].

\n\n

HPC and 3nm Drove the Revenue Mix Higher

\n\n

High-Performance Computing (HPC) revenue grew 22% sequentially in 3Q26, lifting HPC's share of total revenue from 66% in 2Q26 to 68%. Smartphone revenue also rose 10% quarter-on-quarter, but because it grew more slowly than HPC, its share of total revenue edged down from 22% to 21% [2].

\n\n

By wafer revenue, advanced nodes (7nm and below) accounted for 79% of the total. The 3nm node climbed to 33%, overtaking 5nm at 28% (with 7nm at 10%). Notably, 2nm jumped from 3% of wafer revenue in 2Q26 to 8% in just a single quarter [2][3].

\n\n

Demand Was Strong, Yet the CapEx Budget Stayed Unchanged

\n\n

What stood out to me in this release was capital discipline. Even with demand running ahead of supply, TSMC kept its 2026 annual capital expenditure budget unchanged at US$60 billion to US$64 billion, spending US$13.84 billion during the third quarter (US$40.64 billion cumulative over 1Q–3Q). In NTD terms, 3Q26 operating cash flow was NT$919.77 billion against NT$443.21 billion in capital expenditures, leaving NT$476.56 billion in quarterly free cash flow [2][3][4].

\n\n

I read management's commentary on excess AI accelerator demand and 9x/14x reticle CoWoS preparation as supportive for the high-bandwidth memory (HBM) and advanced back-end packaging ecosystems that get co-packaged alongside those logic dies. Combined with leading-edge pricing power and capital discipline, visibility into core operating profit remains high.

\n\n

What I Will Check Next Quarter

\n

\n
• How quickly TSMC absorbs the rising depreciation burden as 2nm expands its share of wafer revenue
\n
• How quickly initial startup costs at overseas fabs are absorbed into gross margin
\n

\n\n\n\n

Sources

\n

\n
1. TSMC 3Q26 Earnings Release, 2026.10.15.
\n
2. TSMC 3Q26 Quarterly Management Report, 2026.10.15.
\n
3. TSMC 3Q26 Earnings Conference Presentation, 2026.10.15.
\n
4. TSMC 3Q26 Earnings Conference Transcript, 2026.10.15.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/tsmc-3q26-earnings-n2-ramp-cowos-supply-audit/\n

\n

\n

\n Semiconductors & HBM · KRX: 000660 · DART K-IFRS & IR · 2026-10-29\n

\n

\n Executive Thesis: A 100T+ KRW revenue quarter (104.6155T KRW), an 80% operating margin, and a ~100T KRW FY26 shareholder return target all arrived in a single release. Here is how operating profit, net profit, HBM4 crossover, and the 3Q26 board resolutions break down.\n

\n \n \n

A 100-trillion-KRW revenue quarter, an 80% operating margin, and a ~100-trillion-KRW annual shareholder return target all landed in the same release.

\n

As of 2026.10.28 (Source: SK Hynix 3Q26 business results released 2026.10.28) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

SK Hynix's third-quarter revenue crossed 100 trillion KRW in a single quarter for the first time in the company's history. In the same release, management laid out a target of returning roughly 100 trillion KRW to shareholders for FY2026. I am writing this note to separate three things clearly in my own records: how large operating profit was, why net profit fell compared with the prior quarter, and how much of that 100-trillion-KRW return target was actually resolved by the board this quarter [1][2].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Item	3Q26 Summary
Revenue	104.6155 trillion KRW, +32% QoQ
Operating Profit	83.3338 trillion KRW, operating margin 80%
Net Profit	39.7756 trillion KRW, +216% YoY, -58% QoQ
Product Mix	Management expects HBM4 revenue to overtake HBM3E in 4Q26
Long-Term Agreements	Over 50% of total 2027–2028 production capacity already contracted
Balance Sheet	Cash & equivalents 114.5T KRW, borrowings 15.8T KRW, net cash 98.7T KRW
Net Cash Increase	+29.3 trillion KRW vs. end of 2Q26 (Calc.)
FY2026 Return Target	~100T KRW (≥30T KRW cash dividends, ≥60T KRW share buybacks & cancellations)
3Q26 Board Resolutions	14T KRW dividend (20,000 KRW/share), 20T KRW buyback (only the 10T KRW cancellation tranche counts toward the return total)

\n\n

Quarterly Revenue Doubled in Six Months

\n\n

Consolidated 3Q26 revenue was 104.6155 trillion KRW, up 32% from 79.3187 trillion KRW in 2Q26 and up 328% from 24.4489 trillion KRW in 3Q25. Considering 1Q26 revenue was 52.5763 trillion KRW, quarterly revenue doubled in just half a year [1][3][4]. I view that jump as the combined result of AI memory demand, rising shipments of higher-value products, and higher average selling prices.

\n\n\n

Left: SK Hynix quarterly revenue and operating profit progression (3Q25–3Q26). Right: Balance-sheet cash vs. borrowings and net cash expansion. Source: SK Hynix quarterly releases [1][3][4].

\n\n

Operating profit reached 83.3338 trillion KRW—an all-time high in both absolute terms and margin. It rose 38% from 60.5426 trillion KRW in 2Q26 and 632% from 11.3834 trillion KRW in 3Q25. Profit grew faster than revenue [1].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Quarter	Operating Margin (Company Release)
3Q25	47%
4Q25	58%
1Q26	72%
2Q26	76%
3Q26	80%

\n\n

Operating margin stepped up every single quarter to reach 80% in 3Q26—33 percentage points higher than a year ago and 4 percentage points (or 3 percentage points on the rounded print; roughly 3.33 percentage points on my unrounded calculation) above 2Q26 [1][3][4]. A richer mix of high-value memory products was the main driver lifting profitability.

\n\n

Net Profit Dropped Quarter-on-Quarter

\n\n

Net profit was 39.7756 trillion KRW, for a net margin of 38%. While that is up 216% from 12.5975 trillion KRW a year ago, it fell 58% compared with 93.9226 trillion KRW in 2Q26. Over the same period, operating profit rose 38%, so core operating profit and bottom-line net profit moved in opposite directions [1][3].

\n\n

In 2Q26, net profit (93.9226T KRW) was actually larger than operating profit (60.5426T KRW). From the earnings release tables I reviewed for this note alone, I could not verify the exact non-operating items that created that gap in 2Q26. To check actual cash inflow after working-capital movements, I plan to open the full cash flow statement and inspect operating cash flow separately.

\n\n

HBM4 Shipments Ramped in Earnest

\n\n

In DRAM, shipments of HBM4—the next-generation high-bandwidth memory attached to AI accelerators—ramped up in earnest after mass shipments began in 2Q26. Management said customer demand for HBM4 is rising rapidly, so the company plans to maximize capacity at its Cheongju M15X fab through year-end and expects HBM4 to overtake HBM3E in both shipment volume and revenue in 4Q26. Even during a product-generation transition quarter, operating margin reached 80% [1][2].

\n\n

Server memory expanded alongside HBM. Sales of high-density server DDR5 modules for data centers rose sharply in DRAM, and shipments of 321-layer TLC enterprise SSDs (eSSD) grew in NAND. As AI spreads into enterprise workloads like coding, customer service, and search, replacement demand for standard server DRAM and power-efficient enterprise SSDs is expanding together with HBM.

\n\n

More Than Half of 2027–2028 Capacity Is Already Under Contract

\n\n

Management disclosed that it has signed long-term agreements (LTAs) with major global cloud customers covering more than 50% of its total 2027–2028 production capacity based on already secured volumes. Pricing is structured to reflect market conditions within an agreed band. Because volumes are locked in while prices partially track market conditions within a range, I do not read these contracts as locking in future profit margins [1].

\n\n

The company also projected that memory demand will exceed supply through 2030, citing the rising technical difficulty of leading-edge nodes and the long lead times required to build new fabs. Because that is management's projection, actual supply-demand balance could still shift if competitors accelerate capacity additions or customers slow their investment pace.

\n\n

Net Cash Swelled, Though Major Outflows Are Also Scheduled

\n\n

Quarter-end cash and cash equivalents reached 114.5 trillion KRW, up 26.5 trillion KRW in three months from 88.0 trillion KRW at the end of 2Q26. Over the same span, borrowings decreased by 2.8 trillion KRW (from 18.6T to 15.8T KRW). Subtracting borrowings from cash leaves 98.7 trillion KRW in net cash (+29.3T KRW QoQ, Calc.) [1].

\n\n

With the Yongin Phase 1 fab scheduled to begin operations next year, management expects 2027 capital expenditures to increase substantially compared with 2026. Net cash is ample today, but dividend payouts, share buyback settlements, and Yongin fab CapEx are all lined up—so I plan to keep tracking quarterly cash inflows and outflows closely.

\n\n

~100 Trillion KRW Is the Full-Year Target; This Quarter's Board Resolutions Are Specific

\n\n

For FY2026, SK Hynix is pursuing total shareholder returns of roughly 100 trillion KRW, aligned with its policy of returning at least 50% of free cash flow (operating cash flow minus CapEx). The floor targets are at least 30 trillion KRW in cash dividends and at least 60 trillion KRW in share buybacks and cancellations. The 40 trillion KRW buyback announced in August was completed last week and is proceeding to cancellation [1].

\n\n

On top of that, the board passed two new resolutions in 3Q26:

\n

\n
• 14 trillion KRW in 3Q cash dividends (20,000 KRW per share): Comprising a 10,000 KRW/share base quarterly dividend plus a 10,000 KRW/share special dividend sharing gains from the Kioxia investment.
\n
• 20 trillion KRW in new treasury share acquisitions: Management explicitly separated this into 10 trillion KRW to be retired in full for shareholder value enhancement and 10 trillion KRW reserved for employee compensation (which is excluded from the ~100 trillion KRW shareholder return total).
\n

\n\n

I viewed the transparency of this capital allocation favorably: management immediately distributed Kioxia investment gains via a special dividend and clearly separated shares bought for retirement from shares bought for employee compensation.

\n\n

Solidigm IPO Review Was Halted

\n\n

CEO Kwak Noh-jung stated that after comprehensively weighing recent financial market conditions, the company concluded it would be difficult to achieve the originally expected benefits and halted the IPO review for its NAND subsidiary Solidigm [1].

\n\n

With hyperscalers locking in multi-year volumes early, the pace at which HBM4 yields and packaging stabilize will be a key gauge for the broader memory and foundry value chain.

\n\n

What I Will Check in 4Q26 Results

\n

\n
• Whether HBM4 revenue actually overtakes HBM3E in 4Q26, and by how much
\n
• How quickly HBM4 production yields stabilize
\n
• Cheongju M15X utilization and ramp progress
\n
• ASP trends on volumes shipped under long-term agreements
\n
• Why 2Q26 net profit exceeded operating profit, alongside full operating cash flow figures
\n
• Ending net cash after paying out dividends, buybacks, and capital expenditures
\n

\n\n

Sources

\n

\n
1. SK Hynix 3Q26 Business Results Press Release (KR), 2026.10.28.
\n
2. SK hynix 3Q26 Financial Results Press Release (EN), 2026.10.28.
\n
3. SK Hynix 2Q26 Business Results Press Release (KR), 2026.07.29.
\n
4. SK Hynix 1Q26 Business Results Press Release (KR), 2026.04.23.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/sk-hynix-3q26-earnings-hbm4-80pct-opm-audit/\n

\n

\n

\n Accounting & Footnotes · NASDAQ: TSLA · Form 10-Q Note 11 · 2026-10-29\n

\n

\n Executive Thesis: Part of Tesla's 3Q26 automotive and energy margin rebound came from IEEPA tariff refunds in Form 10-Q Note 11 (~16.8% Auto ex-credit and ~23.8% Energy when reversed), while R&D surged 69.4% YoY and 9M CapEx hit $13.01B toward a >$20B annual outlook.\n

\n \n \n

Part of the third-quarter automotive margin rebound came from tariff refunds, while operating expenses grew much faster than gross profit.

\n

As of 2026.10.28 (Source: Tesla 3Q26 Form 8-K Exhibit 99.1 and Form 10-Q) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

In Tesla's third-quarter release, automotive gross margin excluding regulatory credits rose to 18.1%. According to Note 11 of the Form 10-Q, however, that figure includes cost reductions from one-time tariff refunds. I wrote this memo to record how much margin remains when those refunds are backed out, and how cash flow held up while AI R&D and capital expenditures climbed [1][2].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Item	3Q26 Figures
Total Revenue	$28.046B, -0.2% YoY (deliveries -2.1%, Services & Other revenue +32.5%, growth rates Calc.)
Total Gross Margin	19.6%, gross profit $5.494B (+160bp YoY, vs. 16.8% in 2Q26)
Automotive Margin (Ex-Credits)	Reported 18.1%; ~16.8% after reversing $260M tariff refund in COGS (Calc.)
Energy Margin	Reported 33.1%; ~23.8% after reversing $526M COGS refund and $339M customer refund (Calc.)
Operating Expenses	$4.776B (+39.2% YoY, Calc.) / R&D $2.761B, SG&A $2.015B
Operating Income (GAAP)	$718M, -55.8% YoY (Calc.) / Operating margin 2.6% (vs. 5.8% in 3Q25 and 1.4% in 2Q26)
Operating Cash Flow & CapEx	OCF $6.263B (+33.3% QoQ, Calc.) / CapEx $4.732B (1Q–3Q cumulative $13.01B)
Free Cash Flow & Ending Cash	+$1.531B (vs. -$1.092B in 2Q26, -61.6% YoY, Calc.) / Cash & ST investments $45.45B

\n\n

The figures marked (Calc.) for segment margins simply reverse the disclosed tariff refund line items from Note 11; they are my arithmetic checks rather than company-published non-GAAP or normalized margin metrics.

\n\n

Operating Expenses Grew Faster Than Gross Profit

\n\n

Third-quarter gross profit was $5.494 billion, up $440 million (+8.7%) from a year ago and up 15.6% sequentially [1].

\n\n

Automotive gross margin excluding regulatory credits (18.1%) also improved from 16.3% in 2Q26 and 15.4% in 3Q25. Regulatory credits—sold to other automakers that need to comply with emissions standards—totaled $113 million in 3Q26. As noted above, however, reported automotive cost of revenues benefited from a $260 million reduction tied to tariff refunds [1][2].

\n\n

FSD Supervised—Tesla's driver-assistance feature that requires active driver supervision—reached 1.74 million active users, up 67.3% year-on-year (combining upfront purchases and monthly subscriptions, excluding free trials). Under Tesla's segment reporting, FSD revenue recognition flows into the Automotive segment. Separately, Services & Other generated $734 million in quarterly gross profit on growth in used vehicle sales, maintenance and repair, and paid Supercharging [1][2].

\n\n\n

Left: Reported 3Q26 segment gross margins vs. simple arithmetic reversal of Note 11 IEEPA tariff refunds (Calc.). Right: 3Q26 bridge from GAAP operating income to operating cash flow and free cash flow. Source: Tesla Form 8-K & Form 10-Q [1][2].

\n\n

Meanwhile, operating expenses jumped by $1.346 billion year-on-year, rising from $3.430 billion in 3Q25 to $4.776 billion in 3Q26 (+39.2%). R&D rose 69.4% from $1.630 billion to $2.761 billion, and SG&A rose 29.0% to $2.015 billion. Stock-based compensation expense increased from $663 million a year ago to $1.165 billion, reflecting CEO performance award recognition and related items [2].

\n\n

As a result, GAAP operating margin edged up from 1.4% in 2Q26 to 2.6% ($718M operating income), but stood 322 basis points lower than a year ago (5.8%, $1.624B). I read this income statement as Tesla plowing the gross profit earned in its core business straight into AI training compute clusters, the Optimus humanoid robot, and autonomy R&D.

\n\n

What Are Segment Margins Without the Tariff Refunds?

\n\n

In February 2026, the U.S. Supreme Court invalidated certain tariffs imposed under the International Emergency Economic Powers Act (IEEPA). Following that ruling, Tesla recognized refunds in 3Q26 for tariffs paid in prior periods: deducting $260 million from Automotive cost of revenues and $526 million from Energy Generation and Storage cost of revenues, while deducting $339 million from Energy revenue for contractual customer refund obligations. These entries are disclosed in Note 11 of the 3Q26 Form 10-Q [2].

\n\n

When I simply add the $260 million tariff refund back to Automotive cost of revenues, ex-credit Automotive gross margin steps down from 18.1% to roughly 16.8%. In Energy, adding the $526 million COGS refund back to cost of revenues and adding the $339 million customer refund back to Energy revenue lowers the segment gross margin from 33.1% to roughly 23.8%. Because 3Q26 blended genuine unit-cost improvements with one-time tariff refunds, the key question for 4Q26 is whether margin recovery continues without a refund tailwind.

\n\n

CapEx More Than Doubled, While Free Cash Flow Flipped Back to Positive

\n\n

The shift of capital toward AI compute infrastructure is just as visible on the cash flow statement. Third-quarter CapEx was $4.732 billion, up 110.5% from $2.248 billion in 3Q25. Cumulative 1Q–3Q CapEx reached $13.01 billion, more than double the $6.13 billion spent in the first nine months of 2025 [1][2].

\n\n

At Gigafactory Texas, Tesla reported that its AI training clusters Cortex 1 (>105MW) and Cortex 2 (>120MW) are operational, with plans to expand in-house AI training compute capacity toward 400MW by year-end [1]. In the Form 10-Q, management stated that it expects full-year 2026 capital expenditures to exceed $20 billion. Having spent $13.01 billion across 1Q–3Q, meeting that outlook implies spending more than $6.99 billion in 4Q26 alone [2].

\n\n

Operating cash flow rose by $1.566 billion sequentially from $4.697 billion in 2Q26 to $6.263 billion in 3Q26 (bringing trailing 12-month OCF to $18.71 billion). Because OCF expanded by $1.566 billion while CapEx stepped down by $1.057 billion from 2Q26's $5.789 billion, free cash flow swung from a -$1.092 billion outflow in 2Q26 to a +$1.531 billion inflow in 3Q26 (though still down 61.6% compared with 3Q25) [1].

\n\n

Quarter-end cash and short-term investments increased by $1.92 billion sequentially to $45.45 billion. In the 10-Q, Tesla noted that as capital-intensive projects expand, liquidity beyond operating cash flow may be required, and disclosed that in September 2026 it entered into credit facilities with a total commitment limit of $30 billion (an undrawn commitment line rather than borrowed debt) [2].

\n\n

What I Will Check Next

\n

\n
• Ex-credit Automotive gross margin and Energy gross margin in a quarter without tariff refunds
\n
• How quickly AI CapEx and R&D spending translate into commercial revenue from Cybercab and robotics
\n

\n\n\n\n

Sources

\n

\n
1. Tesla Q3 2026 Update (Form 8-K Exhibit 99.1), 2026.10.21.
\n
2. Tesla Q3 2026 Form 10-Q (SEC EDGAR), 2026.10.22.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/tesla-3q26-10q-ieepa-tariff-refund-margin-audit/\n

\n

\n

\n Semiconductors & HBM · KRX 000660 · TWSE 2330 · NASDAQ TSLA · 2026-10-30\n

\n

\n Executive Thesis: Putting the 3Q26 filings of TSMC (58.7% OPM), SK Hynix (80.0% OPM), and Tesla (2.6% GAAP OPM) side by side shows where AI infrastructure spending is currently captured as operating profit across the hardware value chain.\n

\n \n \n

Operating profits concentrated at leading-edge foundry nodes and HBM, while Tesla front-loaded capital into compute and R&D.

\n

As of 2026.10.29 (Source: TSMC, SK Hynix & Tesla 3Q26 earnings releases and Form 10-Q) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

In my previous three notes, I looked at the third-quarter filings of TSMC, SK Hynix, and Tesla one by one. Here I put all three filings side by side to check where the capital flowing into AI infrastructure is currently being captured as operating profit across the physical supply chain [1][3][4].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Value-Chain Layer	3Q26 Earnings & Capital Expenditure
Leading-Edge Foundry (TSMC)	Revenue US$46.72B (+41.2% YoY), gross margin 66.4%, operating margin 58.7%, HPC share 68%
Advanced Nodes & Packaging (TSMC)	3nm 33%, 2nm 8% (≤7nm combined 79%), 1Q–3Q cumulative CapEx US$40.64B
AI Memory (SK Hynix)	Revenue 104.62T KRW (+328% YoY), operating profit 83.33T KRW, operating margin 80.0%
Memory Contracts & Cash (SK Hynix)	HBM4 expected to overtake HBM3E in 4Q26, >50% of 2027–2028 capacity under LTA, net cash 98.7T KRW
AI Cluster & Vehicles (Tesla)	Revenue US$28.05B, gross margin 19.6%, operating margin 2.6%, R&D US$2.761B (+69.4% YoY)
Downstream Infrastructure (Tesla)	3Q26 CapEx US$4.732B (1Q–3Q US$13.01B), Cortex 1 & 2 live, targeting 400MW by year-end

\n\n

SK Hynix and TSMC Captured the Margin; Tesla Spent Ahead of Revenue

\n\n

In 3Q26, operating margin was 80.0% at SK Hynix and 58.7% at TSMC [1][3]. HBM is the high-bandwidth memory mounted directly alongside AI accelerator dies, and TSMC supplies both the leading-edge wafer foundry nodes and the CoWoS advanced packaging that binds them together. The companies sitting on those two physical bottlenecks captured the lion's share of current operating profits.

\n\n

Tesla sits at the opposite end of the chain. Building out its own AI training compute clusters and robotics platforms requires putting capital into equipment and R&D first. Third-quarter R&D rose 69.4% year-on-year to $2.761 billion, leaving GAAP operating margin at 2.6% [4].

\n\n

How Downstream CapEx Connects to Upstream Discipline

\n\n

Tesla spent $4.732 billion on capital expenditures in 3Q26 alone (+110.5% YoY; $13.01 billion across 1Q–3Q) and expects full-year 2026 CapEx to exceed $20 billion [4]. Upstream, TSMC spent US$40.64 billion across 1Q–3Q (US$13.84 billion in 3Q26) while keeping its full-year CapEx budget unchanged at US$60B–$64B even though management said AI demand continues to exceed supply [1][2]. I view downstream CapEx as feeding upstream orders and cash flows—though the filings do not disclose how much of Tesla's specific CapEx flows directly into TSMC or SK Hynix revenue.

\n\n

TSMC's Revenue Continues Shifting Toward Leading-Edge Nodes

\n\n

TSMC's HPC platform revenue grew 22% sequentially to reach 68% of total revenue. By wafer revenue, 3nm rose from 23% in 3Q25 and 30% in 2Q26 to 33% in 3Q26, while 2nm jumped from 3% to 8% (putting ≤7nm nodes at 79%) [1]. On the call, management noted that it is developing 9x and 14x reticle-size large-substrate technology in CoWoS packaging and plans to more than double 2nm production capacity in 2027 [2].

\n\n

SK Hynix Locked In Long-Term Contracts as It Transitions to HBM4

\n\n

SK Hynix expects HBM4 to surpass HBM3E in both volume and revenue in 4Q26 as M15X capacity maximizes by year-end. More than 50% of its total 2027–2028 production capacity is already contracted under long-term agreements with market-linked price bands. That means as the Yongin Phase 1 fab comes online and CapEx rises in 2027, more than half of the company's 2027–2028 output already has committed buyers [3].

\n\n

Will Upstream Margins Hold Once New Capacity Comes Online After 2027?

\n\n

The question I care about most from here is whether current upstream margins hold after 2027, when the next wave of fab capacity comes online. TSMC has already flagged that gross margin dilution from overseas fabs could widen from ~2 percentage points in 2026 to 3–4 percentage points per year thereafter [2]. SK Hynix enters its 2027 CapEx step-up holding 114.5 trillion KRW in cash and 98.7 trillion KRW in net cash [3], while for Tesla the open question is how quickly capital invested in Cortex clusters turns into autonomy, software, and services cash flow [4].

\n\n

In next quarter's filings, I will check three items first:

\n

\n
• TSMC's overseas fab margin dilution, pricing pass-through, and N2 yield stabilization pace
\n
• Whether SK Hynix's HBM4 yield ramp and 321-layer eSSD demand stay on track
\n
• How quickly Tesla's 400MW year-end cluster target translates into quarterly free cash flow
\n

\n\n\n\n

Sources

\n

\n
1. TSMC 3Q26 Quarterly Management Report, 2026.10.15.
\n
2. TSMC 3Q26 Earnings Conference Transcript, 2026.10.15.
\n
3. SK Hynix 3Q26 Business Results Press Release, 2026.10.28.
\n
4. Tesla Q3 2026 Form 10-Q (SEC EDGAR), 2026.10.22.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/ai-supply-chain-3q26-crosscheck-tsmc-skhynix-tesla/\n

\n

\n

\n Big Tech & AI CapEx · NASDAQ: GOOGL · MSFT · META · AMZN (10-Q/8-K) · 2026-10-31\n

\n

\n Executive Thesis: Across Microsoft, Alphabet, Meta, and Amazon, combined 3Q26 gross cash PP&E CapEx rose 107.6% YoY to ~$201.9B while operating cash flow grew 10.4% YoY—pushing combined CapEx/OCF from 61.2% to 115.1%. Only Microsoft stayed below 100%.\n

\n \n \n

As I read the filings, downstream hyperscalers are still front-loading physical capacity ahead of cash flow.

\n

As of 2026.10.30 (Source: MSFT, GOOGL, META & AMZN Jul-Sep 2026 SEC filings and transcripts) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

In the July–September 2026 quarter (3Q26; FY27 Q1 for Microsoft), the cash that the four U.S. hyperscalers—Microsoft, Alphabet, Meta, and Amazon—spent to buy property and equipment (CapEx) rose 107.6% year-on-year. Over the same period, cash generated from core operations (operating cash flow, OCF) grew 10.4%. Among the four companies, Microsoft was the only one that spent less on CapEx than it brought in from operating cash flow [1][8][11][15].

\n\n

This note continues my AI supply-chain series. Up to the previous post, I looked at how AI infrastructure profits have concentrated upstream in foundry (TSMC) and HBM (SK Hynix). Here I organize the 3Q26 filings of the four downstream hyperscalers buying those chips to build data centers, and lay out two explicit hypotheses along with the exact criteria I will use to score them in upcoming filings.

\n\n

Across the Four Companies Combined, CapEx Exceeded Operating Cash Flow

\n\n

To compute the 4-company aggregate on a uniform basis, I summed gross cash purchases of property and equipment from each company's statement of cash flows. Finance leases—long-term leases accounted for similarly to financed purchases—are excluded from this cash line, and for Amazon I used gross PP&E purchases before deducting proceeds from property sales and vendor incentives. On that uniform gross cash basis, 4-company 3Q26 CapEx totaled roughly $201.9 billion [1][8][11][15].

\n\n

The combined CapEx-to-OCF ratio jumped from 61.2% in 3Q25 to 115.1% in 3Q26. Crossing 100% means the group as a whole spent more cash on equipment and buildings during the quarter than it generated from operations. On the earnings call, Microsoft's CFO noted that the majority of CapEx went toward shorter-lived assets such as CPUs and GPUs and reflected higher memory and storage pricing—which aligns with our earlier supply-chain view that downstream dollars are flowing straight into logic and memory (I did not re-open the upstream filings for this post) [3].

\n\n

Company by Company, Only Microsoft Is Below 100%

\n\n

In the company-by-company breakdown below, I follow each company's own cash-basis CapEx definition: Meta includes principal payments on finance leases, and Amazon nets out property proceeds and incentives. Because of those definition differences, the four individual numbers below do not sum to the $201.9 billion gross total above.

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Company	3Q26 Cash-Basis CapEx (YoY Growth, Calc.)
Microsoft (FY27 Q1)	~$39.4B (+103.0%) · CapEx/OCF: 67.9%
Alphabet (3Q26)	~$51.4B (+114.4%) · CapEx/OCF: 121.4%
Meta Platforms (3Q26)	~$34.9B (+80.2%) · CapEx/OCF: 106.5%
Amazon (3Q26)	~$75.9B (+121.7%) · CapEx/OCF: 178.9%

\n\n\n

Left: 3Q26 cash-basis CapEx/OCF ratio by company vs. the 4-company combined ratio (115.1%, up from 61.2% in 3Q25). Right: Microsoft YoY growth rate comparison across cash PP&E (+103.0%), reported CapEx incl. finance leases (+29.5%), and OCF (+28.7%), plus Alphabet QoQ uncommenced lease growth (+53.2%) vs. cash CapEx (+14.3%). Source: SEC EDGAR filings, author calculations [1][3][5][8][9][10][11][15].

\n\n

At all four companies, CapEx grew faster year-on-year than operating cash flow, and at Alphabet operating cash flow actually declined year-on-year. On each company's own free cash flow definition (OCF minus CapEx), Alphabet and Meta were negative in 3Q26, Amazon was negative on a trailing 12-month basis, and only Microsoft remained positive (and remained below 100% CapEx/OCF even when finance leases are included) [1][3][8][11][15].

\n\n

Microsoft's Growth Picture Changes Depending on the Accounting Basis

\n\n

Headline CapEx reported by Microsoft in its earnings presentation includes assets acquired under finance leases, while excluding operating leases [4]. Depending on whether we use the cash flow statement (\"cash basis\") or the headline presentation number including finance leases (\"reported basis\"), Microsoft's year-on-year trajectory looks completely different: cash-basis CapEx rose 103.0% YoY, whereas reported-basis CapEx rose 29.5% YoY—virtually matching its 28.7% YoY operating cash flow growth [1][3][5].

\n\n

Even if we strip finance leases out of the reported figure, growth is 58.8% YoY (Calc.) due to timing differences between when equipment is received and when cash is paid. Furthermore, the share of finance leases inside CapEx declined compared with a year ago while new assets added via operating leases rose—consistent with Microsoft's July 2026 accounting update revising the expected useful life of data center buildings (shifting more new data center leases into operating lease classification) [2][3][4][5].

\n\n

What the Four Companies Guided for Next Quarter and 2027

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Company	Disclosed Forward CapEx Direction
Microsoft	Oct–Dec higher than Jul–Sep (reported basis); acceleration in Jan–Jun 2027 and full FY27 growth (calendar 2026 trimmed vs. July solely due to third-party data center delivery delays) [3][4]
Alphabet	2026 technical infrastructure investment to increase substantially over 2025 (no specific dollar figure printed) [9]
Meta	Full-year 2026 range implies 4Q26 > 3Q26 (~$49.2B 4Q floor, Calc.); aggressive CapEx growth guided for 2027 [11]
Amazon	2026 full-year outlook maintained, implying ~$47.8B in 4Q26 (Calc., below 3Q26); meaningful increase guided for 2027 [15][16]

\n\n

Hypothesis 1: Will Companies Whose CapEx Outpaces OCF Slow Down Investment or Cut Buybacks First?

\n\n

First, I corrected an initial assumption in my draft notes regarding share buybacks: Alphabet, Meta, and Amazon all repurchased $0 of common stock across January–September 2026 (whereas in Jan–Sep 2025 Alphabet bought back ~$39.8B, Meta bought back ~$26.2B, and Amazon bought $0) [8][9][11][12][16]. Microsoft repurchased ~$4.38 billion in 3Q26, down 22.5% YoY [1].

\n\n

Meanwhile, external financing has stepped up: Alphabet cited AI infrastructure CapEx in its June equity issuance and issued both equity and corporate bonds in 3Q26; Meta entered into preferred stock and common stock forward contracts in October [14]; Amazon raised long-term debt in 3Q26 [15]; and Microsoft priced a bond offering on October 29 (settling November 4, though because Microsoft's CapEx is inside OCF I do not classify its debt as plugging a CapEx deficit) [7].

\n\n

Hypothesis 2: When Uncommenced Lease Commitments Grow Faster Than Cash CapEx, Is True Investment Larger?

\n\n

Uncommenced lease commitments represent signed data-center and infrastructure leases that have not yet commenced (combining operating and finance leases; Microsoft does not split them and Meta reports them together) [2][12]. Because a commitment balance spans multiple years while cash CapEx is a quarterly flow, I compare only their sequential growth rates (end-September vs. end-June commitment balance against 3Q26 vs. 2Q26 cash CapEx):

\n

\n
• At Alphabet, uncommenced lease commitments jumped 53.2% QoQ while cash CapEx rose 14.3% QoQ [8][9][10].
\n
• At Meta, uncommenced leases also grew faster than cash CapEx; at Microsoft, the two growth rates were similar; only at Amazon did cash CapEx grow faster [6][13][17].
\n
• Why I withhold judgment on 3Q26: Starting in 3Q26, both Alphabet and Meta expanded the scope of their commitment footnote to include energy infrastructure, without breaking out how much of the increase came from the scope change [9][12]. Therefore, I will only count uncommenced lease growth as supporting evidence in a quarter where the footnote definition does not change.
\n

\n\n

Falsification Rubric for the Next Two Quarters (4Q26 & 1Q27)

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Observed Outcome in 4Q26 / 1Q27 Filings (±5% Tolerance Band)	How I Will Score It
Amazon & Meta: 4Q26 CapEx is not >5% below implied baseline (AMZN ~$47.8B, META floor ~$49.2B) AND 2027 guidance is maintained/raised through 1Q27	Against Hypothesis 1 (regardless of external financing mix)
Alphabet: 2027 CapEx plan and 1Q27 filing maintain or raise current investment pace	Against Hypothesis 1
Microsoft: Quarterly share buybacks do not drop by >5% below 3Q26 (~$4.38B)	Against Hypothesis 1 (if AMZN, GOOGL & META all score Against, Hypothesis 1 is falsified)
Full-year guidance cut by >5% (excluding delivery delays), 4Q26 actual >5% below baseline, weaker 2027 plan, or MSFT buybacks down >5%	Supports Hypothesis 1
Uncommenced lease balance rises, yet new assets added under operating + finance leases do not exceed 3Q26	Against Hypothesis 2 (Meta excluded if new leased assets are not disclosed)
Uncommenced lease balance shrinks without new leased assets commencing (cancellations or unmet conditions)	Against Hypothesis 2

\n\n

What I Will Check Next

\n

\n
1. Whether the 4-company combined CapEx/OCF ratio stays above 100% or drops back below 100%
\n
2. Whether Microsoft crosses 100% CapEx/OCF, or whether any of the other three drops below 100%
\n
3. The gap between Amazon's actual 4Q26 CapEx and the ~$47.8B implied figure
\n
4. Microsoft's quarterly share buyback dollar amount relative to ~$4.38B
\n
5. Sequential growth rates of uncommenced lease commitments vs. cash CapEx (on an unchanged footnote scope)
\n
6. Whether Microsoft quantifies the cash-vs-delivery timing gap or shifts its lease mix again
\n

\n\n

Sources

\n

\n
1. Microsoft FY27 Q1 Earnings Release (Form 8-K Exhibit 99.1), 2026.10.28.
\n
2. Microsoft Form 10-Q (Ended Sept 30, 2026, Note 12 Leases), 2026.10.28.
\n
3. Microsoft FY27 Q1 Earnings Call Transcript, 2026.10.28.
\n
4. Microsoft FY26 Q4 Earnings Call Transcript, 2026.07.29.
\n
5. Microsoft FY26 Q1 Earnings Call Transcript, 2025.10.29.
\n
6. Microsoft FY26 Form 10-K (Leases Footnote), 2026.07.29.
\n
7. Microsoft Bond Pricing Term Sheet (Filing FWP), 2026.10.29.
\n
8. Alphabet 3Q26 Earnings Release (Form 8-K Exhibit 99.1), 2026.10.28.
\n
9. Alphabet Form 10-Q (Ended Sept 30, 2026, MD&A & Note 4 Leases), 2026.10.29.
\n
10. Alphabet Form 10-Q (Ended June 30, 2026, Note 4 Leases), 2026.07.23.
\n
11. Meta Platforms 3Q26 Earnings Release (Form 8-K Exhibit 99.1), 2026.10.21.
\n
12. Meta Platforms Form 10-Q (Ended Sept 30, 2026, Note 10 Commitments), 2026.10.22.
\n
13. Meta Platforms Form 10-Q (Ended June 30, 2026), 2026.07.30.
\n
14. Meta Platforms Preferred Stock & Forward Contract Form 8-K, 2026.10.26.
\n
15. Amazon 3Q26 Earnings Release (Form 8-K Exhibit 99.1), 2026.10.29.
\n
16. Amazon Form 10-Q (Ended Sept 30, 2026, MD&A & Note 4), 2026.10.30.
\n
17. Amazon Form 10-Q (Ended June 30, 2026, Note 4 Commitments), 2026.07.31.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/big-tech-3q26-ai-capex-scorecard-201b-audit/\n

\n

\n

\n Big Tech & AI CapEx · NASDAQ: AAPL · SEC Form 10-K / 10-Q · 2026-11-01\n

\n

\n Executive Thesis: In the aligned April-June 2026 quarter, Apple spent 2.24% of revenue on cash CapEx while Microsoft, Alphabet, Meta, and Amazon spent 27.02%-49.53%. Ahead of Apple's Nov 2 FY26 Q4 call, here are my two falsifiable hypotheses and the $3.534B-$3.599B Q4 CapEx baseline.\n

\n \n \n

In the aligned April–June 2026 quarter, Apple spent 2.24% of revenue on cash CapEx while the four cloud hyperscalers spent 27.02%–49.53%.

\n

As of 2026.10.30 (Source: Apple FY26 Q2-Q3 financial statements, FY25 Q4 statements & Big Tech SEC XBRL/8-K) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

I originally planned to write a review of Apple's fourth-quarter results, but because the numbers are not out yet, I turned this into a pre-earnings baseline note using Apple's third-quarter filings and management's fourth-quarter outlook. While the cloud hyperscalers are front-loading AI infrastructure CapEx, I wanted to lock in the baseline numbers for Apple before the release drops.

\n\n

Throughout this note, capital expenditures (CapEx) refer to cash paid to acquire property, plant, and equipment. Quarter labels follow Apple's fiscal calendar: FY26 Q3 is the April–June quarter ended June 27, 2026, and FY26 Q4 is the July–September quarter ended in late September 2026.

\n\n

FY26 Q4 Numbers Are Not Out Yet

\n\n

Apple's FY26 Q4 earnings call is scheduled for November 2, 2026 at 14:00 PT (November 3 at 07:00 KST) [1]. I had intended to review Q4 total revenue, Products vs. Services revenue and gross margins, CapEx and CapEx-to-revenue, operating cash flow, share repurchases and dividends, and next-quarter guidance—but because none of those have been reported yet, I did not estimate any of them.

\n\n

What we do have right now is the official outlook management gave on the July 30, 2026 earnings call: FY26 Q4 total revenue was guided to grow 9%–11% year-on-year, operating expenses (R&D plus SG&A) were guided to $19.1B–$19.4B, and gross margin was guided to 47%–48% (which includes roughly 1 percentage point of benefit from tariff refund recognition). Management also commented on iPhone and Services directional growth, OI&E, and the tax rate, but did not provide a specific CapEx guidance figure (I verified the call remarks via The Motley Fool's published transcript rather than the audio replay) [2]. Because there is no company CapEx guidance, I will compare actual Q4 CapEx against a baseline range I calculated below.

\n\n

Hypothesis 1: Does Apple Spend Less on CapEx Than Hyperscalers, and Does Services Margin Mask the AI Burden?

\n\n

My starting premise across this series is that companies buying semiconductors to build data centers and AI services are front-loading equipment ahead of cash flow, and that those outlays compress margins. From that premise, Hypothesis 1 has two parts: Part 1 is that Apple's cash CapEx-to-revenue ratio is materially lower than that of the four hyperscalers (Microsoft, Alphabet, Amazon, and Meta)—if Apple's ratio approaches theirs, Part 1 is falsified. Part 2 is that Apple's high Services gross margin makes its AI investment burden less visible on the consolidated income statement.

\n\n

Because Apple reports its cash flow statement only on a cumulative year-to-date basis, I calculated standalone FY26 Q3 (Apr–Jun) cash CapEx ($2.455B) by subtracting 6-month cumulative CapEx from 9-month cumulative CapEx ($6.799B) and divided it by FY26 Q3 revenue ($109.417B). The four hyperscaler figures cover the aligned three months ended June 30, 2026 [3][4][5].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Company	Aligned Apr–Jun 2026 Cash CapEx / Revenue (Calc.)
Apple (FY26 Q3)	2.24% ($2.455B / $109.417B)
Amazon (2Q26)	27.02%
Alphabet (2Q26)	37.50%
Microsoft (FY26 Q4)	39.78%
Meta Platforms (2Q26)	49.53%

\n\n\n

Left: Aligned Apr–Jun 2026 cash CapEx/Revenue across Apple (2.24%) vs. Big Tech peers (27.02%–49.53%). Right: Apple FY25 Q4 cash CapEx ($3.242B) vs. the FY26 Q4 revenue-paced baseline range ($3.534B–$3.599B). Source: Apple financial statements & SEC XBRL/8-K data, author calculations [2][3][4][5][6][8].

\n\n

In the July–September 2026 quarter, the four hyperscalers' cash CapEx-to-revenue ratios moved even higher, spanning 38.26% to 52.54% (Amazon 38.26%, Alphabet 40.25%, Microsoft 42.66%, Meta 52.54%) [5][6]. Because Apple's July–September figure has not been released yet, placing it directly next to the 38.26%–52.54% range would mix quarters (note: the 4-peer Apr–Jun ratios were computed from SEC XBRL data without opening each peer's 2Q Form 10-Q body text) [5][5].

\n\n

In the aligned April–June data, Apple shows none of the physical front-loading visible across the four cloud peers—so Part 1 of Hypothesis 1 holds so far. However, Apple records part of its AI spending outside capital expenditures: the FY26 Q3 Form 10-Q notes that R&D ($11.729B, +32% YoY) increased due to infrastructure-related costs including AI, and on the earnings call the CEO noted that AI outlays also sit inside cost of sales [2][7]. A low CapEx ratio does not automatically mean total AI spending is small.

\n\n

Why Part 2 (Whether Services Margin Masks the AI Burden) Cannot Be Decided From the Filings

\n\n

In FY26 Q3, Apple's Services gross margin was 75.6%, flat year-on-year [7]. However, Apple does not break out AI or data-center CapEx separately, blends AI operating costs into R&D and cost of sales, and does not disclose dollar payments for third-party cloud compute. The public record confirms that Services gross margin is high, but because AI costs are not broken out, Part 2 cannot be tested from the filings—so I withhold judgment on Part 2.

\n\n

Hypothesis 2: Will FY26 Q4 CapEx Grow Materially Faster Than Revenue?

\n\n

Hypothesis 2 states that if Apple relies on external cloud partners and on-device processing (running AI models directly on iPhones and Macs) rather than building hyperscale data centers itself, its CapEx growth rate should not materially exceed its revenue growth rate. If FY26 Q4 CapEx comes in substantially above the baseline below, Hypothesis 2 is falsified.

\n\n

Standalone Q4 CapEx is obtained by subtracting 9-month cumulative CapEx ($6.799B in FY26) from full-year CapEx. Applying the same method to FY25 (full-year CapEx $12.715B minus 9-month CapEx $9.473B) gives FY25 Q4 cash CapEx of $3.242 billion [3][8]. Scaling $3.242 billion by management's guided FY26 Q4 revenue growth rate (+9%–11% YoY) yields a revenue-paced baseline range of $3.534 billion to $3.599 billion [2][8].

\n\n

Two important caveats apply to this test: first, I have not pre-set a rigid percentage cutoff for \"materially above,\" so the November 2 print will be a first-pass check rather than a mechanical verdict. Second, because this line item is total company PP&E (and Apple uses both external cloud and its own data centers), landing inside $3.534B–$3.599B does not prove the external-cloud/on-device premise is true—only a large overshoot serves as a falsifying signal in the opposite direction.

\n\n

What I Will Check When FY26 Q4 Results Drop

\n

\n
• If Apple's Q4 CapEx/Revenue ratio moves toward the 4-hyperscaler band, I will revise my view that Apple is not front-loading equipment
\n
• If Apple discloses line-item AI or cloud infrastructure costs, I will revisit my withheld judgment on Services gross margin
\n
• If actual Q4 revenue growth lands outside 9%–11% YoY, I will recalculate the $3.534B–$3.599B CapEx baseline using the actual revenue growth rate
\n
• Any new FY27 Q1 guidance or CapEx commentary on the call [6][6][6]
\n

\n\n

Sources

\n

\n
1. Apple Investor Relations: FY26 Q4 Earnings Release & Call Schedule, accessed 2026.10.30.
\n
2. The Motley Fool: Apple (AAPL) Q3 2026 Earnings Call Transcript (Call held 2026.07.30), published 2026.08.07.
\n
3. Apple FY26 Q3 Condensed Consolidated Financial Statements (PDF), 2026.07.30.
\n
4. Apple FY26 Q2 Condensed Consolidated Financial Statements (PDF, 6M Cash Flows), 2026.04.30.
\n
5. SEC EDGAR XBRL Company Facts API: Microsoft, Alphabet, Amazon, Meta Platforms, accessed 2026.10.30.
\n
6. Jul–Sep 2026 Earnings Releases (Form 8-K Exhibit 99.1): Microsoft (2026.10.28), Alphabet (2026.10.28), Amazon (2026.10.29), Meta (2026.10.21).
\n
7. Apple Form 10-Q for the Quarterly Period Ended June 27, 2026 (MD&A Gross Margin & R&D), 2026.07.31.
\n
8. Apple FY25 Q4 Condensed Consolidated Financial Statements (PDF), 2025.10.30.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/apple-fy26-q4-10k-capital-light-ai-services-margin/\n

\n

\n

\n Accounting & Footnotes · SEC Form 10-Q / 8-K Footnote Methodology · 2026-10-31\n

\n

\n Executive Thesis: Before taking headline operating profit at face value, I strip out one-time tariff refunds, non-cash stock compensation, shareholder-return double-counting traps, and initial fab depreciation dilution across the 3Q26 filings of Tesla, SK Hynix, and TSMC.\n

\n \n \n

Before taking headline operating income at face value, I separate one-time tariff refunds, non-cash expenses, and initial depreciation dilution.

\n

As of 2026.10.29 (Source: Tesla 3Q26 Form 8-K & Form 10-Q, SK Hynix & TSMC 3Q26 releases) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

When you read the third-quarter filings of Tesla, SK Hynix, and TSMC all the way through the footnotes, several numbers emerge that headline earnings tables never show: one-time tariff refunds, non-cash expenses, and initial depreciation dilution from brand-new fabs. This note records how I separated each of those items across the three companies' actual 3Q26 disclosures [1][2][3][4].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Checklist Item	Figures Verified in 3Q26 Filings
One-Time COGS & Revenue Adjustments	Tesla 10-Q Note 11: IEEPA tariff refund reduced Auto COGS by $260M and Energy COGS by $526M; customer refund obligation reduced Energy revenue by $339M
Reported vs. Ex-Refund Margins	Tesla ex-credit Auto GM: 18.1% reported vs. ~16.8% ex-refund (Calc.) / Energy GM: 33.1% reported vs. ~23.8% ex-refund (Calc.)
Accounting Profit vs. Cash Flow	Tesla GAAP operating income $718M, stock-based compensation $1.165B, operating cash flow $6.263B, free cash flow +$1.531B
Income Statement vs. Balance Sheet Cash	SK Hynix operating profit 83.33T KRW, net profit 39.78T KRW, ending cash 114.5T KRW (+26.5T KRW QoQ), net cash 98.7T KRW
New Fab Initial Dilution	TSMC gross margin 66.4% (-1.3%p QoQ on initial 2nm ramp); overseas fab dilution guided at ~2%p in 2026 and 3–4%p thereafter
Separating Shareholder Return Tranches	SK Hynix FY26 return target ~100T KRW (incl. 40T KRW completed buyback); 3Q board resolved 14T KRW dividend + 20T KRW buyback (10T cancel / 10T employee comp)

\n\n

1. Tesla's Margin Rebound Blended Unit Economics With One-Time Tariff Refunds

\n\n

A company's reported gross margin moves when core ASPs or manufacturing yields change—but it can also jump when prior-period costs are refunded and credited against current-quarter cost of revenues. That is why the footnotes in the Form 10-Q matter more than the summary tables in the Form 8-K press release.

\n\n

In 3Q26, Tesla's automotive gross margin excluding regulatory credits rose to 18.1% (up from 16.3% in 2Q26 and 15.4% in 3Q25), and Energy Generation and Storage gross margin reached 33.1% [2]. Turning to Note 11 of the Form 10-Q, however, reveals that following the February 2026 U.S. Supreme Court ruling on IEEPA tariffs, Tesla recognized tariff refunds in 3Q26 that reduced Automotive cost of revenues by $260 million and Energy cost of revenues by $526 million, alongside a $339 million contractual customer refund deducted from Energy revenue [1].

\n\n

Simply reversing those disclosed refund and rebate entries lowers ex-credit Automotive gross margin to roughly 16.8% and Energy gross margin to roughly 23.8% (Calc.). Keeping those adjusted figures as my baseline prevents overstating the underlying unit-cost improvement.

\n\n

2. Why Operating Income and Operating Cash Flow Diverged So Sharply

\n\n

Operating income on the income statement is calculated after deducting depreciation and stock-based compensation (SBC). Because SBC compensates employees with equity rather than cash, it reduces GAAP operating profit without an immediate cash outflow; the statement of cash flows adds SBC back to arrive at operating cash flow.

\n\n

In 3Q26, Tesla's GAAP operating income was $718 million (a 2.6% operating margin). Once $1.165 billion of stock-based compensation and depreciation/amortization were added back alongside working-capital movements, operating cash flow reached $6.263 billion. That allowed Tesla to fund $4.732 billion in 3Q26 CapEx ($13.01B across 1Q–3Q) and still post +$1.531 billion in positive free cash flow [2]. At the same time, because SBC jumped from $663 million a year ago to $1.165 billion, I track it separately as an equity dilution cost to shareholders.

\n\n

3. Even at Highly Profitable Companies, Profit, Cash, and Buyback Tranches Move Differently

\n\n

At SK Hynix, 3Q26 operating profit was 83.33 trillion KRW (80% OPM), while net profit after tax and non-operating items was 39.78 trillion KRW (38% net margin). On the balance sheet, cash and equivalents rose by 26.5 trillion KRW sequentially (from 88.0T to 114.5T KRW) and borrowings fell by 2.8 trillion KRW (from 18.6T to 15.8T KRW), lifting net cash to 98.7 trillion KRW [3].

\n\n

Shareholder return disclosures also require careful separation so the same capital is not counted twice: management's ~100 trillion KRW FY2026 shareholder return target (≥50% of FCF) already includes the 40 trillion KRW buyback announced in August and completed prior to the release. On top of that, the 3Q26 board newly resolved a 14 trillion KRW cash dividend and a 20 trillion KRW treasury stock acquisition—explicitly splitting the 20T KRW buyback into 10T KRW for share cancellation (included in the shareholder return total) and 10T KRW for employee compensation (excluded from the return total) [3].

\n\n

4. Separating Initial Fab Depreciation Dilution From Core Pricing Power

\n\n

When a capital-intensive manufacturer ramps a next-generation node or opens overseas fabs, initial depreciation weighs on gross margin before yields reach maturity. TSMC's 3Q26 gross margin dipped 1.3 percentage points sequentially from 67.7% to 66.4% as 2nm ramped to 8% of wafer revenue (landing inside its 65%–67% guidance band), and management guided 4Q26 gross margin to 65%–67% (66.0% midpoint, -40bp QoQ) while quantifying overseas fab dilution at roughly 2 percentage points in 2026 and 3–4 percentage points per year thereafter [4]. Recording those disclosed dilution bands upfront makes it much easier in future quarters to tell normal ramp depreciation apart from any genuine erosion in pricing power.

\n\n

What I Will Check Next

\n

\n
• Whether TSMC's 4Q26 gross margin stays inside management's 65%–67% guidance band
\n
• Whether tariff refund or customer rebate entries reappear in Tesla's next quarterly footnote
\n
• Whether Tesla's quarterly stock-based compensation continues to climb
\n

\n\n\n\n

Sources

\n

\n
1. Tesla Q3 2026 Form 10-Q (SEC EDGAR, Note 11), 2026.10.22.
\n
2. Tesla Q3 2026 Update (SEC Form 8-K Exhibit 99.1), 2026.10.21.
\n
3. SK Hynix 3Q26 Business Results Press Release, 2026.10.28.
\n
4. TSMC 3Q26 Earnings Conference Transcript, 2026.10.15.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/earnings-season-accounting-footnote-checklist-3q26/\n

\n

\n

\n Semiconductors & HBM · KRX 000660 · TWSE 2330 · US 4-Peer Cash Flow · 2026-11-01\n

\n

\n Executive Thesis: In a capital-cycle framework, CapEx expansion turns into systemic risk when companies spend beyond operating cash flow and plug the gap with debt. In 3Q26, Tesla, TSMC, and SK Hynix funded their CapEx out of operating cash flow and balance-sheet cash buffers.\n

\n \n \n

Across the three companies I reviewed here, capital expenditure is currently resting on operating cash flow and accumulated cash rather than debt.

\n

As of 2026.10.29 (Source: Tesla 3Q26 Form 8-K & Form 10-Q, SK Hynix & TSMC 3Q26 releases) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

The capital cycle describes how industries expand capacity during boom times and eventually face oversupply when that capacity comes online. From a capital-cycle perspective, a surge in capital expenditures turns into systemic financial risk when companies invest more than the cash they generate from operations and bridge the deficit with external debt. With that test in mind, I placed the 3Q26 CapEx, operating cash flow, cash balances, and debt figures of Tesla, TSMC, and SK Hynix side by side [1][2][3].

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Check Item	3Q26 Disclosed Figures
Tesla Cash Flow	OCF $6.263B, CapEx $4.732B (1Q–3Q cumulative $13.01B), FCF +$1.531B (swung positive from -$1.092B in 2Q26)
Tesla Liquidity Buffer	Cash & short-term investments $45.45B (+$1.92B QoQ, Calc.), plus $30B credit facility limit signed in Sept 2026
TSMC Cash Flow	OCF NT$919.77B, CapEx NT$443.21B (US$13.84B), FCF NT$476.56B, full-year CapEx budget US$60B–$64B
SK Hynix Net Cash	Ending cash 114.5T KRW (+26.5T KRW QoQ), borrowings 15.8T KRW (-2.8T KRW QoQ), net cash 98.7T KRW
SK Hynix Returns & Reinvestment	FY26 return target ~100T KRW (incl. 40T completed buyback), 3Q board resolved 14T dividend + 20T buyback, Yongin fab starts in 2027
Upstream Order Visibility	TSMC raised 2026 USD revenue growth outlook to ~44%; SK Hynix secured LTAs for >50% of 2027–2028 capacity

\n\n

Tesla's Free Cash Flow Flipped Back to Positive in a Single Quarter

\n\n

Tesla is concentrating capital into its in-house AI training clusters and robotics lines. In 2Q26, free cash flow (operating cash flow minus CapEx) dipped into negative territory (-$1.092B) as quarterly CapEx spiked to $5.789 billion. In 3Q26, operating cash flow rose 33.3% sequentially to $6.263 billion, covering $4.732 billion of quarterly CapEx out of operating cash flow and leaving +$1.531 billion in free cash flow [1].

\n\n

Cash on the balance sheet also rose: ending cash and short-term investments reached $45.45 billion, up from $43.52 billion in 2Q26 and $41.65 billion in 3Q25. Management expects full-year 2026 CapEx to exceed $20 billion (with $13.01B spent across 1Q–3Q). In the Form 10-Q, Tesla noted that as capital-intensive projects expand, liquidity beyond operating cash flow may be required, and disclosed that in September 2026 it entered into credit facilities totaling $30 billion. Because a credit facility is an undrawn commitment limit rather than borrowed debt, Tesla has effectively built a two-layer liquidity buffer—$45.45 billion in cash plus a $30 billion standby line [1].

\n\n

TSMC and SK Hynix Are Expanding Capacity and Returning Cash Simultaneously

\n\n

Upstream in foundry and memory, where AI infrastructure profits are currently concentrated, cash is accumulating even during heavy capacity expansion. TSMC spent NT$443.21 billion (US$13.84 billion; US$40.64 billion across 1Q–3Q) on CapEx in 3Q26, yet because operating cash flow was NT$919.77 billion, it generated NT$476.56 billion in free cash flow [3]. Even while raising its full-year 2026 USD revenue growth outlook to ~44%, TSMC kept its annual CapEx budget unchanged at US$60B–$64B—a sign of capital allocation discipline against future oversupply.

\n\n

At SK Hynix, quarter-end cash and equivalents rose by 26.5 trillion KRW sequentially to 114.5 trillion KRW while borrowings declined by 2.8 trillion KRW (from 18.6T to 15.8T KRW), bringing net cash to 98.7 trillion KRW [2]. Under its policy of returning at least 50% of free cash flow, the company targets roughly 100 trillion KRW in FY2026 shareholder returns (including the 40T KRW buyback completed after August, plus the newly resolved 14T KRW cash dividend and 20T KRW buyback—10T for cancellation and 10T for employee compensation).

\n\n

At the same time, SK Hynix halted the IPO review for its NAND subsidiary Solidigm, locked in long-term agreements covering more than 50% of its 2027–2028 production capacity, and prepared for the 2027 Yongin Phase 1 fab ramp from a position where it can fund both capacity expansion and shareholder returns internally without tapping external equity markets [2].

\n\n

What I Will Check in Upcoming Filings

\n

\n
• Whether Tesla continues to cover quarterly CapEx out of operating cash flow or begins drawing down its $30B credit facility
\n
• Whether ASPs and shipment volumes absorb the rising depreciation when TSMC doubles 2nm capacity and ramps overseas fabs, and when SK Hynix brings Yongin Phase 1 online and maximizes M15X in 2027
\n
• How LTA price-band clauses and CoWoS packaging tightness feed into upcoming gross margin guidance
\n

\n\n

Sources

\n

\n
1. Tesla Q3 2026 Form 10-Q (SEC EDGAR), 2026.10.22.
\n
2. SK Hynix 3Q26 Business Results Press Release, 2026.10.28.
\n
3. TSMC 3Q26 Quarterly Management Report, 2026.10.15.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/ai-capex-capital-cycle-liquidity-buffer-view-3q26/\n

\n

\n

\n Macro & FOMC · Federal Reserve FOMC & U.S. Treasury · 2026-11-01\n

\n

\n Executive Thesis: The FOMC held the federal funds target range at 3.75%-4.00% on a 10-2 vote, with Lisa D. Cook and Beth M. Hammack dissenting for a 25bp hike. Economic growth wording stepped up from 'solid' to 'strong', while the Sept FOMC minutes noted heavy hyperscaler AI bond issuance.\n

\n \n \n

Growth wording stepped up from \"solid\" to \"strong,\" while the capital investment assessment and balance-sheet directives stayed unchanged.

\n

As of 2026.10.30 (Source: Federal Reserve FOMC statements, SEP, minutes & U.S. Treasury yield curve) · Growth rates, ratios, and baselines not printed directly in the primary releases were calculated by me from the disclosed financial statements and are marked (Calc.) in the tables.

\n\n

In my previous capital-cycle note, I wrote that AI capital expenditure at the companies I reviewed was being funded out of internal cash flows rather than debt. The September FOMC minutes contain a record of heavy private bond issuance for AI infrastructure—which runs against that broader premise—though because the minutes do not provide a numeric ratio of debt to internal cash flow, I withhold a final verdict below.

\n\n

On October 28, 2026, the Federal Reserve held the federal funds target range steady at 3.75%–4.00% (3-3/4 to 4 percent, midpoint 3.875%), keeping the range to which it had raised rates by 0.25 percentage points in September [1][3]. Here I place the October statement wording, dissents, and balance-sheet directives side by side with September, and record two falsifiable hypotheses for the December 8–9 meeting.

\n\n

Both Dissents Favored an Additional 25bp Rate Hike

\n\n

The Committee vote shifted from 12–0 in September to 10–2 in October. Both dissenting members—Lisa D. Cook and Beth M. Hammack—preferred raising the target range by an additional 1/4 percentage point (25bp) at the October meeting [1][3]. The policy debate inside the Committee sat between holding steady and hiking further; there were zero dissents in favor of a rate cut. If a cut dissent appears in December or the hike dissents disappear, I will update this reading (and I will check the full roster of the 10 majority voters when the October minutes are released).

\n\n

Economic Growth Wording Stepped Up, and the Policy Sentence Was Replaced

\n\n

Comparing the first three paragraphs of the September and October statements line by line [1][3]:

\n

\n
• Economic activity: Upgraded one notch from \"expanding at a solid pace\" in September to \"expanding at a strong pace\" in October.
\n
• Policy forward sentence: Completely replaced. September's sentence (\"Today's policy action will support a timelier return of inflation to 2 percent\") was replaced in October with \"The accumulation of more information will help in assessing the appropriate stance of monetary policy\"—shifting from explaining a hike to waiting for more data.
\n
• Productivity & CapEx: Kept verbatim across both meetings: \"Productivity growth is strong, and capital investment is robust.\"
\n

\n\n

Nothing Changed on the Balance Sheet at This Meeting

\n\n

In the Implementation Note released alongside the statement, the operational directives for the Fed's securities holdings were unchanged from September: roll over in full all principal payments from maturing Treasury securities at auction, reinvest agency debt and agency MBS principal payments into Treasury bills, and purchase Treasury bills as needed to maintain an ample level of reserves (with IORB at 3.90%, ON RRP at 3.75% with a $160B per-counterparty daily limit, and Standing Repo / Primary Credit at 4.0%) [2][4].

\n\n

According to the September minutes, reserve-management bill purchases were paused and followed no preset path [8]. In his October 28 press conference (preliminary transcript), Chair Warsh explained that a dedicated task force is reviewing balance-sheet policy and is expected to bring recommendations to the FOMC around early Q1 2027 (timing not yet finalized, and unlikely to materially alter the current policy stance) [6]. Until then, I track weekly holdings directly in the Fed's H.4.1 release [10].

\n\n

Testing My Prior Capital-Cycle Premise Against BEA Q3 GDP and the Sept FOMC Minutes

\n\n

At the press conference, Chair Warsh noted that capital investment represents a historically high share of GDP led by high-tech acceleration, while also pointing out that interest rates have affected investment and credit spreads have widened on some assets [6]. The day after the meeting (October 29), the Bureau of Economic Analysis (BEA) released its Q3 2026 advance GDP estimate: real GDP grew at a 3.4% annual rate (up from 2.2% in Q2 2026, with H1 revised from 1.8% to 2.4%), and equipment investment was led by information processing equipment, primarily computers and peripheral equipment [11].

\n\n

While the October statement and BEA GDP release describe the size and composition of CapEx without addressing how it is funded, the September FOMC minutes explicitly discuss financing: in the System Open Market Account (SOMA) Desk report, market commentary cited heavy private bond issuance to fund AI infrastructure as one factor pushing Treasury yields higher, noting that long-dated corporate bonds issued by cloud hyperscalers in large volumes kept credit spreads wide; several participants also cited expected growth in AI-related borrowing as an upward pressure on longer-term Treasury yields [8].

\n\n

That record in the September minutes is adverse evidence against a blanket \"AI CapEx is self-funded\" view (and matches what we saw in our Big Tech 4-company CapEx note, where Alphabet, Meta, and Amazon issued debt/equity while TSMC, SK Hynix, and Tesla funded CapEx from cash flow). Because the minutes do not provide a numeric breakdown of debt versus internal cash flow across the economy, I record the minutes as adverse qualitative evidence while withholding a macro-level verdict until numeric funding-share data is available.

\n\n

Hypothesis 1: Will the Dot Plot Path Stay Elevated as Long as Capital Investment Holds Up?

\n\n

First, I corrected an initial error in my pre-note draft, which had assumed the Fed was cutting rates and running balance-sheet runoff (QT): in reality, the Fed hiked in September and held in October, and is rolling over maturing principal in full [1][2][3]. My revised Hypothesis 1 is that as long as business capital investment supports growth, the FOMC dot plot median rate path will not shift lower.

\n\n\n

Left: FOMC SEP federal funds rate median projections in June 2026 vs. September 2026 (end-2026 and end-2027 both at 4.1%; unrounded 2026 median = 4.125%, vs. current midpoint 3.875%). Right: 1-day U.S. Treasury par yield change on Oct 28, 2026. Source: Federal Reserve SEP & U.S. Treasury [1][7][9].

\n\n

Because October had no SEP, the active baseline is the September 16, 2026 SEP: the rounded median federal funds projection for end-2026 rose from 3.8% in June to 4.1% in September (and in Figure 2 of the SEP, 12 of 18 participants placed their dot at 4.125%, including the 9th and 10th middle dots—one 25bp hike above the current 3.875% midpoint). The end-2027 median also rose from 3.6% in June to 4.1% in September (with 2028 at 3.9% vs. 3.4% in June, 2029 at 3.6%, and longer-run at 3.2% vs. 3.1%; 2026 real GDP median 2.3%, unemployment 4.1%, PCE inflation 3.7%, core PCE 3.4%) [7].

\n\n

Falsification rule for December 8–9, 2026: If the December FOMC statement retains the clause \"capital investment is robust\" yet the December SEP median for end-2027 drops below September's 4.1%, I will mark Hypothesis 1 as falsified [5][7].

\n\n

Hypothesis 2: Do Balance-Sheet Decisions Move Long-End Treasury Yields More Than Rate Decisions?

\n\n

Hypothesis 2 tests whether balance-sheet policy announcements move long-term Treasury yields (which proxy for long-duration equity discount rates) more strongly than policy-rate decisions alone. Because October left the balance-sheet directive unchanged, I use the October 28 rate-decision day move in U.S. Treasury par yields (~15:30 ET quotes, after the 14:00 ET FOMC release) strictly as the baseline hurdle for future balance-sheet decision days [9]:

\n

\n
• Oct 27 → Oct 28 (FOMC decision day): 2-Year fell 0.03%p (4.70% → 4.67%), 10-Year rose 0.06%p (4.99% → 5.05%), and 30-Year rose 0.07%p (5.30% → 5.37%).
\n
• Oct 29 (BEA Q3 GDP release day): 2-Year 4.61%, 10-Year 4.98% (-0.07%p), and 30-Year 5.32% (-0.05%p) [9][11].
\n
• Falsification rule: On the future FOMC day when a balance-sheet decision is announced (provided it does not coincide with a policy-rate change or a major GDP/jobs/CPI print), if the 1-day absolute change in both the 10-Year and 30-Year Treasury yields is smaller than 0.06%p and 0.07%p respectively, Hypothesis 2 is falsified.
\n

\n\n

What I Will Check Next

\n

\n
• November 18, 2026 (3-week release rule, Calc.): Check the October FOMC minutes for the full voter breakdown and discussion behind the two hike dissents [5]
\n
• December 8–9, 2026 FOMC & SEP: Check whether \"capital investment is robust\" is kept and whether the end-2027 median dot stays at or above 4.1% [5]
\n
• Early Q1 2027: Watch for the Balance Sheet Task Force recommendation mentioned by Chair Warsh [6] and weekly H.4.1 reserve/asset tables [10]
\n

\n\n

Sources

\n

\n
1. Federal Reserve FOMC Statement (October 28, 2026), 2026.10.28.
\n
2. Federal Reserve Implementation Note (October 28, 2026), 2026.10.28.
\n
3. Federal Reserve FOMC Statement (September 16, 2026), 2026.09.16.
\n
4. Federal Reserve Implementation Note (September 16, 2026), 2026.09.16.
\n
5. Federal Reserve FOMC Meeting Calendars & Minutes Release Rules, accessed 2026.10.28.
\n
6. Chair Warsh FOMC Press Conference Transcript (Preliminary PDF, pp. 1, 12, 14–15), 2026.10.28.
\n
7. FOMC Summary of Economic Projections (September 16, 2026, Table 1 & Figure 2), 2026.09.16.
\n
8. Minutes of the Federal Open Market Committee (September 15–16, 2026), released 2026.10.07.
\n
9. U.S. Department of the Treasury: Daily Treasury Par Yield Curve Rates (October 2026), accessed 2026.10.30.
\n
10. Federal Reserve Statistical Release H.4.1 (Factors Affecting Reserve Balances), 2026.10.29.
\n
11. Bureau of Economic Analysis (BEA): Gross Domestic Product, 3rd Quarter 2026 (Advance Estimate PDF), 2026.10.29.
\n

\n\n

---

\n

\n Canonical English Research Portal: https://deep-growth-investing.pages.dev/posts/fomc-october-2026-statement-sep-dot-plot-comparison/\n

\n