INTCIntel Corporation

INTC — dossier, 2026-09-23

Slot names-ai-capex-4. Drafted by Claude in an unattended cloud routine. Nothing here is human-verified. Every figure carries the document it was read from; anything from background knowledge is marked [background]. Quotes were located by substring search in the source and copied, never re-typed. Never edited after today.

Price reference: $123.86 close, 2026-09-22, volume 107,946,745 (ROIC.ai NASDAQ:INTC).

---

The one-line version

Intel's operating business turned in Q2 2026 — first positive operating income in the series read here, gross margin back over 40%, data-centre demand ahead of supply — and its reported earnings got worse, by $12.5 billion, because its own share price went up. Both facts come out of the same 10-Q, and the second is a consequence of how the U.S. government's stake is accounted for. Meanwhile the thing the equity is actually priced for — an external foundry anchor customer — is still $293 million a quarter and still unnamed.

---

S1 — Source

watchlist (owner, Apple Stocks, 2026-09-20), row: {ticker: "INTC", label: "Intel Corporation", coverage: sec_domestic, cluster: ai_semis, cik: 50863}. No Tier 0 promotion; INTC is not in queue.json promoted.

S2 — Kill test

Read from XBRL company facts (data.sec.gov/api/xbrl/companyfacts/CIK0000050863.json, User-Agent sent) and confirmed against the filings.

Quarterly, Q2 2026 against Q2 2025 (10-Q Consolidated Condensed Statements of Operations): revenue 16,128 vs 12,859 (+25.4%), gross profit 6,509 vs 3,542 (GM 40.4% vs 27.5%), operating income 1,796 vs (3,176), and net loss attributable to Intel (11,033) vs (2,918).

Survives-or-dies. Intel does not die on liquidity and does not die on the product business. Cash and short-term investments are $29,727m against total debt of $50,537m — net debt ~$20.8bn on $9.7bn of annual operating cash flow — plus a $7.0bn revolver available to February 2029, a $3.0bn facility to January 2029, $10.0bn of commercial paper authorisation undrawn, and roughly $23.0bn of fresh equity raised in August 2026. Capex has been cut 43% from the FY2023 peak while operating cash flow has held. H1 2026 revenue is up 16.4% with gross margin up 770bp. What can kill this equity is not solvency; it is that ~$655bn of market value is being paid for an outcome — Intel Foundry winning external customers at scale — that has $293m of quarterly evidence behind it, while the share count has risen 27% in five years and two more claims (143m escrowed shares, a 241m-share warrant struck at $20.00) sit outside that count.

The share-count trap, in order. 4,212m (FY2023) → 4,280m (FY2024) → 4,530m (FY2025) → 5,104m diluted weighted-average in Q2 2026 → 5,044m actually outstanding at 2026-07-17 (10-Q cover) → +242,105,262 sold at $95.00 on 2026-08-10/11 → ~5,286m. Add the 143m escrowed shares not yet released and the 241m DOC warrant shares and the fully-loaded number is ~5,670m.

S3 — Filings, gate-minimum

Read: FY2025 10-K (filed 2026-01-23) risk factors and MD&A; Q2 2026 10-Q (filed 2026-07-24) MD&A delta and Notes 2, 3, 4, 6, 10, 11, 15; the 2026-08-12 8-K; the Q2 2026 earnings call transcript (turns 11–57 of 57). Not read: the DEF 14A filed 2026-03-23 (located, not opened — see the scorecard's next action), the second 10-K, the Q1 2026 10-Q, any transcript before Q2 2026, and the efts.sec.gov auditor-change / restatement full-text sweep.

The three biggest risks management itself discloses

1. "We are making significant, long-term and inherently risky investments in R&D and manufacturing facilities that may not realize a favorable return." — 10-K FY2025, Risk Factors, heading. This is the foundry bet stated as a risk by the company making it.
2. "There are a number of risks and uncertainties associated with the U.S. government's acquisition of significant equity interests in us." — 10-K FY2025, Risk Factors, heading. Its first bullet is the one that has no analogue anywhere else on this watchlist: "The legislative, judicial or executive branches of the U.S. government could determine in the future that all or a portion of the transactions were unauthorized, void or voidable." A later bullet is simply "The transactions are dilutive to existing stockholders."
3. "Our alternative financing arrangements and pursuit of government grants involve risks and may not be successful." — 10-K FY2025, Risk Factors, heading. The body says both SCIP arrangements "are expected to significantly and increasingly impact our net income (loss) attributable to Intel and earnings (loss) per share attributable to Intel in future periods". Intel paid $14.2bn on 2026-04-08 to end one of them.

The Escrowed Shares mechanism — why the loss and the share price move together

• The derivative liability for the Escrowed Shares was $2.7bn at 2025-12-27 and $15.6bn at 2026-06-27 (10-Q Note 4).
• Q2 2026 carried a $12.5bn loss and H1 2026 $13.6bn, inside "Interest and other, net", which was $(12,576)m for the quarter against $(95)m a year earlier.
• The 10-Q's MD&A says it plainly: "we recognized a $12.5 billion loss related to the net change in the fair value of shares of our common stock released from escrow during the quarter and remaining in escrow at the end of Q2 2026, driven by an increase in our stock price."

So: for as long as escrowed shares remain unreleased, a rising Intel share price produces a larger reported GAAP loss. At 2026-09-22 the stock is $123.86. Unless it fell back between 2026-06-27 and the Q3 close, Q3 2026 carries another mark of the same kind. Anyone reading Intel's GAAP EPS as a business signal right now is reading a price chart upside down.

Two structural consequences, both from Note 4:

• 143 million Escrowed Shares were not yet released at 2026-06-27; 71m are in basic EPS as not contingently issuable, 71m are excluded until the DOC disburses under Secure Enclave.
• Warrants over 241 million shares at $20.00 become exercisable "if we cease to directly or indirectly own at least 51 % of our foundry business". A Foundry spin-off — the structural fix most bulls and bears both reach for — triggers a ~241m-share claim struck 84% below the current price. That is a disclosed, quantified toll on the strategic option.

Accounting the SEC staff touched

The FY2025 10-K discloses that Intel's accounting conclusion for the U.S. Government Agreement, as presented in the Q3 2025 10-Q, "was subsequently adjusted based upon our consultation with the staff of the SEC on this matter, which concluded in our fourth quarter of fiscal 2025" — i.e. after that 10-Q was filed. Intel concluded the adjustments were immaterial. This is not a restatement and is not an auditor change; it is the staff moving the number on the single largest and newest item on the balance sheet, and it is the reason the FY2025 figures are the first clean read of it.

Ireland SCIP: off-balance-sheet financing bought back at a price

Apollo took 49% of the Fab 34 VIE in 2024. On 2026-04-08 Intel reacquired the whole minority interest for $14.2bn cash, funded from cash, short-term investments and a $6.5bn term loan, which was then repaid with $6.5bn of new senior notes — $1.0bn 4.65% 2031, $1.0bn 5.00% 2033, $2.2bn 5.30% 2036, $1.8bn 6.13% 2056, $500m 6.20% 2066. Total debt went $46,585m → $50,537m in the half. Read together with the risk factor above, this is Intel paying to undo a financing structure whose cost was about to show up in EPS.

Transcript — Q2 2026 call, 2026-07-23 (read Q&A-first, SKILL §3.1)

Turns 11–57 of 57. Turns 1–10 (safe-harbor, CEO and CFO prepared remarks) were not read — one transcript call at this depth, and §3.1 puts the Q&A first. The CFO's guidance below is what he restated inside the Q&A.

Guidance and changes.

• Q3 2026 client revenue "flattish"; CCPG up a bit, driven by edge. The end market is softer — "that market is struggling with the memory dynamics in the marketplace".
• Gross margin guided flat quarter over quarter. The CFO's framing of the year: "our number one goal this year … was to get gross margins comfortably into the 40s in every quarter."
• Q4 2026 revenue up, as supply unlocks — "we start to see some of the log jam break towards the end of the third quarter". He also says Intel will still be behind demand in Q4.
• Capex: 2026 tooling +40% on 2025. 2027 will be up; no number — "I purposely didn't mention a number for 27".
• ASIC business "approaching a $2 billion run rate", with "$4 billion run rate" named as the next step, against a stated "$100 billion TAM".
• Data centre: "well north of a double digit CAGR for us in terms of growth rate over the next few years."
• 14A: PDK 0.5 complete, PDK 0.9 "on track for October", risk production H2 2027, committed volume production 2028.
• Liquidity as management frames it: "We have over $30 billion of cash. We have a $10 billion revolver. So we've got $40 billion of liquidity", plus "roughly call it $10 billion of what are called non-core assets".

What the analysts asked about — six questioners, and four of the six went at the same thing from different sides: can you fund this, and for whom are you building it. Joe Moore (Morgan Stanley) on server share and on gross-vs-net capex; Stacy Raskin (Bernstein) on whether client strength was just price; Timothy Curry (UBS) on the Q4 step-up and on gross-margin drop-through; Vivek Arya (BofA) on when foundry confidence becomes announced customers and then on the balance sheet; CJ Muse (Cantor) on a capex framework against free-cash-flow goals; Aaron Rakers (Wells Fargo) on ASIC diversity and margin, and on memory.

Dodges.

• External foundry customers. Arya: "when will that confidence be backed by actual customer announcements". Lip-Bu Tan answered with 18A/18AP/14A milestones, PDK dates, SRAM yield and defect density, and said "the engagement, the customer, the feedback have been very positive". No external customer was named and no date was given. The question was about announcements; the answer was about process.
• ASIC margins. Rakers asked for "the margin profile of ASICs as they expand". Tan answered TAM; Zinsner answered run rate. Neither addressed margin.
• Server share. Moore asked what happens to server share over five years against AMD and Arm. Tan answered roadmap, and recast Arm as "a great partner and customer". No share direction.
• Deferral, not a dodge: the 2027 capex number, explicitly refused.

The sequence worth keeping. Arya's follow-up was whether the product business funds the plan or "will something else be required". Zinsner: "if we're super successful, which we're driving to, you know, we may need to tap the capital markets to drive some more investment. And, you know, we'll stay tuned if we if we need to do that." Eighteen days later, on 2026-08-10, Intel priced 210,526,315 shares at $95.00, and the underwriters took the full 31,578,947-share option the next day — ~$23.0bn, off a shelf registered the same day. Nothing improper: it is an accurate reading of a CFO who told analysts the answer three weeks early and was not pressed on it.

S5 — Ownership

13F register at 2026-06-30 (Equibles GetTopHolders; 3,591 institutions, 3,867,398,113 shares, $539.6bn):

Two things this register says that the usual index-fund top ten does not.

NVIDIA is Intel's fourth largest institutional holder. The routine's standing cross-cutting finding — first written at names-ai-capex-1, extended at names-ai-capex-2 and inverted at CBRS — is that in this complex the strong balance sheet underwrites the weak one: sellers finance buyers (NVDA/SB Energy, AMD's OpenAI warrants, AVGO/XPV), then buyers finance sellers (OpenAI lending CoreWeave $1.0bn). Intel is a third shape again: the complex's strongest balance sheet has bought a strategic stake in its oldest competitor, alongside the U.S. government and SoftBank. Intel's register is now a policy object.

The largest holder of all is missing from the table. The U.S. government is not a 13F filer, so its Escrowed-Share position appears nowhere in the institutional register. Any ownership screen run on INTC understates the real holder list by the single most consequential name on it. The evidence for that stake is in the 10-Q, not the 13F.

Corrected 3spread register, post-fix (CUSIP 458140100, 2,556 filers, 2,428,586,342 shares at 2026-06-30, change vs 2026-03-31; every row carries its SEC filing URL):

• BlackRock 426,478,962 (−21,315,170, trimmed) · State Street 216,849,675 (+2,528,657) · NVIDIA 214,776,632 (+0, held) · Invesco 157,902,890 (+3,942,408) · Susquehanna 107,877,059 (+6,988,545) · SoftBank 86,956,522 (+0, held) · Jane Street 85,597,640 (+17,135,431) · PRIMECAP 75,250,958 (−511,010) · Citadel 59,638,211 (+4,452,665) · T. Rowe 53,547,772 (+1,486,726).
• NVIDIA and SoftBank both show +0 QoQ — the stakes were already in place at 2026-03-31 and neither moved through the quarter. Neither is trading the position.
• The buyers in the quarter are market-makers and multi-strats (Jane Street +17.1m, Susquehanna +7.0m, Citadel +4.5m, Assenagon +21.2m, BofA +8.3m, Goldman +4.8m); the sellers are index and long-only (BlackRock −21.3m, Van Eck −12.4m, Barclays −14.5m, L&G −3.8m). That is a register rotating from holders toward traders, which is what a name about to issue $23bn of stock looks like.
• The Sixth Street / Schwab defect recurs here. Sixth Street Partners Management Company shows 33,298,214 shares as NEW and Charles Schwab Investment Management shows the identical 33,298,214. That is the same defective-filing pattern names-insurance-1 documented on 2026-09-22 for ACGL, KNSL, AX and QCOM (accession 0001752724-26-000051 carrying Schwab's information table). INTC is the fifth issuer it has appeared in. The script warns on it automatically; the row is not carried.
• Ten large prior holders have no 2026-06-30 13F ingested yet — Vanguard Capital Management (564,022,570 prior), Geode (109,534,637), Vanguard Portfolio Management (85,427,273), Capital World (79,877,133), Capital Research Global (77,301,863), FMR (57,515,133), Morgan Stanley (53,989,387), UBS AM (52,771,052), UBS Group (45,274,954), Northern Trust (44,887,372). No exit is read from any of them. Note the consequence: Equibles' table above shows Vanguard Capital at 285,230,183 for the same date. The two vendors disagree on that manager by ~279m shares, which is another reason to treat any single-quarter 13F delta on this name as provisional.
• 13D/G: the fifteen rows 3spread returns under cik 50863 are all filings where Intel is the filer, on other issuers. No 13D/G on Intel appears in this source.
• Form 4: 84 filings in the 365 days to 2026-09-23, clustered at 2026-05-15 (eleven filings in a day — the annual grant/vest window) and again 2026-08-03/14. No individual transaction sizes were pulled; the cadence is routine-compensation-shaped, not signal-shaped.
• Short interest (FINRA, via Equibles): 141,034,491 (2026-05-15) → 127,787,110 (2026-06-30) → 120,963,181 (2026-07-15) → 125,100,834 (2026-07-31) → 135,685,200 (2026-08-14) → 152,248,752 (2026-08-31), days to cover 1.0 → 1.7. The position grew 25.9% from mid-July through the equity offering. On ~5.3bn shares this is under 3% of the float; the direction is the point, not the size. Equibles also returns a model estimate for the 2026-09-15 settlement (~150.4m, P10–P90 131.8–168.7m) which is an estimate, not a FINRA figure, and is not used here.
• Shelf / ATM / buyback: a new automatic shelf (S-3ASR, 333-298165) was filed 2026-08-10 and immediately drawn on for the ~$23.0bn equity sale; a second S-3ASR was filed 2026-01-23. The commercial paper authorisation is $10.0bn, undrawn at 2026-06-27. No ATM programme and no buyback activity were identified; Intel is issuing, not retiring. Absence of an ATM in what was read is not proof there is none — GetAtmPrograms was not called, to stay inside the Equibles budget.

Tool defect found in this slot

python ledger/research/tools/threespread.py holders INTC prints # MariaDB plc (INTC, CUSIP G5920M100) — 13F holders as of 2026-06-30 / filers: 0. It does not error. coverage INTC returns INTEL CORP, cik 50863, with 843 insider filings and populated institutional_holdings, so the API has the issuer. The CUSIP resolution is the fault: per ledger/research/README.md, "The 13F stream is keyed by CUSIP, resolved from the issuer's 13D/G rows", and owners INTC shows all fifteen 13D/G rows under cik 50863 have INTEL CORP as the filer. The CUSIP therefore belongs to one of Intel's venture holdings.

The raw response makes it unambiguous. /v1/beneficial-ownership?ticker=INTC returns, in order: MariaDB plc / G5920M100 / filer INTEL CORP, then Intel Corp / 458140100 / filer INTEL CORP, then Mobileye Global Inc. / 60741F104, then Joby Aviation, Inc. / G65163100, then Kaltura. The endpoint keys on the filer, the old code took limit=5 and the first row with a CUSIP, and Intel's own CUSIP was row two.

Prior runs recorded 3spread returning near-empty tables for ADR CUSIPs (ARM, OMAB at names-ai-capex-2; ASML at names-ai-capex-3). This is a different and worse failure mode: a confidently-labelled table for a different company. Any name that files 13D/Gs on others — every corporate venture arm on this watchlist — was exposed to it.

Fix applied in this slot (ledger/research/tools/threespread.py, cusip_for()): pull 50 rows rather than 5; ask /v1/coverage/by-issuer/<ticker> what issuer the ticker is; take the CUSIP from the row whose issuer_name normalises to that issuer; and, when no row matches, exit with the explanation rather than printing another company's holder table. Normalisation strips punctuation and corporate suffixes and also strips the embedded spaces some CUSIPs carry (98980G 102 on ZS). Verified on INTC (now Intel Corp / 458140100 / 2,556 filers) and it leaves the ZS and MRVL paths unchanged.

S8 — Valuation and sizing ⏳ OPEN (user)

Today. $123.86 (2026-09-22). Shares ~5,286m after the August sale → ~$655bn of market value, against FY2025 revenue of $52,853m and no net income. Adding net debt of ~$20.8bn and the $15.6bn escrow derivative gives an enterprise value near $691bn on ~$60bn of run-rate revenue. There is no meaningful trailing earnings multiple; the P/E is undefined.

Out-year model. Assumptions, all stated, none of them management's: FY2026 revenue ~$64bn (H1 actual $29.7bn, H2 guided flattish-then-up); +12% in each of FY2027 and FY2028, which is inside the "well north of a double digit CAGR" the CFO claimed for data centre and generous to a flat-to- down client market; gross margin climbing from 40% toward the high 40s as 18A/18AP yields mature; operating expense held near $18.5bn; net interest ~$2.5bn; a 12% effective tax rate while the domestic valuation allowance persists; ~5.5bn shares including escrow release.

What that says. The traded price sits above the base case and roughly 9% below the bull case. To justify $123.86 on FY2028 earnings you have to underwrite the bull column — which is to say, you have to underwrite the external foundry customer that management would not name when asked directly three weeks before it sold $23bn of stock. That is the same shape names-ai-capex-2 found at MRVL and ARM, where base cases modelled below the traded price, and it is now true of the third name in a row in this cluster.

Buy price and size are owner-only (skill §5). Liquidity is not a constraint: 107.9m shares traded on 2026-09-22, ~$13.4bn of value, so any plausible position clears in minutes.

S9 — Monitoring

• Next earnings: Q3 2026, guided on the call for October 2026 — "provide additional update in October". A precise date was not pulled (ROIC's calendar needs a paid plan); the filing cadence puts it in the last ten days of October.
• Feed: the Tier 0 EDGAR sweep already covers cik 50863 (watch_edgar.py, 06:40 UTC daily).
• Leading indicator between quarters: Intel Foundry third-party revenue, disclosed in Note 2 of every 10-Q — $22m → $293m in a year. It is the only published series that measures the external foundry thesis directly, and it moves quarterly rather than annually.
• Trigger that converts this to a buy: a named external foundry anchor customer with committed 14A volume, disclosed in an 8-K or on a call — plus Foundry third-party revenue crossing ~$500m in a quarter, so the announcement is corroborated by the segment note rather than by a press release. Absent that, the base case is below the price and the name stays a watch.
• Second trigger, in the other direction: PDK 0.9 not confirmed complete by the Q3 2026 call, or the H2 2027 risk-production date moving. Management put the dates on the record; a slip is theirs.

---

Access and evidence limits, this run

• DEF 14A not read. Insider ownership %, compensation structure and what the bonus is paid on are all missing from S3. This is the largest single gap and it is the scorecard's next action.
• One transcript, not two. Q2 2026 only, turns 11–57 of 57; prepared remarks not read. Equibles budget.
• No efts.sec.gov full-text sweep for auditor changes, "changes in accountants", restatement language or NT 10-K. The SEC-consultation item in the 10-K was found by reading the 10-K, not by a sweep, so a second such item could exist unfound.
• No GetAtmPrograms / GetBuybackPrograms / GetGoingConcernStatus / GetExecutiveChanges calls — budget. The shelf and offering came from EDGAR directly, which is the better source anyway; going concern is plainly not in question here.
• Form 4 detail not extracted. Only the filing cadence (84 in 365 days) was read, not the transaction sizes or the buy/sell split.
• 3spread holders unusable for this name — see above. S5's register is Equibles'.
• No analyst-coverage count.
• The out-year model is the routine's own, built on stated assumptions, not on company guidance beyond one quarter. Treat the three cases as a frame, not a forecast.