Intel Fab Watch, Issue 01: the baseline issue

Summary

Issue 01 is the baseline. It records where Intel’s manufacturing buildout stands as of 17 July 2026, so later issues have a fixed point to measure against, and draws no conclusion from a single reading.

Two Arizona milestones landed inside the period, on their stated dates: the first high-volume logic chips shipped on ASML’s High-NA EUV tools, and 18A-P entered risk production. 18A yields are improving at roughly 7–8% a month against a stated path to industry-standard levels in early 2027 — until then Intel is capping CPU output at Fab 52 and leaving capacity idle.

The commercial side has not moved with it. External foundry revenue is $174M, 1.3% of the total, and the confirmed external customer list did not grow this period. Capital intensity ran 36.6% of revenue against TSMC’s 38.0%, while free cash flow was −$3,867M and net debt $12,242M.

What resolves that gap is 14A, and its window is open now rather than later: the v0.9 external PDK is due October 2026, management has pointed to external customer commitments in H2 2026, and the pause-or-cancel decision runs through H1 2027. Nothing in this issue settles it either way.

Financials

What changed: Q1 2026 results (10-Q filed 2026-04-24) are the baseline quarter; two large interim financing events in April will first appear on the balance sheet in the Q2 filing, expected around July 23.

Free cash flow — the cash left after operations and capital spending — remains deeply negative, which is the expected arithmetic of a heavy buildout phase: the thesis question this publication tracks is whether that spending converts into revenue-producing capacity, and FCF is where the answer will eventually show up.

Segment detail: Intel Foundry (the manufacturing arm) posted revenue of $5,421M, up 16% year over year, but the vast majority of that is Intel manufacturing for itself (intersegment). Its operating loss widened slightly to $(2,437)M from $(2,320)M a year earlier. External foundry and assembly/test revenue was $174M vs $31M in Q1 2025.

Interim events not yet on the balance sheet: in April, Intel reacquired Apollo’s 49% stake in the Ireland fab co-investment vehicle for ~$14.2B cash, then on April 30 issued $6.5B of senior notes in five tranches (coupons 4.65%–6.20%, maturities 2031–2066), which appears to refinance the bridge loan used for that buyout (inference — the filing does not state use of proceeds). Both will be visible in the Q2 10-Q.

Government stake mechanics: the US government holds a 9.9% equity stake (converted from CHIPS Act grants in August 2025). The deal’s warrant and escrow features sit on Intel’s books as a $3.6B derivative liability, revalued quarterly — Intel’s rising share price produced a $1.1B non-cash loss in Q1. The government also holds warrants for 241M shares at $20, exercisable only if Intel’s ownership of its foundry business drops below 51% — a structural disincentive to spinning off or selling control of the foundry.

Fab buildout

What changed: Two Arizona manufacturing firsts, and a €5B capital commitment to Ireland rather than a US site.

Nodes & customers

What changed: Node execution hit its stated dates; the confirmed customer list did not grow; 14A’s make-or-break window is now open.

Node progress

18A yields are improving roughly 7–8% per month per CEO Lip-Bu Tan, with industry-standard levels expected in early 2027 — until then Intel is capping CPU output at Fab 52 and leaving some capacity idle. A sell-side estimate (BlueFin Research Partners) puts combined Arizona/Oregon 18A output at ~30,000 wafers per month with defect density approaching mature-node norms; a separate analyst rumor claims 85% yields. Neither figure is an Intel disclosure. 18A-P hit its risk-production date; 14A’s v0.9 external PDK — the process design kit, the technical rulebook external customers need before committing chips to a node; Intel calls this the gating milestone for customer sign-off — is targeted for October 2026, and Cadence signed a multi-year agreement on June 8 to co-optimize 14A designs. (Yield/wafer figures: analyst estimates, not confirmed.)

Chip reception

A gap this period: no major new independent benchmarks. Panther Lake’s review cycle front-loaded at its January CES launch; Clearwater Forest (Xeon 6+, 288 cores, launched June 1) is still awaiting independent lab benchmarks rather than Intel’s own performance claims.

Foundry customers

Confirmed roster unchanged: Microsoft (18A), AWS (18A AI fabric), US DoD (RAMP-C/Secure Enclave), MediaTek and Tower on mature nodes. The July 14 KeyBanc-sourced report claiming AMD, Nvidia, Microsoft, Marvell, Micron, and OpenAI as 18A/14A design wins — plus June reporting of an Apple evaluation — remains entirely unconfirmed by Intel or the named companies. (Speculation.) For 14A the stakes are explicit: Intel has told investors the node may be paused or cancelled without a major external commitment in the H2 2026–H1 2027 window, and the CEO has said he expects foundry customer commitments in the second half of 2026. Why this matters mechanically: external customers committing to 14A would validate the foundry model beyond Intel’s own products and underpin the demand case for the US fab capacity now under construction.

Competitive & ecosystem trust

What changed: First data points logged for both tracking charts (trend lines accumulate from here); EDA certification now spans both major vendors on 18A; fab ownership is consolidating toward 100% Intel.

The foundry chart baselines two things. Capex intensity — capital spending as a share of revenue, a proxy for how hard each player is investing relative to its size: Intel 36.6% vs TSMC 38.0% vs Samsung’s semiconductor division 12.5%. Intel is spending at near-TSMC intensity on a much smaller revenue base. And external foundry revenue share: 1.3% of Intel’s total. Rising foundry revenue share over future issues would indicate external customers actually placing volume orders, not just signing letters of intent — this is the single line most directly tied to the buildout converting into a business.

The product chart baselines x86 CPU unit share (Mercury Research, Q1 2026): Intel ~70% client / 66.8% server, AMD ~30% / 33.2% — though AMD’s server revenue share hit a record 46.2%, meaning AMD sells fewer server units at higher prices. These products fund the buildout, so share trends here feed directly into how much capex Intel can sustain. In AI accelerators, Intel is currently a non-factor (Nvidia ~75–80%, AMD scaling); Intel’s next entry is Crescent Island, an inference-focused GPU due H2 2026.

Ecosystem and structure. Both major chip-design-software (EDA) vendors — Synopsys and Cadence — have certified their tools for 18A/18A-P, and Cadence has extended to 14A; whether Synopsys follows on 14A is a marker to watch. On ownership: Ireland’s Fab 34 is now 100% Intel (Apollo bought out), Arizona remains 49% Brookfield, Ohio was always wholly Intel-owned. Consolidating ownership concentrates both the capital risk and the economics. The policy backdrop shifted in January: Section 232 tariffs impose a 25% duty on certain advanced chips, with tariff-free quotas tied to US fab construction — a regime that structurally favours US-located capacity (Intel’s, but also TSMC Arizona and Samsung Texas). (The “steering” effect is interpretation, not a reported fact.)

Worth watching next

Sources: Intel Q1 2026 10-Q and 8-K filings; Intel IR materials; named-source trade press (Tom’s Hardware, EE Times, CNBC, TechSpot, EDN, Mercury Research via trade press). Items marked speculation are unconfirmed by Intel or the companies named. This note characterises no trend as favourable or unfavourable and contains no investment recommendation.