Home / Research / Company deep dive
Company deep dive · The foundry deals, labelled
Intel Foundry: who signs first?
Everybody wants to know which giant Intel's chip factories will win first — Apple, Nvidia, Tesla. The honest answer starts with one number: of the $6.697 billion the foundry took in last quarter, only $411 million came from outside customers. This deep dive goes through every possible deal and labels each one from the filings, the reporting and the rumors: signed, in talks, rumored only, or denied.
Position disclosure — November 8, 2026: of the names this piece covers, the desk currently holds AVGO, TSLA, GOOGL. Its full position book (QCOM · AVGO · TSLA · GOOGL · SNDK · MU) prints in the open in the current editions.
Fact checked — figures verified against primary sources at publication · November 2026.
Section 01The verdict
The most believable path is not the glamorous one — it is packaging. Advanced packaging (Intel's EMIB and EMIB-T) is the only external foundry business with shipping customers today: Amazon and Cisco have been packaging customers since 2022, because a customer can keep its chips at TSMC and move only the package to Intel — no redesign, no new design kit to learn — while TSMC's own packaging lines carry waits of 52 to 78 weeks for everyone but its biggest customers.
Second: a small, government-brokered Apple deal for a low-end chip, still preliminary — reported by The Wall Street Journal in May 2026, confirmed by neither company. Third: the Tesla and SpaceX Terafab, which names Intel's next process, 14A — but SpaceX's own filing calls it a framework whose real contracts may never be signed.
Not happening: a TSMC-run joint venture (denied in public by TSMC's chief executive, twice), a merger with GlobalFoundries (a rumor from February 2025 — the companies never held talks), and a spin-off (penalized by the U.S. government's warrant, which switches on if Intel's foundry ownership drops below 51 percent).
Labels used throughout: Signed announced by Intel and/or the counterparty, or in an SEC filing. Reported by reputable press (Reuters, WSJ, Bloomberg, Semafor, CNBC), no signing announced. Estimate an analyst figure, labelled as such. Rumored only channel checks, supply-chain press or single-source speculation, unconfirmed by either party. Nothing below presents an estimate as a reported figure.
Section 02The anchor: one quarter that disciplines everything
External revenue is climbing — $174 million in Q1 2026, $293 million in Q2, $411 million in Q3 — but Intel's own filing says the Q3 jump came primarily from one contract that will not repeat, and it does not name the customer. A good part of the nine-month growth is Altera, a business Intel sold a majority of last year, buying from its old parent. The break-even recipe management gave in May 2025 — operating break-even sometime in 2027 on "low- to mid-single-digit billions" of annual external wafer revenue — was no longer repeated on the Q3 2026 call. Treat it as a lapsed target, not a current guide. What management does still demand: at least a 20% return on investment on every new node.
Section 03The deal table: wafers
Every credible report of an external wafer customer for Intel's 18A, 18A-P or 14A processes, through November 2, 2026.
| Customer | What | Status | What the source actually says |
|---|---|---|---|
| Microsoft | Custom chip on 18A | Signed (2024) | Satya Nadella said in February 2024 that Microsoft had "chosen a chip design that we plan to produce on Intel 18A." The "$15 billion" figure from that event was the total across all foundry commitments, not a Microsoft contract. No product, volume or revenue has been disclosed since. |
| Amazon / AWS | AI fabric chip on 18A; Xeon 6 on Intel 3 | Signed (2024 framework) | A "multi-year, multi-billion-dollar framework" (September 2024). No quantified volume or start date; the Ohio production it pointed to cannot host wafers before 2030. |
| Apple — low-end M-series | 18A-P, MacBook Air / iPad class | Reported — preliminary | WSJ (May 8, 2026): a preliminary agreement after more than a year of talks, with the U.S. government helping bring Apple in; neither company has confirmed it. Analyst Ming-Chi Kuo: shipments from mid-2027 at 15–20 million units a year — Estimate. Intel's Q3 2026 filing still names no external wafer customer. |
| Apple — non-Pro iPhone | 14A | Rumored only | One analyst line (GF Securities, from 2028), extrapolated from the M-series trial. No company source. |
| Tesla / SpaceX — Terafab | 14A + packaging, Texas | Reported — framework | Intel publicly joined Terafab on April 7, 2026; Musk named 14A on Tesla's Q1 call. SpaceX's own S-1 describes a framework whose definitive agreements may never be signed. The $16.8 billion Phase 1 figure is SpaceX and Tesla's construction budget — not a dollar of Intel revenue. Musk confirmed talks with TSMC about Terafab in October 2026. |
| Nvidia | 18A/14A testing; joint products | Reported — testing | Reuters reported 18A manufacturing tests in 2025. The signed Nvidia deal (September 2025) is equity plus joint CPUs and packaging for joint products — explicitly not Intel making Nvidia's main GPU dies. A DigiTimes report on 2028 Feynman I/O dies is Rumored only. |
| Broadcom | 18A tests; networking switches | Rumored only | Reuters reported in 2024 that Broadcom's test wafers did not meet expectations at that stage. An October 2026 analyst channel check claiming Intel had "won" Broadcom contracts is unconfirmed by either company — and Intel's filing two weeks later named no customer. |
| AMD | Unspecified | Reported — disputed | Semafor (October 2025) reported early-stage talks; SemiAccurate contradicted the report days later. Both companies declined to comment. |
| IPU co-development | Signed (product) | April 2026: expanded joint development of infrastructure chips. No process node, no volume — and the revenue books in Intel's product segment, not Foundry external revenue. Google's packaging interest is in Section 04. | |
| MediaTek | Intel 16 (mature node) | Signed (2022) | Smart-edge chips on a mature process, with MediaTek stressing TSMC remains its advanced-node partner. Its 2026 Intel relevance is mainly as Google's TPU design partner. |
| U.S. defense programs | Secure Enclave, RAMP-C, SHIP | Signed | Up to $3 billion of program funding; prototype customers include Boeing and Northrop Grumman. Real, trusted-supply revenue — niche volume, not commercial scale. |
| Fortinet | Security processor | Signed (2026) | Co-development announced July 2026; no node or volume in Intel's release. Symbolically useful, economically small. |
A warning that applies to several rows at once: an October 15, 2026 analyst channel check claimed Intel had already won contracts from Broadcom, Google, Apple, AMD, Microsoft and Nvidia, with very large dollar figures attached. Intel has confirmed none of those names. Every claim sourced only to that check is classified rumored only here, however directionally interesting.
Section 04Packaging: the business that is already shipping
Advanced packaging — wrapping finished chips and connecting them to the rest of the machine — is where the confirmed external activity actually is. CNBC reported in April 2026 that Intel's packaging customers include Amazon and Cisco, customers since 2022, plus a new commitment from SpaceX and Tesla. Intel's CFO has raised his outlook from hundreds of millions to "billions of dollars per customer" (a management characterization, not booked revenue), and management targets a revenue ramp in the second half of 2027.
The big swing item is Google's TPU v9, reported by SemiAnalysis and TrendForce to use Intel's EMIB-T packaging instead of TSMC's CoWoS — with an order reported above three million units for 2028. Neither Google nor Intel has confirmed it, and the reported substrate yields conflict wildly (a 45–50% analyst estimate against claims near 98% in some press). It is the single largest credible external item in this report — and it is still unconfirmed.
Why packaging is different: porting a chip design to a new process costs tens of millions of dollars and one to two years per design. Moving only the package costs a fraction of that and skips the redesign entirely. That is why every credible near-term external win sits in this section — and why the buyers who need Intel most need its packaging and its U.S. address more than its transistors.
Section 05What is not happening
A TSMC-operated joint venture — denied. Reports in early 2025 described a White House-initiated plan for TSMC to take a stake and help run Intel's fabs. TSMC's chief executive C.C. Wei said on the record that TSMC "is not engaged in any discussion with other companies regarding any joint venture, technology licensing or technology transfer" — and TSMC reiterated the denial in September 2025 after a fresh report.
A GlobalFoundries merger — a rumor, never talks. The February 2025 speculation was triggered by an executive reshuffle at GlobalFoundries, not by any contact between the companies. No reputable outlet has ever reported that the two held talks.
A spin-off — blocked by the government's warrant. Washington bought 433.3 million Intel shares at $20.47 — about $8.9 billion, roughly a tenth of the company — in August 2025. Attached is a five-year warrant for about 5% more of the company at $20 a share, exercisable only if Intel ceases to own at least 51% of its foundry business. Intel could sell up to 49% without triggering it; its CFO has said the motivation to cross the line probably is not there. Fab-level partnerships remain open — Intel has sold 49% stakes in individual fabs before (and bought one back in 2026) — which is the realistic way foundry economics get shared without tripping the wire.
Section 06The constraints and the calendar
Three facts discipline every headline above. First, Tan's gating rule: 14A capacity gets built only against committed demand, and Intel's own 10-K warns that without a significant external customer, 14A "will not be economical" and could be paused. Second, the competitive wall: TSMC held about 72.5% of global foundry revenue in Q2 2026 (TrendForce), and Intel Foundry's external revenue is roughly 1% of TSMC's quarterly total — Samsung is out competing for the same hedge demand, including a July 2026 memorandum with Broadcom. Third, the calendar: no 14A wafer revenue before 2028, no Ohio wafers before 2030, and no Apple volume before the second half of 2027 even if everything holds. The only external revenue that can move the 2026–27 loss line is packaging, mature nodes and defense work — precisely where the confirmed activity is.
Until a name is signed and disclosed, Intel's foundry story is a packaging business with a famous shareholder. That is a better story than the rumors — it is just a smaller one than the headlines.
Hear it: this deep dive is also a podcast episode of The Degenerates Journal — listen to "Deep Dive: Intel Foundry — Who Signs First?", or find it on Spotify and YouTube.
Section 07What we could not verify
The October 2026 channel check claiming Intel had "won" six named customers — no confirmation from Intel or any customer, and contradicted in tone by Intel's own filing two weeks later. Broadcom's reported "cancellation" and Nvidia's reported "halt" of 18A testing — both circulate in secondary summaries; the primary reports could not be retrieved. Google's TPU v9 packaging adoption and its reported order size — unconfirmed by Google, MediaTek or Intel. The Apple preliminary agreement's terms — no product, node, volume or date in the reporting, and neither company has ever commented. Terafab's definitive terms — no contract value, volume or Intel capital commitment appears in any filing. Intel's yields for 18A and 14A — Intel publishes none; every yield figure in circulation is an estimate or a management characterization.
Sources — Intel Q3 2026 8-K and 10-Q (October 29–30, 2026) and FY2025 Form 10-K, read in full; Intel announcements on the U.S. government investment (August 2025), the Nvidia collaboration (September 2025) and Terafab (April 2026); The Wall Street Journal on the Apple preliminary agreement (May 2026); Reuters on the 18A tests, the TSMC joint-venture pitch and MediaTek; TSMC statements via DataCenterDynamics; CNBC on packaging customers (April 2026); Ming-Chi Kuo's published supply-chain notes; TrendForce on TPU v9 and foundry market share; SpaceX's S-1 description of Terafab; Semafor on the AMD talks. Researched November 2–3, 2026. Figures are as reported by those sources; analyst figures are labelled estimates where they appear.