Last updated: 7 August 2026. Intel Foundry has stopped being a roadmap story and started being a manufacturing one. Intel 18A is in high-volume production and shipping in two product families. Intel 14A had its mass-production target pulled forward a full year. And the business still lost $2.09 billion in a single quarter while booking just $293 million of genuine external foundry revenue.
This page tracks where Intel Foundry actually stands, with dated figures from Intel’s own filings. Throughout, we separate what Intel or a customer has confirmed from what has only been reported — a distinction most coverage of this company blurs, and one that matters enormously here.
The state of play in 60 seconds
- Structure: Intel Foundry is still an internal reporting segment of Intel Corporation. It is not a separate legal subsidiary and has no separate board. A spin-off is effectively blocked until 2030.
- Leadership: Lip-Bu Tan has been CEO since March 2025. Since a reorganisation on 18 June 2026, Foundry is run by two executive vice presidents — Naga Chandrasekaran (front-end) and Seok-Hee Lee (advanced packaging and back-end) — reporting directly to Tan. There is no single Foundry CEO.
- 18A: in high-volume manufacturing at Fab 52 in Arizona and D1X in Oregon. Shipping in Core Ultra series 3 (Panther Lake) and Xeon 6+ (Clearwater Forest).
- 14A: PDK 0.9 due October 2026, risk production in H2 2027, high-volume manufacturing committed for 2028 — roughly a year earlier than the previous plan.
- External customers: Fortinet is the first named foundry customer under Tan, announced 21 July 2026 — but on Intel 4, not a leading-edge node. Zero external customers are formally committed to 14A.
- The wildcard: an Apple–Intel manufacturing agreement that a US President has publicly confirmed but neither company has.
The financials: revenue is climbing, losses are shrinking, external revenue is tiny
| Period | Foundry revenue | Foundry operating loss |
|---|---|---|
| Q4 2025 | $4.5B (+4% YoY) | $(2.509)B |
| FY2025 | $17.8B (+3%) | $(10.318)B |
| Q1 2026 | $5.4B (+16% YoY) | $(2.437)B |
| Q2 2026 | $5.8B (+31% YoY, +6% QoQ) | $(2.089)B |
The trend is genuinely improving: revenue up 31% year on year and the quarterly loss narrowing by $348 million sequentially. But there is a number underneath that headline which explains why the segment is still nowhere near sustainable.
Intel Foundry’s external revenue in Q2 2026 was $293 million. Everything else is intersegment — Intel manufacturing chips for Intel. Annualised, third-party foundry revenue is running at roughly $1.2 billion against a $2.1 billion quarterly operating loss. That gap is the whole story.
For context, Intel as a whole reported Q2 2026 revenue of $16.1 billion, up 25% year on year, with Data Centre and AI up 59%. The GAAP net loss of $11.0 billion looks catastrophic until you read the cause: a $12.5 billion non-cash mark-to-market charge on escrowed shares held for the Department of Commerce under the Secure Enclave agreement. Intel posted a huge GAAP loss largely because its own share price rose. On a non-GAAP basis it earned $2.2 billion.
When does Foundry break even?
Intel’s stated target is 2027. CFO David Zinsner first gave that date in May 2025 and reiterated confidence in March 2026, saying Foundry needs only “low- to mid-single-digit billions” in annual external revenue to get there, and that the path runs through 14A rather than 18A. Worth noting for anyone tracking this: the 2027 target was not restated on the Q2 2026 earnings call in July.
Intel 18A: shipping, and the yield picture
This is the part of the story that has genuinely landed. Intel’s Q2 2026 release states plainly that Intel Foundry “has entered high-volume manufacturing for a subset of Intel Core Ultra Series 3 processors.” Two product families now ship on the node:
- Panther Lake / Core Ultra series 3 — unveiled October 2025, launched at CES in January 2026, at retail from 27 January 2026. Built at Fab 52 in Chandler, Arizona, operational since October 2025. We covered the architecture in detail in our Intel Panther Lake guide.
- Clearwater Forest / Xeon 6+ — launched 1 June 2026 at Computex, Intel’s first server-class product on 18A. The flagship 6990E+ carries 288 cores across twelve 24-core compute tiles on 18A, stacked on three Intel 3 base tiles.
- 18A-P entered risk production in mid-June 2026, claiming up to 9% more performance or 18% lower power than 18A. Diamond Rapids is the lead product.
Intel also disclosed that Q2 18A output ran about 25% above internal target and more than 50% higher quarter on quarter, with Panther Lake per-SKU cost down roughly 50% year to date.
On yields, be careful what you read. Intel has never published an 18A yield figure. Every number in circulation is an analyst estimate: Morgan Stanley put it near 50% in mid-2025; BlueFin reported in July 2026 that the wafer-to-wafer variability problem had been resolved and that each of Fab 52 and D1X was running 12,000–15,000 wafer starts per month; KeyBanc estimated roughly 85% in July 2026, against about 90% for TSMC N2 and 50–60% for Samsung SF2. Tan’s own on-record statement in May 2026 was that yields were improving 7–8% per month.
One confirmed data point does more than any estimate: KeyBanc reported that 80–90% of Nova Lake compute tiles have been moved to Intel 18A, up from an original plan that put 60–70% at TSMC N2. Intel would not shift volume back in-house if the node were not working.
Intel 14A: the node the whole business rests on
If 18A proved Intel can manufacture, 14A has to prove Intel can sell. The current status:
| Milestone | Status as of August 2026 |
|---|---|
| PDK 0.5 | Complete |
| PDK 0.9 (the version customers need to tape out) | On track for October 2026 |
| Risk production | H2 2027 |
| High-volume manufacturing | 2028 — pulled forward roughly one year |
| Development fab | Oregon D1X; volume capacity earmarked for Ohio |
| Key technology | RibbonFET GAA, PowerDirect backside power, High-NA EUV insertion |
| Committed external customers | Zero |
The stakes were set out in Intel’s own filings. Since July 2025 the company has carried risk language warning of a “potential pause or discontinuation of our pursuit of Intel 14A and other next generation leading-edge process technologies if we are unable to secure a significant external customer.” That is Intel telling regulators it might exit leading-edge foundry altogether.
The tone has since reversed hard. At CES in January 2026 Tan said Intel was going “big time into 14A.” On the Q2 2026 call in July he confirmed Intel had “made the decision in Q2 to fully commit to high volume ramp in 2028.” Capex followed the words: 2026 capital spending was raised to more than $20 billion on 23 July 2026, with 2027 described as “significantly higher.”
There is also a hard deadline nobody talks about. The 35% US Advanced Manufacturing Investment Credit only applies to fab construction that begins before 31 December 2026. Projects breaking ground in 2027 get nothing. That converges with Intel’s stated H2 2026 window for customer decisions, which is why the back half of this year is the real inflection point.
Who is actually buying? Confirmed versus reported
Confirmed
| Customer | Node | Announced | Notes |
|---|---|---|---|
| Fortinet | Intel 4 | 21 July 2026 | Security Processor 6 firewall ASIC. First named external customer under Tan — but on a mature node, not 18A or 14A. |
| US Department of Defense (RAMP-C) | 18A | Programme completed 28 July 2026 | Defence-industrial-base partners completed tape-outs and prototype testing. Individual customers not named. Focus shifts to Secure Enclave. |
| Microsoft | “18A” | February 2024 | Officially announced custom processor. The product has never been publicly identified. |
| Amazon / AWS | 18A AI fabric chip + Intel 3 | September 2024 | Multi-year, multi-billion-dollar framework. No status update found in 2025 or 2026 — treat as stale. |
| Ericsson | Intel 4 | November 2023 | RAN Compute processors. |
Reported but not confirmed
- Apple. The Wall Street Journal reported a preliminary agreement on 8 May 2026; Intel shares jumped 13–19% to an all-time high. On 20 June 2026 President Trump confirmed the deal publicly, with reporting pointing to 18A-P, lower-end iPhone, iPad and Mac processors rather than flagship silicon, first chips around Q2–Q3 2027, and TSMC retaining over 90% of Apple’s supply. Neither Apple nor Intel has issued a formal statement.
- Tesla / Terafab. Elon Musk said on Tesla’s Q1 call (22 April 2026) that Terafab would use Intel 14A. Intel declined to comment. Critically, this looks like a process-technology licensing arrangement rather than a foundry wafer contract — SpaceX would build and operate the high-volume facility. That is why Intel could still report zero committed 14A customers three months later. Note also that Tesla’s AI5 chip taped out at Samsung’s Taylor fab on SF2 in July 2026; these are separate programmes.
- Microsoft Maia 2 on 18A/18A-P — single-sourced to SemiAccurate, October 2025. Unconfirmed by either company.
- Nvidia. Despite a $5 billion equity investment, reporting from January 2026 indicates Nvidia’s engagement is limited to preliminary advanced-packaging discussions. There is no confirmed Nvidia wafer commitment. See our Nvidia AI coverage for what that partnership actually involves.
- An Intel–TSMC joint venture. Reported by Reuters in April 2025 at a 20% TSMC stake. TSMC publicly denied it, and no credible corroboration has appeared since. Treat claims of a live 2026 JV as unsubstantiated.
The money that changed everything in 2025–26
| Investor | Amount | Terms | Date |
|---|---|---|---|
| US Government | $8.9B | 433,323,000 shares at $20.47 — 9.9% at the time, 8.4% as of March 2026 after dilution. Passive: no board seat, no governance rights. | 22 August 2025 |
| SoftBank | $2.0B | Common stock private placement | August 2025 |
| Nvidia | $5.0B | 214.7M shares at $23.28 (~2.9%), plus joint development of x86 RTX SoCs and custom Nvidia data-centre x86 CPUs. Not a foundry commitment. | Announced Sept 2025, closed Dec 2025 |
The government deal carries the clause that shapes Intel’s strategic options: a five-year warrant for a further ~5% of Intel at $20 per share, exercisable only if Intel drops below 51% ownership of Intel Foundry. Zinsner has said plainly that Intel’s motivation will be not to sell below 51%. A partial sale of up to 49% remains possible; a full spin-off or Foundry IPO is off the table until roughly 2030.
On paper the government position has worked out extraordinarily well — a stake bought for $8.9 billion was reportedly worth over $60 billion by late June 2026 after the Apple news.
The fab map: what is built, delayed and cancelled
| Site | Status |
|---|---|
| Fab 52, Chandler AZ | Operational since October 2025. 18A high-volume manufacturing. |
| D1X, Hillsboro OR | Operational. 18A volume plus 14A development. |
| Fab 62, Arizona | Under construction, ready around 2028. 51/49 joint venture with Brookfield. |
| Ohio One, New Albany | Delayed twice. Module 1 now 2030–2031, Module 2 in 2032. Originally slated for 2025 production. Earmarked for 14A. |
| Fab 34, Leixlip, Ireland | Intel repurchased Apollo’s 49% stake for $14.2 billion (announced 1 April 2026), returning it to full ownership. A further €5 billion expansion was announced 13 July 2026. |
| Magdeburg, Germany | Cancelled, July 2025. |
| Wrocław, Poland | Cancelled, July 2025. |
| Fab 38, Kiryat Gat, Israel | Paused since mid-2024. No restart announced. |
| Fab 9, Rio Rancho NM | Advanced packaging (Foveros). Operational since January 2024. |
| Penang, Malaysia | Advanced packaging for EMIB and Foveros. Reported 99% complete as of April 2026, first phase operating later in 2026. |
Alongside the fabs, Tan has cut hard. Headcount is around 81,000 as of July 2026, down from roughly 109,000 at the end of 2024. Management layers went from 12 to about six, and VP numbers were halved from roughly 400 to 200.
Advanced packaging: the underrated part of the story
The 18 June 2026 reorganisation is the tell. Intel carved advanced packaging out as a focused business with dedicated leadership under a new EVP — the former SK hynix CEO Seok-Hee Lee. Companies do that when they intend to run something as a standalone profit centre.
The portfolio is real: EMIB-T (2.5D with through-silicon vias, production rollout planned for 2026), Foveros-R, Foveros-B and Foveros Direct 3D hybrid bonding at sub-5-micron interconnect pitch, already used in Clearwater Forest. Zinsner told investors in March 2026 that advanced-packaging commitments could produce “billions of dollars per year in revenue” — an explicit upgrade from a previous estimate in the hundreds of millions.
The catch: Intel has not publicly named a single external advanced-packaging customer. Google, Amazon, Apple, Nvidia and Qualcomm have all been reported as being in discussions during 2026, and all declined to comment. The strategic logic is strong; the signed contracts are not yet visible.
How Intel compares to TSMC and Samsung right now
| Intel Foundry | TSMC | Samsung Foundry | |
|---|---|---|---|
| Q1 2026 foundry market share | Not in the top 10 | 72.3% | 6.5% |
| Leading node | 18A in HVM | N2 in volume since Q4 2025 | SF2 |
| Reported yield (analyst estimates) | ~85% | ~90% | 50–60% |
| Next node | 14A — HVM 2028 | A14 — volume 2028 | SF2P+ — 2027 at Taylor |
| 2026 capex | over $20B | $60–64B | — |
| Marquee customer signal | Fortinet on Intel 4 | Q2 2026 revenue $40.2B | Tesla AI5 taped out at Taylor |
Two things jump out. Intel’s reported 18A yield is better than Samsung’s SF2 — yet Samsung has the signed Tesla customer and Intel does not. And Intel 14A and TSMC A14 are scheduled to reach volume in the same year, meaning Intel gets no free run at the leading edge.
What to watch in the rest of 2026
- October 2026: Intel 14A PDK 0.9 ships to external customers. This is the version needed to tape out. Slippage here would be a serious signal.
- H2 2026: Tan has said customers should begin making firm supplier decisions in this window. Two prospective customers are running 14A test chips.
- 31 December 2026: deadline for fab construction to begin and still qualify for the 35% investment tax credit.
- Any formal statement from Apple or Intel confirming the manufacturing agreement in their own words, with a node and a volume attached.
- A named advanced-packaging customer. Given Zinsner’s “billions per year” framing, this may be where the first meaningful external revenue actually shows up.
Frequently asked questions
Is Intel spinning off Intel Foundry?
No, and it largely cannot until around 2030. Under the August 2025 agreement with the US government, Intel granted a five-year warrant for roughly 5% of the company at $20 per share that becomes exercisable only if Intel’s ownership of the foundry business falls below 51%. CFO David Zinsner has said Intel’s motivation will be to stay above that threshold. A minority sale of up to 49% is still possible.
Is Intel 18A actually in production?
Yes. Intel’s Q2 2026 earnings release confirms high-volume manufacturing for a subset of Core Ultra series 3 processors. Panther Lake reached retail on 27 January 2026 and the 288-core Clearwater Forest Xeon 6+ launched on 1 June 2026, both built on 18A.
Does Intel Foundry have any external customers?
A few, but very few at the leading edge. Fortinet was named on 21 July 2026 as the first external customer under Lip-Bu Tan — on Intel 4, a mature node. Microsoft, Amazon and Ericsson were announced earlier. External foundry revenue was just $293 million in Q2 2026, and no external customer has formally committed to Intel 14A.
Has Apple confirmed it will use Intel Foundry?
No. The Wall Street Journal reported a preliminary agreement in May 2026, and President Trump publicly confirmed it in June 2026. Neither Apple nor Intel has issued a formal statement, and the node, products, volumes and timing all come from press reporting rather than the companies.
Why did Intel report an $11 billion loss in Q2 2026 if revenue grew 25%?
The loss was almost entirely non-cash. Intel booked a $12.5 billion mark-to-market charge on escrowed shares held for the Department of Commerce under the Secure Enclave agreement — a charge driven by Intel’s own share price rising. On a non-GAAP basis Intel earned $2.2 billion in the quarter.
When will Intel Foundry break even?
Intel’s stated target is 2027, and Zinsner has said it requires only low- to mid-single-digit billions in annual external revenue, arriving mainly via 14A. Two caveats: external revenue is currently annualising near $1.2 billion, and the 2027 target was not restated on the July 2026 earnings call.
Will Nvidia manufacture chips at Intel Foundry?
There is no confirmed wafer commitment. Nvidia invested $5 billion in Intel stock and agreed to co-develop x86 RTX system-on-chips and custom data-centre x86 CPUs, but that is a product partnership, not a foundry contract. Reporting from January 2026 suggests Nvidia’s manufacturing engagement is limited to preliminary advanced-packaging talks.
