Essay No. 058 · Intel Turnaround & Foundry Proof
Intel Bet the Farm. Now It Has to Prove the Foundry. Original analysis Not investment advice
In 2021, Pat Gelsinger chose pain, capex, and "5 nodes in 4 years" instead of managing Intel for near-term margins. In 2026, the bet is no longer theoretical: 18A is in production, foundry losses are real, expansion is being disciplined, and Intel has to prove American leading-edge manufacturing can be economically viable.
The turnaround is no longer about ambition. It is about execution, utilization, and trust.
Intel's 2021 plan sounded insane because it probably had to.
After years of manufacturing misses, delayed process nodes, AMD share gains, Apple leaving Intel CPUs, and hyperscalers designing more of their own silicon, Intel had two paths. Protect margins, milk x86, cut spending, and slowly become less relevant. Or spend aggressively, rebuild process leadership, stand up Intel Foundry Services, and try to become a leading-edge foundry again.
Pat Gelsinger chose the second path. That was the farm bet. The 2026 question is no longer whether Intel can spend. It is whether Intel can earn trust, fill fabs, prove 18A, and turn foundry ambition into economics.
The 2021 bet was about process credibility. The 2026 test is economic credibility.
Section 01 What the 2021 Intel article got right
The 2021 SemiAnalysis piece on Intel's farm bet is the historical anchor for this essay[1]. It framed Pat Gelsinger as betting the company on future technological dominance against AMD pressure, Apple's decoupling, and hyperscaler custom silicon. The financial picture it described was unforgiving. Revenue was expected to fall from $77.7B to roughly $74B in 2022. Gross margins were expected to drop into the 51% to 53% range for several years. Capex was guided to $25B to $28B annually, with R&D around $15B per year. Intel had hired about 6,000 engineers in the first eight months after Gelsinger returned, and had reshuffled design, product, and manufacturing leadership.
The plan was the "5 nodes in 4 years" roadmap, running Intel 7, Intel 4, Intel 3, Intel 20A, and Intel 18A on a punishing cadence, with Intel Foundry Services as a core tenet of the strategy. The 2021 article was skeptical about execution but recognized that the alternative was long-term decline[1]. The chart on page 5 spelled out the long-term outlook: revenue above $74B in 2022, 10% to 12% revenue CAGR over the following 4 to 5 years, gross margins of 51% to 53% for 2 to 3 years, and $25B to $28B capex.
- Intel revenue expected to fall from $77.7B to ~$74B.
- Gross margins expected at 51-53% for several years.
- Capex guided to $25B-$28B annually.
- R&D guided to ~$15B annually.
- ~6,000 engineers hired in Pat's first eight months back.
- Design, product, and manufacturing groups reorganized.
- Roadmap: Intel 7, Intel 4, Intel 3, Intel 20A, Intel 18A.
- "5 nodes in 4 years" cadence.
- Intel Foundry Services framed as a core strategic tenet.
- Long-term outlook: 10-12% revenue CAGR over 4-5 years, 51-53% GM for 2-3 years, $25-28B capex.
The 2021 article was not wrong to be skeptical. Intel was promising one of the hardest turnarounds in semiconductor history. The honest scorecard four years later requires comparing those promises against the public record, not against the wish list.
Section 02 The 2021 financial outlook did not happen
Intel's FY2025 results put numbers on the gap between the 2021 outlook and the 2026 reality. FY2025 revenue was $52.9B, GAAP gross margin was 34.8%, non-GAAP gross margin was 36.7%, GAAP net income was a loss of roughly $0.3B, and non-GAAP net income was $1.9B. Full-year Intel Products revenue was $49.1B, full-year Intel Foundry revenue was $17.8B (mostly internal and intersegment), and Intel 18A ramped to high-volume manufacturing in Arizona and Oregon[2].
The original recovery model was too optimistic. Intel did not glide back to a 10% to 12% CAGR from a $74B base with 51% to 53% gross margins. It became smaller, lower-margin, and more financially constrained than the 2021 outlook implied. That is the honest financial scorecard.
Section 03 But the bet did not simply fail
The financial scorecard is not the whole scorecard. Intel's own newsroom confirms that Panther Lake is the first client SoC built on Intel 18A and that Panther Lake is in production, that Clearwater Forest is Intel's first 18A-based server processor, that Intel 18A uses RibbonFET and PowerVia, and that Intel claims up to 15% better performance per watt and 30% improved chip density versus Intel 3. Foveros provides the multi-chiplet integration path on top[5].
That bundle is the technical proof point the 2021 plan promised. 18A combines gate-all-around RibbonFET transistors with backside power delivery via PowerVia, in a high-volume product, on schedule for the late stretch of the 5-nodes-in-4-years cadence. Process credibility is no longer hypothetical.
18A makes the turnaround real. It does not make the turnaround complete.
Section 04 Products versus Foundry
The Q1 2026 numbers separate the technical story from the commercial story. Intel reported Q1 2026 revenue of $13.6B, up 7% year over year, with a non-GAAP gross margin of 41.0%. CCG revenue was $7.7B. DCAI revenue was $5.1B, up 22% year over year. Intel Foundry revenue was $5.4B[3]. Intel's Q1 prepared remarks added the harder edge: Intel Foundry operating loss was $2.4B, external foundry revenue was only $174M, and yields improved across Intel 4, Intel 3, and 18A while early-ramp costs continued to absorb foundry margin[4].
Intel Products
- Client CPUs (CCG) and data-center CPUs (DCAI).
- x86 ecosystem, software, and OEM relationships.
- Q1 2026: CCG $7.7B, DCAI $5.1B (+22% YoY).
- Carries most of the company's margin profile.
- Has to keep funding the foundry build-out.
Intel Foundry
- Intel 18A in HVM, with 14A on the roadmap.
- Advanced packaging, Foveros, multi-chiplet integration.
- Q1 2026: revenue $5.4B, operating loss $2.4B.
- External revenue $174M, the harder test.
- CHIPS Act funding and U.S. capacity expansion.
Intel Products is still the cash engine. Intel Foundry is still the strategic burden.
Section 05 The foundry business is not just manufacturing
The 18A ramp is a process achievement. It is not, by itself, a foundry achievement. A foundry has to clear a long checklist before external customers commit at scale. Competitive process technology. High yield. Stable cycle time. Clean design rules. EDA support. IP libraries. Customer support. Pricing discipline. Packaging integration. Capacity availability. Trust with external customers. And a credible answer to the most uncomfortable question, which is whether the foundry will manufacture chips for companies that compete with its own product line.
Process technology
Yield
EDA & IP
Packaging
External customers
Fab utilization
Cost discipline
Roadmap trust
Intel has manufactured world-class chips for itself before. The harder test is manufacturing world-class chips for others, repeatably, at a competitive cost, with the disciplines that a foundry-customer relationship demands.
Section 06 Lip-Bu Tan changed the strategy
The clearest change in the 2025 to 2026 cycle is leadership tone. Lip-Bu Tan's reset memo says Intel invested too much, too soon without adequate demand. The factory footprint became fragmented and underutilized. Intel decided not to move forward with previously planned projects in Germany and Poland. Intel planned to slow Ohio construction to align spending with demand. Job number one is ramping Intel 18A at scale. Future Intel 14A investment will be tied to confirmed customer commitments. The memo includes the line that defines the new posture: "There are no more blank checks. Every investment must make economic sense."[6]
The 2021 Intel was trying to prove it could still be bold. The 2026 Intel is trying to prove it can be bold and disciplined at the same time. Those are different skills.
Section 07 CHIPS Act support is real, but not enough
Intel received up to $7.86B in direct funding under the U.S. CHIPS Act, supporting manufacturing and advanced packaging projects in Arizona, New Mexico, Ohio, and Oregon, alongside a separate $3B Secure Enclave contract for trusted leading-edge manufacturing[7]. That confirms Intel's role in U.S. semiconductor industrial policy. It also lowers pressure on the company's balance sheet during the most capital-intensive years of the build-out.
Industrial policy can underwrite the construction phase. It cannot underwrite the customer phase. The CHIPS Act gets fabs built. It does not, by itself, fill them with paying external designs.
Subsidies can help build fabs. They cannot automatically fill them.
Section 08 NVIDIA partnership, important but easy to overstate
NVIDIA's strategic move into Intel is the most-discussed external signal of the past cycle. NVIDIA said it would invest $5B in Intel, with collaboration on custom data-center and PC products. Intel will build NVIDIA-custom x86 CPUs for NVIDIA AI infrastructure platforms, and x86 RTX SoCs for PCs that integrate NVIDIA RTX GPU chiplets[8].
That is meaningful. It is also easy to misread. The deal is a vote of confidence in Intel's x86 platform relevance, not a statement that NVIDIA is moving its main AI GPUs from TSMC to Intel Foundry. The press release describes custom CPUs and SoCs, not a wholesale change in NVIDIA's leading-edge GPU manufacturing. The honest reading is that the partnership helps Intel's product story more than it solves the foundry story.
Section 09 Altera sale shows the new discipline
The Altera transaction is the cleanest signal that Intel is operating in triage mode, not expansion-at-all-costs mode. Intel sold a 51% stake in Altera to Silver Lake, valuing Altera at $8.75B and retaining 49% of the company. Intel framed the move around focus, operational independence for Altera, and strengthening Intel's financial position[9].
The farm bet entered the pruning phase. Less ambitious M&A activity. More balance-sheet discipline. More focus on the parts of the company that actually need to perform.
Section 10 AI makes Intel's chance bigger and harder
AI infrastructure creates demand for CPUs that handle inference, orchestration, networking, and general compute alongside accelerators, for advanced manufacturing, for advanced packaging, for geographic supply diversity, for trusted U.S. manufacturing, and for capacity outside TSMC. All of that is good news for Intel. AI also raises the competitive bar. NVIDIA controls the accelerator platform. TSMC controls the most trusted advanced-manufacturing and CoWoS ecosystem. AMD is strong in server CPUs. Arm server CPUs and hyperscaler custom silicon keep growing. Intel has to compete on performance, cost, packaging, and platform trust at the same time.
The TSMC 2026 Technology Symposium frames the competitive context cleanly. TSMC is producing 5.5-reticle CoWoS, with 14-reticle CoWoS planned by 2028 and expected to integrate around 10 large compute dies and 20 HBM stacks in a single package[10]. ASML's Q4 FY2025 results add the macro backdrop, with strong 2025 results, Q4 net bookings of EUR 13.2B, and EUR 7.4B of EUV bookings, driven by AI-related customer capacity plans[11]. AI demand is real. Intel still has to execute into it.
Section 11 What people got wrong in 2021
The weak interpretation of the 2021 story was that Intel could simply spend its way back. That reading treats capex as strategy and a roadmap as a business model. Neither is true. The better interpretation is that Intel had to spend to stay in the game, and that spending only bought the right to prove execution.
Capex is not strategy by itself. R&D is not execution by itself. A roadmap is not a business model by itself. Foundry requires customers, utilization, cost discipline, and trust. The farm bet bought Intel a second chance. It did not guarantee the harvest.
Spend to survive
- Lower margins, higher capex, more R&D.
- "5 nodes in 4 years" cadence.
- Intel Foundry Services as a core tenet.
- $25B-$28B capex; ~$15B R&D.
- Bet framed as the only path to stay relevant.
Earn the foundry
- 18A in HVM, Panther Lake in production.
- Foundry operating loss ~$2.4B in Q1 2026.
- External foundry revenue still small ($174M Q1).
- No more blank checks. 14A tied to customer commitments.
- Altera 51% sold; Germany/Poland paused; Ohio slowed.
Pat Gelsinger announces aggressive investment and 5 nodes in 4 years
$25B-$28B capex, ~$15B R&D, 6,000 new engineers, and Intel Foundry Services as a strategic tenet[1].
Intel ramps Intel 7, Intel 4, Intel 3 and builds foundry
Process milestones land, IFS evolves into Intel Foundry, but financial reality remains harder than the 2021 outlook.
FY2025 results: smaller base, lower margins
FY2025 revenue $52.9B, GAAP GM 34.8%, non-GAAP GM 36.7%, GAAP net loss ~$0.3B, non-GAAP net income $1.9B[2].
Panther Lake on 18A enters production
First client SoC on Intel 18A using RibbonFET and PowerVia, with Foveros for multi-chiplet integration[5].
Lip-Bu Tan resets strategy: no more blank checks
Germany and Poland projects stopped, Ohio construction slowed, 14A tied to confirmed customer commitments[6].
Intel sells 51% of Altera to Silver Lake
Altera valued at $8.75B; Intel retains 49%. The farm bet enters the pruning phase[9].
Q1 product momentum, large foundry losses
Q1 revenue $13.6B (+7% YoY), DCAI +22% YoY; Intel Foundry operating loss ~$2.4B; external foundry revenue $174M[3][4].
Clearwater Forest and the 18A server proof phase
First 18A-based server processor, with the test being whether 18A scales from client to data center and from internal to external customers[5].
Section 12 The real scorecard for Intel now
The honest scorecard is symmetric. There is a set of conditions under which Intel's farm bet pays off, and a set under which it does not. Both lists are short and specific enough to be tracked over the next several years.
Execution converts ambition
- Panther Lake ramps successfully.
- Clearwater Forest proves 18A in servers.
- 18A yields continue improving.
- Intel Products regain competitiveness in client and data center.
- External customers commit to Intel Foundry.
- 14A earns commitments before massive spending.
- Foundry losses narrow.
- Packaging becomes a real external business.
- CHIPS-backed capacity gets utilized.
- Intel earns trust as a foundry, not only as an IDM.
Capacity outruns customers
- 18A ramps but does not attract external customers.
- Foundry remains mostly internal and loss-making.
- 14A cannot secure commitments.
- Products fail to regain share.
- Capital intensity overwhelms cash generation.
- TSMC remains the default for advanced AI silicon and packaging.
- Customers do not believe Intel can serve them like a neutral foundry.
- Subsidies do not translate into utilization.
Section 13 Risks and limits
The argument above relies on a mix of Intel company materials, partner press releases, and government and industry data. It is worth being explicit about where the case can break.
Intel official claims are company claims and should be treated as guidance, not independent benchmarks.
18A in production does not automatically prove broad external foundry adoption.
Intel Foundry revenue still includes large internal and intersegment activity, which can flatter the headline number.
External foundry revenue is still small in Q1 2026 and may stay small for multiple generations.
Foundry operating losses may stay high through the 18A and 14A ramps.
CHIPS Act funding reduces pressure but does not solve fab utilization.
The NVIDIA partnership should not be misread as a main-GPU foundry win.
AI CPU demand may help Intel but does not guarantee share gains against AMD and Arm.
TSMC remains the manufacturing-trust benchmark; Samsung and others remain serious foundry competitors.
Intel's turnaround depends on product execution and foundry execution running in parallel, which is unusually demanding.
The point is not that Intel has failed. The point is that the proof phase is harder than the spending phase.
Section 14 Final verdict
In 2021, Intel chose the only path that gave it a chance. It accepted lower margins, higher capex, more R&D, and massive execution risk. In 2026, that bet has produced real technical proof in 18A and Panther Lake. But the financial proof is still missing. Intel Foundry is losing billions. External revenue is still small. Expansion is being disciplined. 14A must be tied to customer commitments. The company is no longer trying to prove it can spend. It is trying to prove it can manufacture profitably for others.
The 2021 article was right to argue Intel had to bet. The 2026 question is whether the harvest matches the bet. That answer is still being written, but the test is now economic, not ambitious.
The farm bet bought Intel a second chance. It did not guarantee the harvest.
Pat bet the farm. Lip-Bu has to make the farm produce.
Section 15 Evidence ledger and source notes
| Source | Claim | Why it matters |
|---|---|---|
| SemiAnalysis (2021) | Revenue from $77.7B to ~$74B; 51-53% GM; $25-28B capex; ~$15B R&D; 6,000 engineers; 5 nodes in 4 years; IFS as core tenet. | Anchors the original farm-bet outlook against which the 2026 reality is judged. |
| Intel FY2025 results | $52.9B revenue; GAAP GM 34.8%; non-GAAP GM 36.7%; GAAP net loss ~$0.3B; non-GAAP net income $1.9B; Products $49.1B, Foundry $17.8B; 18A in HVM in AZ and OR. | Quantifies the gap between 2021 outlook and 2025 reality. |
| Intel Q1 2026 results | Revenue $13.6B (+7% YoY); non-GAAP GM 41.0%; CCG $7.7B; DCAI $5.1B (+22% YoY); Foundry $5.4B; Q2 guide $13.8-14.8B. | Shows product momentum returning while foundry remains separate. |
| Intel Q1 2026 prepared remarks | Intel Foundry operating loss $2.4B; external foundry revenue $174M; better yields across Intel 4, Intel 3 and 18A. | The honest commercial scorecard for the foundry strategy. |
| Intel Panther Lake / 18A | Panther Lake first 18A client SoC in production; Clearwater Forest first 18A server CPU; 18A uses RibbonFET and PowerVia; +15% perf/W and +30% density vs Intel 3. | Confirms technical milestones and the GAA-plus-backside-power node. |
| Lip-Bu Tan reset memo | Too much, too soon; Germany and Poland stopped; Ohio slowed; 18A is job one; 14A tied to customer commitments; no more blank checks. | Defines the new operating discipline. |
| Intel CHIPS Act award | Up to $7.86B direct funding; AZ, NM, OH, OR projects; separate $3B Secure Enclave contract. | Industrial-policy support, not commercial proof. |
| NVIDIA-Intel partnership | $5B NVIDIA investment; custom x86 CPUs for NVIDIA AI platforms; x86 RTX SoCs with NVIDIA GPU chiplets for PCs. | Product-side vote of confidence, not a leading-edge GPU foundry win. |
| Altera / Silver Lake | Sold 51% of Altera; Altera valued at $8.75B; Intel retained 49%. | Evidence of focus and balance-sheet discipline in the pruning phase. |
| TSMC 2026 Symposium | 5.5-reticle CoWoS today; 14-reticle CoWoS by 2028 with ~10 compute dies and 20 HBM stacks. | Sets the competitive context Intel has to clear in advanced packaging. |
| ASML Q4 FY2025 | Strong 2025 results; Q4 net bookings EUR 13.2B; EUV bookings EUR 7.4B; AI-driven medium-term demand. | Macro evidence that AI manufacturing demand is real even as Intel restructures. |
Footnotes & sources
- SemiAnalysis, “Intel Betting The Farm — Shrinking Business, Margins Down For Few Years, But Aggressively Investing $40B-$43B A Year And More With Subsidies,” 2021 (PDF supplied by author). Source for the $77.7B-to-$74B revenue framing, the 51-53% gross margin range, $25-28B capex, ~$15B R&D, the 6,000-engineer hiring figure, the design and manufacturing reorganization, the 5-nodes-in-4-years roadmap, the Intel Foundry Services framing, and the long-term outlook chart on page 5.
- Intel Investor Relations, “Intel Reports Fourth Quarter and Full-Year 2025 Financial Results,” intc.com/…/2025. Source for FY2025 revenue, GAAP and non-GAAP gross margin, GAAP net loss, non-GAAP net income, Intel Products and Intel Foundry revenue, and the Intel 18A high-volume manufacturing milestone in Arizona and Oregon.
- Intel Investor Relations, “Intel Reports First Quarter 2026 Financial Results,” intc.com/…/2026 Q1. Source for Q1 2026 revenue, year-over-year growth, non-GAAP gross margin, CCG and DCAI revenue, Intel Foundry revenue, and Q2 2026 revenue guidance.
- Intel Q1 2026 prepared remarks (Intel investor relations). Source for the Intel Foundry operating loss figure, the external foundry revenue figure, the yield-improvement comments across Intel 4, Intel 3 and 18A, and the early-ramp cost commentary used in the Products-versus-Foundry section.
- Intel Newsroom, “Intel Unveils Panther Lake Architecture — First AI PC Platform Built on 18A,” newsroom.intel.com. Source for Panther Lake as the first 18A client SoC in production, Clearwater Forest as the first 18A server CPU, the RibbonFET and PowerVia framing, the up-to-15% perf/W and 30% density claim versus Intel 3, and the Foveros multi-chiplet integration framing.
- Intel Newsroom, Lip-Bu Tan, “Steps in the Right Direction,” newsroom.intel.com/corporate/lip-bu-tan-steps-in-the-right-direction. Source for the too-much-too-soon framing, the Germany and Poland pause, the Ohio slowdown, the 18A-is-job-one framing, the 14A-tied-to-customer-commitments policy, and the no-more-blank-checks language.
- Intel Newsroom, “Intel and CHIPS Act,” newsroom.intel.com/corporate/intel-chips-act. Source for the up to $7.86B in direct CHIPS Act funding, the AZ, NM, OH, and OR footprint, and the separate $3B Secure Enclave contract.
- NVIDIA Newsroom, “NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products,” nvidianews.nvidia.com. Source for the $5B NVIDIA investment in Intel, the custom data-center and PC product collaboration, the Intel-built NVIDIA-custom x86 CPUs for NVIDIA AI infrastructure platforms, and the x86 RTX SoCs for PCs with NVIDIA RTX GPU chiplets.
- Intel Newsroom, Altera and Silver Lake transaction page, newsroom.intel.com/corporate/intel-partner-deal-news-april2025. Source for the sale of a 51% stake in Altera to Silver Lake, the $8.75B Altera valuation, and Intel's 49% retained stake.
- TSMC, “TSMC 2026 Technology Symposium,” pr.tsmc.com/english/news/3302. Source for 5.5-reticle CoWoS in production today and 14-reticle CoWoS planned by 2028 with around 10 compute dies and 20 HBM stacks. Used as competitive context, not as an Intel claim.
- ASML, “Q4 2025 Financial Results,” asml.com/en/news/press-releases/2026/q4-2025-financial-results. Source for the strong 2025 results, Q4 net bookings of EUR 13.2B, EUV bookings of EUR 7.4B, and the AI-related customer capacity plans driving medium-term demand.