Intel’s Foundry Lost $10 Billion in 2025 as CHIPS Act Billions Mask Strategic Collapse

Intel’s foundry unit reported a $10.318 billion operating loss on sales of $17.826 billion in 2025, a deterioration so severe it threatens to invalidate the company’s entire turnaround thesis—even as $7.86 billion in CHIPS Act funding papers over the structural failures. The losses quintupled from 2024, when the foundry lost $2.249 billion, creating a credibility crisis that government subsidies cannot resolve.

Intel’s full-year revenue dropped to $52.9 billion, the weakest result since 2010, while the company narrowly avoided a larger net loss only through CHIPS Act milestone payments that began flowing in Q4. Intel has already received $2.2 billion of its federal grant allocation, with disbursement tied to construction milestones rather than commercial viability. This creates a perverse incentive structure: build capacity regardless of customer demand, collect government checks, report progress.

The Capex Blind Spot Materializes

The semiconductor industry is experiencing what Barclays analysts describe as capital expansion at “tremendous rates” amid “multiple hundreds of billions a year” in infrastructure spending. Barclays estimates hyperscaler capital expenditure in 2027-2028 could exceed current forecasts by more than $225 billion, creating a supply-demand mismatch that rewards disciplined capacity builders.

Intel is not among them. The company expects $27 billion in 2025 Capex, focused on scaling 18A capacity while pausing development of the next-gen 14A node until a major foundry client is secured. This represents the inverse of capital discipline: spending billions on bleeding-edge nodes without confirmed customers, while government subsidies cushion downside risk.

Intel is investing $100 billion to expand its domestic operations across Arizona, New Mexico, Ohio, and Oregon. Fab 52 in Arizona has production capacity of 10,000 wafer starts per week—approximately 40,000 wafer starts per month at full ramp, equipped with four ASML Twinscan NXE Low-NA EUV lithography systems. The facility represents state-of-the-art American manufacturing capability. It also represents uneconomic capacity expansion at scale.

The contrast with TSMC exposes Intel’s problem. TSMC’s EUV-based N5 and N3 process technologies accounted for 63% of wafer revenue in Q4 2025, demonstrating commercial validation at scale. Intel’s advanced nodes represent 10% of foundry revenue, and external foundry revenue was $222 million in Q4—driven primarily by U.S. government contracts rather than market demand.

CHIPS Act as Subsidy for Failure

Intel also secured a $3 billion contract for the Secure Enclave program to expand trusted manufacturing of leading-edge Semiconductors for the U.S. government. Combined with direct funding and a 25% investment tax credit on qualified investments exceeding $100 billion, Intel is receiving approximately $33 billion in public support through grants, credits, and defense contracts.

This creates what one semiconductor analyst called “the nationalization of losses, privatization of upside”—except there is no upside yet. Most customer relationships appear to remain exploratory rather than binding, with the foundry segment posting roughly $7 billion in operating losses in 2023, followed by additional multi-billion-dollar losses through 2024 and 2025.

Geopolitical imperative is replacing market discipline. The U.S. government took a 10% stake in Intel with an $8.9 billion investment in August 2025, primarily from CHIPS Act grants. Additional investments from SoftBank ($2 billion) and Nvidia ($5 billion) followed, though Nvidia didn’t commit to using Intel’s foundry. The capital structure resembles a bailout dressed as Industrial Policy.

AI Infrastructure ROI Enters Question

Intel’s losses arrive as broader AI Infrastructure economics face scrutiny. CEOs are hesitant to invest in compute power capacity at maximum levels due to limited visibility into future demand, with many companies unsure whether large capital expenditures on AI infrastructure today will produce measurable ROI, according to McKinsey research.

IBM CEO Arvind Krishna said today’s figures for constructing and populating large AI data centers place the industry on a trajectory where roughly $8 trillion of cumulative commitments would require around $800 billion of annual profit simply to service the cost of capital, as reported by Tom’s Hardware. That ROI threshold looks increasingly difficult as enterprise adoption remains concentrated in narrow use cases.

Only 29% of executives can measure AI ROI confidently, while only 25% of AI initiatives deliver expected ROI and just 16% have scaled enterprise-wide, per IBM research. The gap between capex commitments and revenue realization is widening, not closing. Sequoia Capital asked, “Where is all the revenue?” noting “a big gap between the revenue expectations implied by the AI infrastructure build-out, and actual revenue growth”.

Intel’s foundry bet assumes this gap will close through government-subsidized capacity waiting for customers who may never arrive at sufficient scale. The company is building first-class fabs to serve a market that increasingly questions whether trillion-dollar infrastructure spend can generate returns.

The 18A customer mirage

Panther Lake is already in production, on track to meet customer commitments—but those customers are internal. For now, Intel’s only major customer is itself, notes CNBC. CEO Lip-Bu Tan stated Intel does not have any external customers that have committed to using the new manufacturing technology.

The pipeline remains thin. Microsoft has publicly confirmed collaboration on custom AI silicon, though specific process nodes have not been disclosed; AWS is working with Intel on custom AI fabric initiatives, but there is no confirmation that Trainium 3 will be manufactured on 18A; Nvidia’s engagement is limited to exploratory discussions about advanced packaging, and 18A testing appears to be paused.

The CEO expects some potential external customers to commit to using the fabrication process in the second half of 2026, with capacity deployment following. This timeline creates an 18-month window between massive capex commitments and commercial validation. Bank of America’s Vivek Arya warned that without a major external customer committing to 18A by late 2026, the foundry business may struggle to become self-sustaining.

The 14A node faces similar uncertainty. Intel is wishy-washy about 14A—promoting the technology on one hand while stating it may pull the plug if it can’t obtain enough foundry business, which doesn’t exude confidence in the foundry customer base, notes industry publication Semiecosystem.

What to Watch

The next 18 months will determine whether Intel’s foundry becomes a viable business or a subsidized monument to industrial policy failure. Three inflection points matter: