Intel: why the profit is so thin

Start with the arithmetic. In the trailing twelve months through the third quarter of 2026, Intel took in about $59.9 billion of sales and kept roughly $1.76 billion of operating profit — about 2.9 cents of profit from every $1 of sales. Divide a profit that small by a capital base as large as Intel's, and you land at a return on invested capital of roughly 1.4%: $1.40 per $100 invested.

Section 01The five drains on the profit

1. The factories lose about $2 billion a quarter

The single biggest drain is Intel Foundry, the manufacturing segment. It lost $10.3 billion in 2025, $13.3 billion in 2024 (restated basis), and about $8.9 billion over the twelve months through Q3 2026. In the third quarter of 2026 alone, Foundry lost $1,853 million on $6,697 million of revenue — a −28% operating margin. In the same quarter, Intel's chip-design businesses earned $5,324 million of operating profit. The factories, plus about $1.1 billion of corporate costs, consumed most of it, leaving $2,522 million for the whole company.

The detail that matters most: of Foundry's $6,697 million of third-quarter revenue, roughly $6.3 billion was Intel selling chips to itself. Only $411 million came from outside customers — and Intel's own filing says that increase over last year came mainly from one non-recurring contract. For all of 2025, outside customers supplied just $307 million of Foundry's $17.8 billion of revenue.

2. The margin on each chip collapsed

Gross profit — what is left from each $1 of sales after the direct cost of making the chips — was once Intel's superpower. It has been cut nearly in half from its 2018 level, and the partial recovery since 2024 rests on one strong quarter: Intel guided the fourth quarter of 2026 back down to 37.1%.

Full-year gross margin (GAAP) from Intel's annual results. The 43.0% point is a single quarter — Q3 2026 — plotted separately; Intel guided Q4 2026 gross margin to 37.1%. Causes Intel itself discloses: under-used factories spreading fixed costs over fewer wafers, the higher cost of its newest 18A wafers, and inventory-related charges.

3. Depreciation on a mountain of new factories

Intel spent $25.75 billion on plant and equipment in 2023, $23.94 billion in 2024 and $14.65 billion in 2025, and guided spending back up to about $20 billion in 2026 and "significantly" more in 2027 — with $19.6 billion of further spending commitments outstanding at the end of Q3 2026. Every one of those buildings and machines is written down year by year: depreciation ran to $8,933 million in the first nine months of 2026 alone, up 12% on a year earlier. The factories were built before outside customers arrived to fill them.

4. Restructuring and write-offs keep recurring

The twelve-month margin also carries a $3.9 billion write-down of Mobileye goodwill in early 2026. It follows a third quarter of 2024 that absorbed $15.9 billion of impairment charges plus $2.8 billion of restructuring, and a $1.9 billion restructuring charge in a single quarter of 2025. Intel's own guidance puts 2026 operating expenses at $22.6 billion on a GAAP basis, including $4.3 billion of restructuring.

5. The denominator: $228.6 billion of assets

At September 26, 2026, Intel's balance sheet carried $228.6 billion of total assets, including $105.3 billion of plant and equipment and $20.5 billion of goodwill, against about $50.1 billion of debt and $54.0 billion of cash and short-term investments. Even a normal chip profit would look small against a base that size. A $1.76 billion profit is roughly 1.4% of it.

Section 02The latest quarter, segment by segment

Intel reported its third quarter of 2026 on October 29. The pattern is the whole story in miniature: the data-centre business is booming, the PC business is shrinking, and the factory loses money on every wafer it sells to itself.

Corporate-unallocated costs and inter-segment eliminations (−$6,291M of revenue) reconcile the segments to the consolidated line. Source: Intel Q3 2026 earnings release, October 29, 2026.

• PC chips: revenue fell 5% because unit volume fell 14% on weaker demand and competition, partly offset by prices up 9%. Management guided PC-chip revenue down more than 20% in Q4 and expects the PC slowdown to last at least into the first half of 2027.
• Data centre: the strongest growth in 15 years, per Intel's CFO — server prices up 54% and volume up 10% on hyperscaler demand. Intel says demand "continues to meaningfully exceed available supply": in servers, it is short of capacity, not customers.
• Company totals: gross margin 43.0%, earnings per share $0.60 (GAAP) / $0.52 (adjusted), operating cash flow $5,082M. Guidance for Q4 2026 — revenue $15.0–16.0 billion and gross margin 37.1% — came in well below the roughly $17.1 billion analysts expected, and the shares fell 28.58% to $65.76 in response.

Section 03How Intel got here: a decade in short

It lost the technology lead. Intel's 10-nanometre process slipped about three years, reaching volume only in 2019, while TSMC — fed orders by Apple, AMD and others — shipped 7nm in 2018 and 5nm in 2020. The manufacturing cadence that had been Intel's moat broke, and the lead passed to Taiwan.

It missed the phone, then lost Apple. Intel never won a real place in smartphones; its one toehold, iPhone modem chips, ended in 2019 when it sold most of that business to Apple for $1 billion, at what it called a multibillion-dollar loss. In 2020 Apple announced the Mac would move to Apple's own chips, and took its flagship endorsement with it.

AMD took the profitable end. Building on TSMC's factories, AMD's server chips took about 41% of server CPU revenue by late 2025 (Mercury Research), and by mid-2026 Intel's share of x86 processors had fallen below 70% for the first time since 1995.

Then it bet the company on catching up. From 2021, under CEO Pat Gelsinger, Intel's IDM 2.0 plan poured roughly $95 billion of spending (2021–2024) into new factories and a foundry business selling capacity to outsiders — "five nodes in four years." The bill arrived before the customers: Intel lost $18.8 billion in 2024, its first annual loss since 1986, and the stock fell about 60% that year. Gelsinger left in December 2024.

The rescue era. Lip-Bu Tan became CEO in March 2025 and cut: headcount from 124,800 (end of 2023) to 81,100 by September 2026, the German and Polish factory projects cancelled outright in July 2025, and the Ohio complex slowed to 2030–31. Three backers wrote cheques: the US government took 9.9% of Intel for $8.9 billion (August 2025, funded from CHIPS Act money, a passive stake with no board seat), SoftBank put in $2 billion, and Nvidia put in $5 billion alongside a product partnership — which, notably, did not commit Nvidia to use Intel's factories.

And in October 2026, an accident. On October 14, an explosion during a gas delivery at Fab 28 in Kiryat Gat, Israel killed two workers and halted the plant for about two weeks. Intel's quarterly filing — not its earnings release — disclosed the accident; management estimated roughly $300 million of lost revenue in Q4 2026 and a similar hit in Q1 2027, and said the full financial impact could not yet be estimated.

Section 04What Intel says has to change

Intel's own answer is consistent: the factories must fill with other companies' chips. Everything else is arithmetic around that single variable.

• Outside foundry revenue. Intel's CFO has said foundry operating break-even — once targeted for 2027 — needs only "a few billion dollars" of additional outside revenue. Set against the current pace, that is the gap: about $307 million of outside revenue in all of 2025. Intel's own investor-relations chief warned in November 2025 that winning its next-generation 14A process a big customer would require spending well ahead of revenue and would likely push break-even past the end of 2027. The 2027 date was not reaffirmed in the Q3 2026 materials.
• No blank cheques for 14A. Intel's filing language: the scale and pace of factory expansion "will ultimately be dictated by the amount of committed demand for Intel 14A" it can obtain, and it could pause or discontinue 14A if it cannot secure a significant external customer. The internal yardstick is at least a 20% return on investment per process generation.
• Customers — named vs rumoured. The only publicly confirmed external name for 14A is Tesla's Terafab project, which Intel joined in April 2026 — a partnership with no binding purchase agreement, while Terafab has also talked to TSMC. Fortinet (July 2026) is the first named outside foundry customer of the Tan era, on an older process. Reports in October 2026 that Google had ordered millions of chips from Intel Foundry for 2028 were not confirmed — Reuters could not verify them and the companies declined to comment. We do not count them as customers.
• The spending tension. At the same time as it cuts, Intel is raising factory spending again — about $20 billion guided for 2026 and "significantly" more in 2027, with no positive free cash flow expected in 2027. Intel's stated logic: finish the capacity, and be ready if committed outside demand arrives in the window management has named — decisions expected from late 2026 into early 2027.

Section 05The numbers, in one place