| What the business is | Intel designs and sells x86 processors for PCs (Client Computing and Physical AI Group, CCPG) and servers (Data Center and AI, DCAI) to PC makers and cloud providers, and runs its own fabs through Intel Foundry, which also seeks outside customers. It owns about 77% of Mobileye and 49% of Altera (Q2 2026 10-Q). FY2025 revenue was $52.9B with 85,100 employees (FY2025 10-K). |
| Industry | Semiconductors: x86 CPU design, a legal duopoly with AMD, plus leading-edge logic foundry, where TSMC took 72.5% of top-10 foundry revenue in Q2 2026 (TrendForce). |
| How it makes money | Intel sells processors on cancellable purchase orders to OEMs, ODMs and cloud providers, net of price concessions, rebates and co-marketing deducted before revenue is reported; the fabs are paid internally at transfer prices that approximate market, so all external cash comes from chip sales. |
| Unit of economics | One processor. Unit counts are not disclosed (unknown); in Q2 2026 client units fell 8% and server units rose 9% (Q2 2026 10-Q). In FY2025 Intel Products earned 25.9% operating margin after paying for wafers, while Intel Foundry lost $10.3B on $17.8B of mostly internal revenue (FY2025 10-K). |
| What protects it | The x86 cross-licence means no third party can make compatible CPUs, and Windows and enterprise software is written for x86; a well-funded entrant cannot buy either. The only US leading-edge logic fabs, backed by a US government shareholder, are a second, policy-based protection that does not yet earn a return. |
| What drives earnings | Server CPU demand from AI inference; processor prices and mix, now lifted by shortage; factory loading and the cost of new nodes (18A); the scale of external foundry and packaging revenue. |
| What to watch | Intel Foundry external revenue and operating loss each quarter; whether a significant external customer commits to Intel 14A; server ASP and x86 share (Mercury Research) once supply normalises. |
| Cycle exposure | PC: high — industry units went from 339.8M (2021) to 241.8M (2023) (Gartner). Server: medium-high, currently in a supply-constrained upswing. Manufacturing: very high operating leverage from fixed fab costs. Now: revenue recovering (1H 2026 +16%) with gross margin at 39.9%, mid-way between the FY2024 trough and the FY2017 peak. |
US dollars. Intel's fiscal year ends on the last Saturday of December; latest full year FY2025 (ended 27 Dec 2025), latest quarter Q2 2026 (ended 27 Jun 2026). (inferred) marks a conclusion reasoned from sourced facts rather than a disclosed figure.
Every PC, every conventional server and much of the cloud needs a general-purpose processor that runs the software its owner already has. For most of that software the instruction set is x86, and Intel is its largest supplier. It sells the processor, plus chipsets, networking silicon and custom chips, to PC makers (OEMs and ODMs) and cloud service providers, who build it into systems (FY2025 10-K, Business). Through Intel Foundry it is also trying to sell the factory behind those chips — wafer fabrication and advanced packaging — to other chip designers.
The unit is one processor. It starts as a wafer processed in an Intel fab or, for some of Intel's most advanced products, at TSMC. Intel 7 carried the majority of Intel's 2025 internal production by revenue, and Intel 18A entered high-volume production at the start of 2026 (FY2025 10-K; Q2 2026 10-Q). Since Q1 2024 Intel Foundry bills each wafer and each assembly-and-test service to Intel Products at prices "intended to approximate market pricing", recognising revenue at wafer sort (FY2025 10-K, Note 3).
Intel Products then sells the finished CPU on purchase orders that customers "can typically cancel, change or delay … with little or no notice to us and without penalty" (FY2025 10-K, Business). Price concessions, rebates and co-marketing are estimated and netted off before revenue is reported (FY2025 10-K, Note 2). Some receivables are sold without recourse, $2.6B in 2025, which pulls cash forward (FY2025 10-K, Note 6). Revenue is therefore transactional, not contracted.
| Segment | What it sells | FY2025 revenue | FY2025 op. income | 1H 2026 revenue | 1H 2026 op. income |
|---|---|---|---|---|---|
| CCPG (ex-CCG) | Notebook and desktop CPUs, chipsets | $32.2B | $9.3B | $16.6B | $4.9B |
| DCAI | Xeon server CPUs, custom ASICs | $16.9B | $3.4B | $11.3B | $4.0B |
| Intel Foundry | Wafers, packaging, test; ~98% to Intel | $17.8B | $(10.3)B | $11.2B | $(4.5)B |
| All other | Mobileye, IMS; Altera to Sep 2025 | $3.6B | $0.3B | $1.3B | $0.3B |
Sources: FY2025 10-K Note 3 (recast for the Q1 2025 fold of NEX into CCG and DCAI); Q2 2026 10-Q Note 2. Eliminations and corporate costs not shown: corporate unallocated was $(5.5)B in FY2025 and $(6.5)B in 1H 2026, the latter including a $3.9B Mobileye goodwill impairment. External foundry share (inferred) = $307M of $17.8B.
Demand is concentrated. Three unnamed customers bought 43% of FY2025 revenue (19%, 12% and 12%), almost all of it Intel Products (FY2025 10-K, Note 3). By billing location, the US was 30% of FY2025 revenue, China 24%, Singapore 18% and Taiwan 15% (FY2025 10-K; percentages inferred). Billing location follows where OEMs and ODMs are invoiced, so it overstates Asia as an end market (inferred).
Over the decade Intel shrank back to its core and sold assets to fund its fabs. It sold NAND memory to SK hynix for $9.0B in two closings (2021 and 2025), McAfee (2017) and most of its smartphone-modem business (2019), and 51% of Altera to Silver Lake in September 2025 for $4.3B net, booking a $5.6B pre-tax gain (FY2025 10-K, Note 10). Altera ran at a 55% gross margin in 1H 2025 against Intel's 32% (8-K 2025-09-15; Q2 2026 10-Q), so it was sold for cash, not because it was a weak business (inferred). Mobileye has been sold down to about 77% ownership (Q2 2026 10-Q, Note 3).
A processor passes through instruction-set IP (the x86 licence, or Arm), chip design, fabrication, packaging, the system maker and finally the end user. Profit collects at two chokepoints: scarce IP and the leading-edge fab share leader. Intel owns a slice of the first and is losing money trying to reach the second.
| Company / segment | Period | Gross margin | Operating margin |
|---|---|---|---|
| NVIDIA | FY ended Jan 2026 | 71.1% | ~60% |
| TSMC | FY2025 / Q2 2026 | 59.9% / 67.7% | 50.8% / 60.3% |
| Qualcomm | FY ended Sep 2025 | 55.5% | ~28% |
| AMD | FY2025 | 50% | 11% |
| Intel Products (CCG + DCAI) | FY2025 | not disclosed | 25.9% |
| Intel Foundry | FY2025 | not disclosed | −57.9% |
Sources: company releases (NVIDIA 25 Feb 2026; TSMC 4Q25 and 2Q26 management reports; Qualcomm Nov 2025; AMD 3 Feb 2026); Intel FY2025 10-K Note 3. Operating margins marked ~ are computed from reported operating income.
The table explains Intel's consolidated result. Measured as if it were fabless, Intel's design arm earns more than AMD as a whole, although AMD's margin is depressed by its GPU ramp, so the comparison flatters Intel (inferred). The factory's loss then consumes most of that profit.
x86 is a legal duopoly. The 2009 Intel–AMD cross-licence is royalty-free, lasts until the last patent expires, and terminates for both parties on a change of control (AMD 8-K, Nov 2009). Within it, Intel's share has fallen steadily. Mercury Research put Intel at 69.7% of x86 client units and 65.5% of server units in Q2 2026; AMD's client share rose from 23.9% to 30.3% in a year (Mercury via Tom's Hardware, Aug 2026). Counting only Xeon against EPYC, AMD had 46.4% of server units, and it took 46.2% of server revenue in Q1 2026 (Mercury).
The bigger long-run threat bypasses x86 altogether. Arm-based chips reached a record 15.3% of client and 13.6% of server CPU units in Q2 2026 (Mercury). For the third year running, more than half of new AWS CPU capacity was its own Arm-based Graviton (Amazon, Dec 2025). Intel concedes the trend: "we have lost market share in recent years, including in both client and data center markets," and in AI accelerators "we have been unsuccessful to date in becoming a meaningful participant" (FY2025 10-K, Risk Factors).
TSMC took 72.5% of top-10 foundry revenue in Q2 2026 at a 60.3% operating margin, and raised its 2026 capex guidance to $60–64B (TrendForce, Sep 2026; TSMC 2Q26). Intel guides to more than $20B (Intel 2Q26 remarks). Being second at the leading edge has not paid: Samsung's foundry has run at an estimated loss, and GlobalFoundries earned a 9.7% operating margin in Q2 2026 (TrendForce, Jan 2026; GlobalFoundries 2Q26 release). In the 2023 downturn TSMC's wafer shipments fell 20.1% but its operating margin stayed at 42.6% (TSMC 4Q23 report), so leading-edge pricing held for the leader (inferred).
Intel Foundry's external revenue was $293M in Q2 2026, mostly Altera, below the tenth-largest foundry's $0.43B (Q2 2026 10-Q; TrendForce). Confirmed external design wins are Microsoft and AWS on 18A (2024) and Fortinet on Intel 4 (July 2026); a reported Apple deal is press rumour only (9to5Mac, May 2026). The binding barriers are not only capital. Yields, design-tool and IP ecosystems, and customer trust all bind, and Intel's product arm competes with the designers it wants as customers.
“We have been unsuccessful to date in securing any significant external foundry customers for any of our nodes.”
Intel FY2025 10-K, Risk Factors
The first protection is the x86 franchise itself. No third party can obtain an x86 licence, and Windows, enterprise applications and OEM product lines are built around it. That barrier still binds in PCs and on-premise servers, where Intel kept about two-thirds of x86 units. It is eroding in the cloud, where hyperscalers control their own software and can recompile for Arm (inferred from Mercury and AWS data).
The second is location and policy. Intel describes itself as "the only company conducting both leading-edge semiconductor logic R&D and related high-volume manufacturing" in the US (FY2025 10-K). The US government now owns about 10% of Intel (433.3M shares for $8.9B, about $20.47 each; 8-K 2025-08-25, inferred from its terms), and the $3.2B Secure Enclave programme is government-directed demand. This protection is real but narrowing, because TSMC's Arizona plan has grown to $265B (TSMC). It also does not yet earn a return.
| Intel's claim | Outside evidence | Verdict |
|---|---|---|
| Only US company doing leading-edge logic R&D and volume manufacturing | US produced no sub-10nm logic in 2022 (SIA); TSMC Arizona expanding | Supported today, narrowing |
| Well positioned in growing CPU demand | DCAI revenue +59% in Q2 2026, but AMD gained share in every segment (Mercury) | Partially supported |
| "Second-largest foundry by 2030" (Apr 2024) | External revenue $293M a quarter vs Samsung $3.26B (TrendForce) | Contradicted on external revenue |
| 18A a leadership node | In high-volume production; yields at industry standard targeted for 2027; no major external 18A wafer customer | Technically plausible, commercially unproven |
Verdicts are this memo's reading of the evidence (inferred).
In CPUs the winner delivers a competitive design on time on the best available process; AMD went from about 1–2% of server units in 2018 to 34.5%, and from one desktop CPU in ten sold in 2016–18 to more than one in three, while Intel's process slipped (Mercury). In foundry the winner is the share leader whose scale funds the next node and whose customers trust it not to compete with them. Intel is the incumbent in the first game and a subscale entrant in the second, trying to fund both from one income statement (inferred).
From FY2021 to FY2025 Intel shrank: revenue fell from $79.0B to $52.9B. Notebook units fell 36% in FY2022, and server units fell 16%, 37% and 10% in FY2022–24 (10-Ks). In FY2025 server units recovered 9% but server ASPs fell 4% because of "pricing actions … driven by a competitive environment" (FY2025 10-K).
The 2026 rebound is price, not volume. In Q2 2026 client revenue rose $1.1B on ASPs up 27% while units fell 8%. Server revenue rose $2.0B on ASPs up 48% with units up 9% (Q2 2026 10-Q). Intel attributes most of the ASP gain to a richer mix of premium products, with "demand-based pricing actions contributing to a lesser extent, in part to offset higher input costs" — while "market demand exceeded our available product supply".
| Period | Reported growth | Excluding Altera | Basis |
|---|---|---|---|
| FY2025 vs FY2024 | −0.5% | −0.7% | Intel Products only; Altera's full-year contribution not disclosed |
| 1H 2026 vs 1H 2025 | +16.4% | +18.7% | Removes Altera: $779M consolidated in 1H25, $320M as foundry customer in 1H26 |
| Q2 2026 vs Q2 2025 | +25.4% | +28.3% | Removes Altera: $428M in Q2 2025, $181M in Q2 2026 |
Sources: FY2025 10-K; Q2 2026 10-Q, MD&A. Ex-Altera rates are inferred arithmetic. Acquisitions were immaterial; the adjustment is a divestiture, so organic growth runs above reported.
The growth drivers, ranked from most to least impactful:
The largest source of 2026 growth. ASPs rose while Intel could not meet demand because of Intel 7 and Intel 3 wafer limits and industry shortages of substrates and memory. Intel expects client constraints to "ease over the second half of 2026" (Q2 2026 10-Q). Customers paid $1.7B of deposits in Q1 2026 to secure future product supply (Q2 2026 10-Q). FY2025 showed the reverse case: with supply adequate and AMD competing, server ASPs fell 4%.
Server units rose 9% in Q2 2026 on hyperscaler demand (Q2 2026 10-Q), and Mercury put x86 server units up about 20%. Intel's CFO said CPU-to-GPU ratios moved from 1:8 to 1:4 (Q1 2026 call, via Tom's Hardware), and AMD doubled its 2030 server-CPU market estimate to $120B. Structural (inferred) but untested through a cycle, and Intel's share of the gain is shrinking.
AMD gained about 6 points of x86 client unit share in a year, and Arm set records in both client and server (Mercury, Q2 2026). Intel grows only where the market grows faster than its share falls.
Gartner forecasts 2026 PC shipments down 10.4% as DRAM and SSD prices rise about 130%; IDC reported Q2 2026 units down 4.9% (Gartner, Feb 2026; IDC, Jul 2026). Mix shifts toward premium systems, which supports Intel's ASP but not its volume.
External revenue was $467M in 1H 2026, about $1.2B annualised on the Q2 run-rate (inferred), and mostly Altera. Management points to an EMIB-T packaging backlog with customer ramps in 2027 (Intel 2Q26 remarks). No packaging customer is named.
Intel and NVIDIA agreed to co-develop custom x86 data-center CPUs and x86 PC chips with NVIDIA graphics over "multiple generations" (FY2025 10-K). No revenue is disclosed yet.
Chip fabs are largely fixed cost, so Intel's margin moves with factory loading and the cost of each new node. Gross margin fell from 62.3% in FY2017 to 32.7% in FY2024 as revenue fell and costs rose. Depreciation climbed from $7.8B (FY2023) to $10.8B (FY2025), and $34.5B of construction in progress still has to be depreciated once in service (FY2025 10-K). New nodes start expensive: 2025 carried $878M of inventory reserves on early 18A output, and in Q2 2026 Foundry's product profit fell $340M because of "higher-cost wafers manufactured on our Intel 18A process node" (FY2025 10-K; Q2 2026 10-Q).
Two other costs weigh on margin. Unused or obsolete capacity creates impairments and accelerated depreciation: $3.3B in FY2024, mostly Intel 7, and $950M in FY2025. Products built at TSMC "have lower margins" because TSMC's profit sits inside Intel's cost (FY2025 10-K, Risk Factors). The internal split isolates the problem: Intel Products' operating margin ran 22.9%, 26.3%, 25.9% and 31.8% across FY2023, FY2024, FY2025 and 1H 2026, while Intel Foundry's ran −38.3%, −76.7%, −57.9% and −40.5% (FY2025 10-K recast; Q2 2026 10-Q). Transfer prices are Intel's own estimate of market, so the split is indicative rather than exact (inferred).
Management has cut the cost base hard. Employees fell from 131,900 (FY2022) to 85,100 (FY2025), including Altera's exit (10-Ks). R&D plus MG&A fell 17% to $18.4B in FY2025, the FY2026 non-GAAP opex target is $16.0B (8-K 2025-09-15), and opex fell to 28.2% of revenue in Q2 2026 from 37.5% (Q2 2026 10-Q). Q2 2026 operating income was $1.8B against a $3.2B loss a year earlier; the GAAP net loss of $11.0B came from a $12.5B non-cash mark on the government's escrowed shares, which rises with Intel's share price.
| $B unless stated | FY2017 | FY2021 | FY2024 | FY2025 |
|---|---|---|---|---|
| Revenue | 62.8 | 79.0 | 53.1 | 52.9 |
| Gross margin | 62.3% | 55.4% | 32.7% | 34.8% |
| Operating income | 18.1 | 19.5 | (11.7) | (2.2) |
| Operating cash flow | 22.1 | 29.5 | 8.3 | 9.7 |
| Capex (additions to PP&E, investing) | 11.8 | 18.7 | 23.9 | 14.6 |
| Diluted shares (millions) | 4,835 | 4,090 | 4,280 | 4,530 |
Sources: 10-Ks FY2018–FY2025. Comparability: FY2017 and FY2021 include the since-sold NAND memory business and Altera; FY2021 operating cash flow is the recast $29.5B (originally $30.0B) after a 2022 reclassification; FY2024 and FY2025 capex exclude $1.2B and $3.0B of equipment financed through extended vendor terms; segment definitions changed every year from 2022 to 2025.
Cash generation no longer covers investment. Adjusted free cash flow was −$11.9B in FY2023, −$2.2B in FY2024 and −$1.6B in FY2025 (FY2025 10-K). The gap was bridged by partners who fund fabs in return for a share of their output economics: Brookfield in Arizona ($5.1B contributed in 2025) and, until April 2026, Apollo in Ireland. Government money also helped: $1.6B of capital incentives in 2025 plus $5.4B of US advanced-manufacturing tax credits netted against PP&E. In 1H 2026 operating cash flow of $8.1B exceeded gross capex of $7.6B (Q2 2026 10-Q).
Over FY2016–FY2025 Intel generated $216.7B of operating cash and spent it in this order: capex $175.0B (plus $6.2B recorded elsewhere), buybacks $47.2B, dividends $43.0B, acquisitions $33.9B, chiefly Altera and Mobileye (10-K cash-flow statements FY2016–FY2025, summed). The two halves tell different stories. In FY2016–20, $142.4B of operating cash funded $67.1B of capex and $71.4B of shareholder returns. In FY2021–25 capex of $107.9B exceeded operating cash of $74.3B.
About $65B of outside capital filled the FY2022–25 gap: partner capital $20.2B, 2025 equity from the US government, SoftBank and NVIDIA $12.7B, divestitures $12.7B, subsidiary share sales $4.9B, government incentives $4.8B and net debt $9.5B (10-K cash-flow statements, summed). The dividend kept rising through FY2022 ($1.46 a share) after free cash flow turned negative. It was then halved in 2023 and suspended from Q4 2024, and the CHIPS agreement bars dividends for two years (FY2024 10-K). Buybacks stopped in Q1 2021. The pattern is a cash-return company that re-levered into a fab build-out, then turned to partners, asset sales and equity when the build-out outran its cash (inferred).
Two 2026 moves post-date the latest balance sheet or reshape it. In April Intel bought back Apollo's 49% of the Ireland Fab 34 venture for $14.2B, paid with cash and a $6.5B bridge refinanced with $6.5B of notes (8-Ks 2026-04-08 and 2026-04-30). That reclaims the fab's future earnings at a large up-front cost. In August, after the Q2 balance sheet, Intel sold 242.1M shares at $95.00, about $23.0B gross (8-K 2026-08-12). Pro forma, cash and short-term investments would rise from $29.7B at 27 June to roughly $52.7B and exceed the $50.5B of debt, before fees and second-half flows (inferred).
PCs are about 60% of Intel Products revenue and servers about 40%, and in 1H 2026 servers grew far faster (table below). Intel itself calls the PC industry "highly cyclical" (FY2025 10-K). Industry PC shipments went from 339.8M units in 2021 to 241.8M in 2023 and back to 270.2M in 2025 (Gartner). Intel's revenue followed, from $79.0B in FY2021 to $54.2B in FY2023.
| End market | 1H 2026 revenue | Share of Intel Products | Change vs 1H 2025 |
|---|---|---|---|
| PCs (CCPG) | $16.6B | 59% | +7% |
| Servers and custom silicon (DCAI) | $11.3B | 41% | +40% |
| External foundry (mostly Altera) | $0.47B | — | from $0.05B |
Source: Q2 2026 10-Q, Note 2 and MD&A. Shares inferred from segment revenue.
Capacity cannot flex to match the cycle, and today the mismatch runs both ways. Node-specific shortages on Intel 7 and Intel 3 coexist with surplus shells: Ohio has been slowed, Germany and Poland cancelled, and "our manufacturing asset portfolio exceeded manufacturing capacity requirements" (FY2025 10-K, Note 6).
Against its own history, Intel is early in a recovery but not near a peak. 1H 2026 revenue annualises to about $59B (inferred), against $79.0B at the FY2021 peak and $52.9B in FY2025. Gross margin of 39.9% sits between the 32.7% trough and the 62.3% peak. Server prices are at a cyclical high because of shortage, while PC units are entering a memory-driven decline. Server is in an upswing, client is turning down, and margin is mid-recovery.
The downside case runs through price. When substrate and memory shortages ease, AMD and Arm designers can buy more capacity at TSMC, and FY2025 showed what competitive pricing does: server ASPs fell 4%. Because Intel Foundry's costs are fixed, a price reset flows almost fully into margin. A second, non-cyclical exposure sits in Israel: "a significant portion of our current and anticipated future revenues" comes from Intel 7 products made there, uninsured against war (Q2 2026 10-Q).
| Indicator | Why it matters | Where published |
|---|---|---|
| Intel Foundry external revenue and operating loss | Tests whether the factory can stand on its own | Intel 10-Q segment note; quarterly release |
| Named 14A or 18A-P external customer | Decides whether 14A capex earns a return | Intel 8-K / press release |
| Server and client ASP and unit % change | Separates price from volume as supply normalises | Intel 10-Q, MD&A |
| x86 and Arm unit and revenue share | Share loss moves volume out of Intel fabs | Mercury Research, via trade press, quarterly |
| PC shipments | Client volume driver | IDC and Gartner, quarterly |
| TSMC node mix, capex, CoWoS capacity | Competitor scale and the packaging opportunity | TSMC quarterly management report |
Before a thesis, an investor would need three things resolved. First, what server pricing looks like once supply normalises. Second, whether 14A secures a committed external customer by mid-2027. Third, what Intel Products' margin would be if Foundry transfer prices were set by an outside party.