Semiconductors: x86 CPUs, data-center accelerators, foundry. INTC / AMD. Run 23 September 2026. Built from each company's FY2023-FY2025 Form 10-K, Q2 FY2026 Form 10-Q and 8-Ks in the research folder, plus cited web sources for guidance, market share and deals. Events swept through 23 September 2026; most recent events checked: Intel $23.0B common stock offering (8-K, 12 Aug 2026) and AMD $4.75B senior notes (8-K, 17 Aug 2026). Figures in USD, as reported; no FX conversion. Not a valuation and not a recommendation.
AMD, clearly. Three cells carry its next five years. In AI accelerators, AMD holds multi-year agreements for up to 6GW with OpenAI (Oct 2025), up to 6GW with Meta (8-K, 24 Feb 2026) and up to 2GW with Anthropic (AMD release, 22 Jul 2026), with MI450 and Helios racks starting to ship in Q3 2026 and stepping up into 2027, so the 2027 ramp is contracted rather than hoped for. In server CPUs, AMD reached 46.2% of x86 server revenue in Q1 2026 (Mercury Research via Tom's Hardware, approx., unverified) and guides server CPU revenue up more than 70% in 2027 (Q2 FY26 call, approx., unverified), because each EPYC generation ships on a TSMC node Intel has only just matched. In client PCs, units rose 34% in Q2 2026 (10-Q). Intel's 2026 recovery is real but different in kind: Q2 client revenue rose on ASPs up 27% while volume fell 8%, and server ASPs rose 48% on volume up 9%, with demand exceeding supply (Q2 FY26 10-Q). Price-led growth in a shortage fades when supply catches up; share and new categories do not. Geographically the mix is similar (China 24% of Intel revenue, 22% of AMD in FY25), but Intel's China billings fell 18% in 2025 while AMD's rose 24% (10-Ks).
AMD at the company level, through a variable cost base: it pays TSMC only for wafers it buys, so higher Data Center mix flows to gross margin (46.1% in 2023, 49.5% in 2025, 53.3% in 1H26, GAAP; 56% non-GAAP guided for Q3 2026). Intel's consolidated margin is held down by manufacturing: Intel Foundry lost $10.3B in 2025 and $4.5B in 1H26 while its 18A ramp adds higher-cost wafers (10-K, 10-Q). The tension worth stating plainly: Intel's design business now out-earns AMD's segments. Intel Products ran a 32% operating margin in 1H26 (DCAI 40% in Q2) against AMD Data Center at 30% and Client and Gaming at 16% (10-Qs, segment basis, both excluding stock comp and acquisition amortization). Intel has a fortress product franchise attached to a loss-making factory; AMD has an average-margin but compounding engine. AMD's own caution: early MI450 revenue will sit slightly below the corporate gross margin (Q2 FY26 call, approx., unverified).
Intel's exposure is cost rigidity plus capital need. Depreciation was $10.8B in 2025, 2026 capex is guided above $20B, and 14A economics depend on landing outside customers (10-K; Q2 FY26 call). In 2024 a 2% revenue decline produced an $11.7B operating loss after $3.3B of fab impairments (FY24 and FY25 10-K). A significant portion of revenue also comes from Intel 7 products made in its Israel fab, which Intel self-insures against war (Q2 FY26 10-Q). AMD's exposure is concentration: Data Center is 58% of revenue, AI GPU volume is anchored by a handful of labs that were paid in warrants for up to 320M AMD shares (about 20% of 1,632M outstanding; 8-K 24 Feb 2026, 10-Q), and every leading-edge wafer comes from TSMC in Taiwan. In a general downturn Intel breaks first, because its costs do not flex; AMD breaks first only in an AI-capex pause or a Taiwan disruption.
These six product cells carry nearly all of both companies' revenue; neither discloses product by region, so each cell is global with the China overlay noted. Every score is argued in the tabs below.
| Cell (product x region) | Intel | AMD | Why (one clause, sourced) |
|---|---|---|---|
| Server CPU, global (China ~22-24% of each) | 2 | 4 | AMD at 46.2% x86 server revenue share Q1'26 and rising; Intel cut server ASPs 4% in 2025 for competitive reasons and holds on its installed base (Mercury via Tom's Hardware, approx., unverified; Intel FY25 10-K). |
| Client PC CPU, global | 3 | 4 | Intel keeps ~69% of client revenue and priced up 27% in Q2'26; AMD unit share rose to 30.3% on Ryzen mobile (Mercury via XenoSpectrum, approx., unverified; 10-Qs). |
| Data-center AI accelerators, US-led hyperscalers and labs | 1 | 3 | AMD is the #2 merchant supplier with 14GW of agreements behind Nvidia; Intel took $922M of Gaudi charges in 2024 and has no shipping successor (10-Ks; AMD releases). |
| Gaming and semi-custom, global | 1 | 3 | AMD is the console SoC incumbent but revenue fell 31% in Q2'26 on the console cycle; Intel's Arc is immaterial (AMD Q2 FY26 10-Q). |
| Embedded, edge and FPGA, global | 2 | 3 | AMD Embedded holds a 36% operating margin (FY25) and is recovering (+19% Q2'26); Intel sold 51% of Altera in Sept 2025 (10-Ks; 8-K 15 Sep 2025). |
| External foundry, US leading-edge capacity | 2 | 0 | Intel is the only US leading-edge logic manufacturer but external revenue was $307M in 2025; AMD is fabless by design (Intel FY25 10-K). |
Scores are anchored to the exhibits in the three tabs and are not summed; the three lenses get their verdicts in the Three Answers above.
Neither company discloses revenue by product and region together, so the cells below are product cells with geography as an overlay; the cross cells are ND. Segments were normalized as follows: server compute = Intel DCAI vs AMD Data Center (AMD's includes Instinct GPUs; the EPYC vs Instinct split is ND); client = Intel CCG (CCPG from 2026) vs AMD Client; embedded and edge = Intel All Other (Mobileye, Altera until 11 Sep 2025, IMS) vs AMD Embedded; foundry = Intel Foundry external revenue only.
The table shows the core contrast. Intel's two big cells were flat from 2023 to 2025: client grew -0.1% a year and server compute 2.9%. AMD's two big cells compounded at 51% and 60%. Intel is still larger in client by 3x, and the two are now the same size in server compute ($16.9B vs $16.6B), which is the clearest single measure of the last decade.
| Cell | Intel rev $m (% total) | Intel 2y CAGR | Intel latest | AMD rev $m (% total) | AMD 2y CAGR | AMD latest | Margin signal / leader |
|---|---|---|---|---|---|---|---|
| Client PC CPU | 32,228 (61%) | -0.1% | +7% (1H26) | 10,640 (31%) | +51.3% | +24% (1H26) | Intel on scale and margin (CCG 29% op margin); AMD on share |
| Server / data-center compute | 16,919 (32%) | +2.9% | +40% (1H26) | 16,635 (48%) | +60.0% | +81% (1H26) | AMD on growth; Intel DCAI 40% op margin Q2'26 vs AMD DC 31% |
| Gaming and semi-custom | ND | ND | ND | 3,910 (11%) | -20.7% | -15% (1H26) | AMD only; console-cycle driven |
| Embedded, edge, FPGA (Intel: All Other) | 3,563 (7%) | -19.2% | -33% (1H26, Altera out) | 3,454 (10%) | -19.4% | +12% (1H26) | AMD (36% op margin FY25) |
| External foundry | 307 (0.6%) | ND | +$414M YTD (Altera as customer) | 0 | n/a | n/a | Intel only; subscale |
USD m, FY2025 reported; growth = 2-year CAGR FY23-FY25 and 1H26 YoY. Intel percentages use consolidated revenue of $52,853m (segment revenue before eliminations). Sources: Intel FY25 10-K Note 3 and Q2 FY26 10-Q; AMD FY25 10-K segment note and Q2 FY26 10-Q.
Server CPU. AMD's driver is product cadence on TSMC's leading nodes and a platform its customers already qualified; Intel's is the Xeon installed base and, from 2026, 18A products. AMD's server revenue share climbed to 46.2% in Q1 2026 while its unit share was 34.5% in Q2 (Mercury Research via press, approx., unverified), so AMD sells the higher-priced parts. Intel's 2025 server ASP fell 4% on pricing actions taken in a competitive environment (FY25 10-K); its 2026 ASP gains come with demand above supply.
AI accelerators. AMD's growth here is mostly organic: ZT Systems (acquired Mar 2025, $4.4B) was kept for rack design and its manufacturing arm was sold to Sanmina for $2.4B (FY25 10-K). Intel has no product in volume: Gaudi produced $922M of inventory charges in 2024 and Crescent Island and Jaguar Shores are in development (FY25 10-K).
Client. Intel's 18A Panther Lake (Core Ultra Series 3) entered volume in early 2026 and is the first product on its own leading node since the delays of 2018-2023; AMD's gains come from Ryzen mobile. The client market itself is under pressure from memory prices (Mercury via Tom's Hardware reports desktop shipments down 20%, approx., unverified).
Foundry. Intel committed to finishing 14A in Q2 2026 (10-Q), with risk production targeted for 2H 2027 and volume in 2028 (Q2 FY26 call, approx., unverified). A preliminary Apple-Intel manufacturing agreement was reported by the Wall Street Journal on 8 May 2026 but Intel has not confirmed a named 14A customer in its filings (TheStreet citing WSJ, approx., unverified).
Intel sells mostly through PC and server OEMs: its three largest customers were 43% of 2025 revenue (19%, 12%, 12%) and it paid incentives in 1H24 that pulled volume forward (FY25 10-K). AMD had no 10% customer in 2024 or 2025, but in AI GPUs it bought access with equity: performance warrants for up to 160M shares each to OpenAI and Meta at a $0.01 strike (8-K 24 Feb 2026; FY25 10-K). Right now Intel holds price better in client (ASP +27% vs AMD -6% in Q2) because Intel is supply-short and AMD is taking units with lower-priced mobile parts.
Intel makes most of its products in its own fabs but uses TSMC for key tiles, depends on ASML as sole EUV supplier and reports industry shortages of substrates and memory that it expects to persist into 2027 (FY25 10-K; Q2 FY26 10-Q). AMD uses TSMC for all wafers at 7nm and below, GlobalFoundries for older nodes, and third parties in China, Malaysia and Taiwan for assembly and test, with $12.2B of purchase commitments at end-2025 and supply prepayments in 1H26 (FY25 10-K; Q2 FY26 10-Q). Intel controls more of its supply but carries its fixed cost; AMD is flexible but single-sourced at the leading edge.
Server CPU: Intel loses more share if 18A Xeons slip, AMD if TSMC allocates scarce capacity toward Nvidia or Apple. AI accelerators: AMD alone is exposed, to an AI capex pause and to export rules (a $800M MI308 charge in 2025, partly reversed; FY25 10-K). Client: both face Arm PCs and memory-driven unit weakness, Intel more because it has more to lose. China: 24% of Intel revenue, falling, vs 22% of AMD, rising (FY25 10-Ks).
x86 installed base: Xeon is 'a significant portion of global server infrastructure' (FY25 10-K). Durability Med: AMD took server revenue share to ~46% on the same instruction set, so the base is portable. US leading-edge manufacturing: the only company doing leading-edge logic R&D and volume production in the US (FY25 10-K). Durability High to replicate, but not yet economic. Policy backing: US government holds 275M shares plus escrow and CHIPS funds were accelerated ($5.7B, 8-K 29 Aug 2025); Nvidia invested $5B and co-develops x86 products. Durability Med.
Design cadence on TSMC's leading nodes with chiplets: proof is share gains in server and client every year (Mercury, approx., unverified). Durability Med-High. Full-stack AI (CPU, GPU, NIC, rack design from ZT, ROCm software): proof is 14GW of multi-year agreements. Durability Med, since ROCm still trails CUDA and access was partly paid in warrants. FPGA and adaptive SoCs from Xilinx (AMD targets >70% share, Analyst Day 11 Nov 2025). Durability High.
Intel: OEM-concentrated (top three customers 43% of 2025 revenue) and a price-taker in 2024-25, then a price-setter in the 2026 shortage. AMD: no 10% customer in 2024-25, broad hyperscaler base for EPYC, but AI GPU demand anchored by OpenAI, Meta and Anthropic, with warrants to the first two worth up to about 20% of shares outstanding. In a softening market, pressure lands on Intel in client and server CPUs (OEMs shift share of wallet) and on AMD in AI GPUs (a few buyers set the pace).
Intel: vertically integrated, but TSMC for tiles, ASML sole EUV source, substrate and memory shortages, helium shortage from the Iran conflict, and an uninsured Intel 7 fab in Israel. Dual-sourcing Med; pass-through Partial (2026 pricing partly offsets input costs; gross margin ~40%). AMD: TSMC single source at 7nm and below, no long-term commitments with some suppliers. Dual-sourcing Low; pass-through Strong in this cycle (gross margin up from 46% to 53% through 2023-1H26).
| Cell | Route control | Pocket price | Continuity | Outcome (share / margin) | Confirming KPI |
|---|---|---|---|---|---|
| Server CPU | Both direct to hyperscalers and OEMs | AMD above Intel (46% revenue share on 34.5% units, inferred) | Intel supply-constrained on Intel 7/Intel 3; AMD tight on TSMC and substrates | AMD share up, margin up; Intel share down, margin up short-term | Mercury server revenue share, quarterly |
| Client PC CPU | OEM channel for both | Intel above AMD in 2026 (ASP +27% vs -6%) | Both constrained 1H26; Intel expects easing 2H26 | Intel share down slowly, margin flat; AMD share up, segment margin down (16% 1H26) | AMD Client and Gaming operating margin |
| AI accelerators | Direct to labs and clouds (AMD); Intel absent | AMD below Nvidia (inferred) | AMD dependent on TSMC CoWoS and HBM | AMD share up from a low base, margin slightly below company average | AMD Data Center AI growth vs 'well over 100%' in 2027 |
The tension: in client, Intel is the price-setter while AMD takes units at lower ASPs, so client is the one cell where AMD's growth and pricing power point in different directions.
Three causes explain most of the difference. (1) Manufacturing model, Major: Intel carries $10.8B of annual depreciation and $3.3B of 2024 impairments; closing it needs external foundry volume on 18A and 14A, a 3-5 year task. (2) Node access and execution 2018-2024, Major: AMD shipped on TSMC's best nodes while Intel slipped; 18A in volume in 2026 narrows this in 2-3 years if 14A follows on time. (3) AI accelerator portfolio, Major: AMD has an annual MI roadmap and rack systems, Intel has none in volume; this is not money-buyable in under three years.
Both report by function under US GAAP, so no by-nature mapping was needed. COGS is total cost of sales (AMD's includes about $1B a year of acquired-intangible amortization; Intel's includes fab depreciation, impairments and under-utilization). SG&A is marketing, general and administrative. Restructuring (Intel) and operating amortization (AMD) sit outside the five ratios and bridge to EBIT. Same fiscal years (late-December year ends), GAAP, USD.
| % of sales, 3y avg | Intel | AMD | Gap (pts) | What drives it |
|---|---|---|---|---|
| COGS | 64.2 | 51.7 | +12.5 | Intel fab depreciation and 2024-25 impairments in cost of sales |
| R&D | 28.9 | 24.8 | +4.1 | Intel funds process technology as well as products |
| SG&A | 9.8 | 10.9 | -1.1 | Intel cut headcount ~15% in 2025; AMD go-to-market spend +52% in 2025 |
| Gross margin | 35.8 | 48.3 | -12.5 | Fixed-cost fabs vs pay-per-wafer at TSMC |
| EBIT margin | -8.7 | 6.6 | -15.3 | Intel restructuring $7.0B (2024); AMD acquired-intangible amortization ~$2.3B a year, mostly Xilinx |
| Year | COGS | R&D | SG&A | GM | EBIT |
|---|---|---|---|---|---|
| 2023 | 60.0 | 29.6 | 10.4 | 40.0 | 0.2 |
| 2024 | 67.3 | 31.2 | 10.4 | 32.7 | -22.0 |
| 2025 | 65.2 | 26.1 | 8.7 | 34.8 | -4.2 |
| 1H26 | 60.1 | 22.7 | 7.4 | 39.9 | -4.5 |
| Year | COGS | R&D | SG&A | GM | EBIT |
|---|---|---|---|---|---|
| 2023 | 53.9 | 25.9 | 10.2 | 46.1 | 1.8 |
| 2024 | 50.6 | 25.0 | 10.6 | 49.4 | 7.4 |
| 2025 | 50.5 | 23.4 | 12.0 | 49.5 | 10.7 |
| 1H26 | 46.7 | 22.6 | 12.2 | 53.3 | 15.9 |
Raw values, USD m (Intel / AMD). Revenue 2023: 54,228 / 22,680; 2024: 53,101 / 25,785; 2025: 52,853 / 34,639. EBIT 2023: 93 / 401; 2024: -11,678 / 1,900; 2025: -2,214 / 3,694. Intel 1H26 EBIT includes $3.9B goodwill impairment; excluding restructuring and other charges it was 9.8%. Capex as % of revenue: Intel 47.5 / 47.3 / 33.4 (gross, incl. financed additions) vs AMD 2.4 / 2.5 / 2.8. Sources: FY23-FY25 10-K income statements and cash flow statements; Q2 FY26 10-Qs.
Gross margin is the persistent structural gap: 12.5 points on the three-year average and 13.4 points in 1H26, present in every period. The mechanism is the manufacturing model from the power map. Intel's depreciation rose from $7.8B to $10.8B between 2023 and 2025 and sits in cost of sales, and 2024 added $3.3B of fab impairments; AMD pays TSMC a margin but only on wafers it buys. The cross-check confirms the power map rather than contradicting it: Intel's product segments earn more than AMD's (Intel Products 32% operating margin in 1H26), so the gap is not weak pricing, it is the factory. That makes the gap closable in principle, by filling fabs with outside customers, but not by cost-cutting: Intel's combined R&D and SG&A were already down 7% in 1H26 after two restructurings. The one line where Intel is leaner, SG&A, reflects those cuts and AMD's go-to-market build-out.
| Metric | Threshold | By when | If it hits, it favors | Where published |
|---|---|---|---|---|
| AMD Q3 2026 revenue | At or above $13.0B (guide $12.7-13.3B), non-GAAP GM ~56% | Early Nov 2026 | AMD | AMD Q3 FY26 release and 10-Q |
| Intel Q3 2026 revenue and GM | Within $15.8-16.8B; non-GAAP GM at or above 42% | Late Oct 2026 | Intel | Intel Q3 FY26 release and 10-Q |
| Intel 14A: PDK 0.9 and named external customer | PDK 0.9 on time (Oct 2026); definitive customer disclosed | By Q2 2027 | Intel | Intel 8-K / 10-Q, earnings call |
| Intel Foundry operating loss | Below $1.0B per quarter | By Q4 2027 | Intel | Intel 10-Q segment note |
| AMD x86 server revenue share | At or above 50% | By Q4 2027 | AMD | Mercury Research via press; AMD target (Analyst Day 11 Nov 2025) |