Get occasional email updates on new AI tools. No spam.
Every AI model runs on chips, and a small group of companies designs, makes and supplies them. This page follows that chain in four layers: who designs the processors, who makes the memory, who manufactures the chips, and who builds the machines that make them.
Every company links its sources, figures are shown as each company reports them, and there are no rankings or investment advice.
Inside an AI chip package
Tap a layer to filter the company list below.
Design → equipment builds the factory → foundry manufactures the chip → memory and advanced packaging are joined to it → systems go into data centers. A delay at any step slows how many AI chips can ship.
Designs CPUs and Instinct AI GPUs, and sells them as full racks (Helios). In July 2026 it announced a deal for Anthropic to deploy up to 2 GW of Instinct MI450 GPUs, and reported Q2 2026 data-center revenue of $6.7B (AMD, Aug 2026).
Makes the machines that lay down and shape thin material layers inside chips, including tools for HBM and advanced packaging. Quarterly revenue was $9.12B for the quarter ended July 2026 (Applied, Aug 2026).
The UK company whose chip designs run in most phones and many servers. In March 2026 it began selling its own data-center processor, the Arm AGI CPU, for the first time; SoftBank is its majority owner (Arm, Mar 2026).
Dutch maker of the EUV lithography machines used to print advanced chips. It reported Q3 2026 net sales of EUR 11.7B on 14 Oct 2026, guided 2026 sales to EUR 44-45B, and says planned 2027-28 supply will not fully meet demand.
Designs custom AI accelerators for large customers and the Ethernet networking chips that link AI clusters. Its AI semiconductor revenue was $16.7B in the quarter ended Aug 2026, and in April 2026 it announced a long-term deal to keep supplying Google’s TPUs (Broadcom filing, Sep 2026).
Beijing-based designer of data-center AI accelerators. It reported first-half 2026 revenue of 5.996B yuan, up 108% (filing relayed by Implicator, Aug 2026); it has been on the US Entity List since December 2022.
Builds wafer-sized AI chips and sells fast inference through its own cloud. It listed on Nasdaq (CBRS) in May 2026 and reported Q2 2026 GAAP revenue of $180M (Cerebras, Aug 2026); speed claims are the company’s own.
Builds inference hardware specialized for transformer models. In August 2026 it shipped its first rack to Jane Street and announced $700M at a $21B valuation (Etched, TechCrunch); it has not published independent benchmarks.
Built LPU inference chips; in December 2025 NVIDIA licensed its technology and hired its founder, and Groq says it stays independent as an inference cloud. It raised $350M at a $3.5B valuation in August 2026 (Groq, TechCrunch).
Designs the Ascend AI chips and sells them in Atlas servers and large clusters. In September 2026 it said its Ascend 950 cluster opens to customers in China from 30 Sept and outside China from 30 Nov; volume figures are reported, not confirmed.
US chipmaker with a foundry arm. Q2 2026 revenue was $16.1B, up 25%; Intel said 18A-P entered risk production and Xeon 6+, its first server chip on 18A, has launched (Intel, Jul 2026).
Makes the etch and deposition machines used to build leading-edge logic and memory chips. Revenue was $6.72B for the quarter ended June 2026 (Lam, Jul 2026), with China about 26% of sales.
Designs custom AI chips and the connectivity chips that move data inside data centers. It reported record quarterly revenue of $2.74B in August 2026 and signed a custom-chip agreement with Google in July 2026 (Marvell filings).
Makes memory chips, including the HBM that sits next to AI processors. It reported $54.2B revenue for the quarter ended September 2026 (Micron, Sep 30); HBM4 mass production for NVIDIA’s Vera Rubin platform was announced in March 2026.
Designs the data-center GPUs and systems AI runs on, and also builds AI models. In August 2026 it said Vera Rubin is in full production; Data Center revenue for the quarter ended 26 July 2026 was $89.0B (NVIDIA).
Best known for phone and PC processors with on-device AI, and now entering AI data centers with AI200 and AI250 inference systems. At its June 2026 investor day it set a target of over $15B in data-center revenue by fiscal 2029.
Japanese foundry building 2nm chips in Hokkaido with government backing. Mass production is planned for 2027 and, as of October 2026, it has not named a customer order (Rapidus; Reuters via TNW).
Makes its own AI inference chips (RDUs) and a cloud service. It announced a first close of $1B at an $11B valuation in July 2026; Intel is an investor and partner, not an owner (TechCrunch, Jul 2026).
Korean memory and foundry company. Its July 2026 results said HBM4 sales scaled up and the first HBM4E samples went to customers; on 8 Oct 2026 it guided Q3 operating profit of about KRW 107.4 trillion (preliminary, Samsung).
Korean memory maker whose HBM is used with AI accelerators. It began mass shipments of HBM4 in Q2 2026 and reported 2Q26 revenue of 79.3 trillion won (SK hynix, Jul 2026; preliminary figures).
China’s largest contract chipmaker. It reported Q2 2026 revenue of about $3.0B, up 36% (Aug 2026), and said AI demand should keep driving orders; it has been on the US Entity List since December 2020.
Chip company led by Jim Keller that sells Blackhole-based AI servers and licenses RISC-V processor designs. Its Galaxy Blackhole server launched in April 2026 from $110,000; performance claims are the company’s own.
Japanese maker of chip-fabrication equipment, including coater/developers, etch and deposition tools. April–June 2026 sales were 732.4B yen, up 33%, with China about 30% (Tokyo Electron, Jul 2026).
A contract chipmaker (foundry) that manufactures chips designed by others, including many for NVIDIA, Apple and AMD. It reported Q3 2026 revenue of US$46.7B on 15 Oct 2026, with 2nm at 8% of wafer revenue, and said in July it will add US$100B in Arizona.
What each company most recently reported, in its own currency and fiscal calendar. Periods differ and nothing is converted, so this table is for reference, not for ranking.
| Company | Period | As reported | Source |
|---|---|---|---|
| AMD | Q2 2026 (to June) | $11.5B revenue; Data Center $6.7B | AMD Q2 2026 results ↗ |
| Applied Materials | Fiscal Q3 2026 (to July) | $9.12B revenue | Applied Q3 FY2026 release ↗ |
| Arm | Fiscal Q1 2027 (to June) | $1.289B revenue | Arm filing (StockTitan) ↗ |
| ASML | Q3 2026 (to Sept) | EUR 11.7B net sales | ASML 6-K (SEC) ↗ |
| Broadcom | Fiscal Q3 2026 (to Aug 2) | $29.6B revenue; AI semiconductors $16.7B | Broadcom 8-K (SEC) ↗ |
| Cambricon | First half 2026 | 5.996B yuan revenue | Implicator ↗ |
| Cerebras | Q2 2026 (to June) | $180.1M revenue (GAAP) | Cerebras Q2 release ↗ |
| Intel | Q2 2026 (to June) | $16.1B revenue; Intel Foundry $5.8B | Intel Q2 release ↗ |
| Lam Research | Fiscal Q4 2026 (to June 28) | $6.72B revenue | Lam release (syndicated) ↗ |
| Marvell | Fiscal Q2 2027 | $2.739B revenue | Marvell results (TelecomTV) ↗ |
| Micron | Fiscal Q4 2026 (to Sept 3) | $54.23B revenue | Micron 8-K (SEC) ↗ |
| NVIDIA | Fiscal Q2 2027 (to July 26) | $96.2B revenue; Data Center $89.0B | NVIDIA 8-K (SEC) ↗ |
| Qualcomm | Fiscal Q3 2026 (to June) | $9.947B revenue | Qualcomm earnings release ↗ |
| Samsung | Q3 2026 (to Sept), preliminary guidance | About KRW 195 trillion sales; operating profit about KRW 107.4 trillion | Samsung guidance ↗ |
| SK hynix | Q2 2026 (to June), preliminary | 79.32 trillion won revenue | SK hynix 2Q26 results ↗ |
| SMIC | Q2 2026 (to June) | About $3.0B revenue | The Standard (Reuters) ↗ |
| Tokyo Electron | Fiscal Q1 2027 (Apr–June) | 732.4B yen net sales | TEL presentation ↗ |
| TSMC | Q3 2026 (to Sept) | NT$1,494.24B (US$46.72B) net revenue | TSMC 6-K (SEC) ↗ |
Dated notes from public reports, checked 15 Oct 2026. Rules and supply numbers change often. “Reported” and analyst-estimate items are not confirmed by the companies or governments involved.
Pulled live from our news sources and filtered for chip stories.
How we make this: company facts come from company releases and filings where we could read them, and from named press where we could not; each card links what we used. Many figures for Chinese companies are reported or company-stated and not independently confirmed. Where a company has not named a customer or a figure, we leave it out. Scale tiers on the Heat Map are editorial estimates. This is information, not investment, legal or trade advice. Spot a mistake? Tell us through the About page and we will check it.