AI demand apex · multi-silicon compute
Anthropic
The AI lab behind Claude and the other apex of generative-AI demand. Google is investing up to $40B (cash + compute, 5GW, 1M TPUs) and Amazon up to $25B (5GW of Trainium), and it spreads its training stack across NVIDIA GPUs, Google TPUs and Amazon Trainium. Run-rate revenue ~$30B. Private (~$350B valuation).
Updated Jun 21, 2026
Money in ▸엔터프라이즈·API 고객◐Revenue ~$30B소비자 구독 (Claude)◐
Anthropic~$30B run-rate revenue
Financials & Segments
- Total revenue · ~$30B run-rate revenue
- Customer concentration · Large compute commitments to Google, Amazon, NVIDIA and Broadcom.
- Segments · Claude API · Enterprise · Subscriptions
Revenue from
- 엔터프라이즈·API 고객 Medium confidence· Revenue ~$30B · 2026 · run-rate revenue
Claude API/enterprise revenue (~$30B run-rate). - 소비자 구독 (Claude) Medium confidence
Claude consumer/pro subscription revenue.
Pays to
- Amazon High confidence
Uses 5GW of AWS Trainium; Amazon invests up to $25B ($100B AWS commitment).Confirms EN Anthropic Published Apr 19, 2026: "Amazon invests up to $25B; Anthropic commits $100B to AWS over 10 years." - Alphabet (Google) High confidence
Uses 5GW of Google Cloud TPUs (1M); Google invests up to $40B.Confirms EN TechCrunch Published Apr 24, 2026: "Google invests up to $40B in Anthropic, securing 1M TPUs / 5GW."Confirms EN Anthropic Published Apr 21, 2026: "Google invests up to $40B in Anthropic, securing 1M TPUs / 5GW."Confirms KO Global Economic Published Feb 19, 2026: "Anthropic relies on Google/Amazon/MS cloud for Claude; Google invests heavily."Confirms KO Wikipedia Published Jan 1, 2026: "Anthropic relies on Google/Amazon/MS cloud for Claude; Google invests heavily." - Broadcom Medium confidence
Expands custom compute via Broadcom, the Google-TPU design partner.Confirms EN Anthropic Published Apr 21, 2026: "Anthropic-Google-Broadcom partnership for multi-GW next-gen compute." - NVIDIA Medium confidence
Uses NVIDIA GPUs as a secondary training/inference substrate. - CoreWeave Medium confidence
Rents GPU-cloud compute from CoreWeave.
Value-chain ripple (tiers 2–3)
Following node-to-node links up to 3 tiers; each firm is shown once, at its nearest tier.
| Company | Chain | '21 | '22 | '23 | '24 | '25 | YTD |
|---|---|---|---|---|---|---|---|
| Anthropic | node | · | · | · | · | · | · |
| CoreWeave | benef1 | · | · | · | · | · | +64.7 |
| Alphabet (Google) | benef1 | +65.3 | -39.1 | +58.3 | +35.8 | +66.1 | +17.8 |
| Amazon | benef1 | +2.4 | -49.6 | +80.9 | +44.4 | +5.2 | +5.9 |
| SanDisk | benef2 | · | · | · | · | · | +820.4 |
| Marvell | benef2 | +84.9 | -57.5 | +63.7 | +83.8 | -22.8 | +265.9 |
| Intel | benef2 | +6.1 | -46.6 | +94.6 | -59.6 | +84 | +263.1 |
| Credo Technology | benef2 | · | · | +46.3 | +245.2 | +114.1 | +88.9 |
| Alchip Technologies | benef2 | +14.3 | -8.7 | +365.1 | -18.9 | +0.8 | +40.2 |
| Arista Networks | benef2 | +97.9 | -15.6 | +94.1 | +87.7 | +18.5 | +29.5 |
| Foxconn (Hon Hai) | benef2 | -5.1 | +2.8 | +7.8 | +80.1 | +26.9 | +21.8 |
| NVIDIA | benef2 | +125.5 | -50.3 | +239 | +171.3 | +38.9 | +13.1 |
| Samsung Electro-Mechanics | benef3 | -11.8 | -18.5 | -1.4 | -2.5 | +110 | +713.6 |
| AT&S | benef3 | +57.5 | -22 | -27.6 | -41.3 | +188.9 | +493.2 |
| Kioxia | benef3 | · | · | · | · | +1082.1 | +408.4 |
| LG Innotek | benef3 | +78.2 | -22 | -28.8 | -22.1 | +66.8 | +370.1 |
| Arm Holdings | benef3 | · | · | · | +64.2 | -11.4 | +302 |
| Micron Technology | benef3 | +24.1 | -46 | +72.1 | -1 | +240.5 | +297.6 |
| SK hynix | benef3 | -0.4 | -25.6 | +53.9 | +48.6 | +361.1 | +204.7 |
| Ibiden | benef3 | +30.4 | -19.6 | +51.6 | -38.6 | +261 | +199 |
| PSK Inc | benef3 | +22.7 | -19.6 | +14.2 | -2.6 | +228.2 | +168.3 |
| SUMCO Corporation | benef3 | -4.3 | -3.9 | +21.3 | -48.1 | +41.7 | +163.6 |
| Unimicron | benef3 | +133.5 | -30.2 | +35.2 | -19.7 | +175.7 | +155.7 |
| ASE Technology | benef3 | +38.3 | -12.8 | +60.9 | +10.1 | +65.6 | +151.9 |
| Astera Labs | benef3 | · | · | · | · | +25.6 | +150.7 |
| Applied Materials | benef3 | +83.6 | -37.5 | +68 | +1.1 | +59.6 | +140.7 |
| Lam Research | benef3 | +53.7 | -40.8 | +88.8 | -6.9 | +139.5 | +128.1 |
| Teradyne | benef3 | +36.8 | -46.3 | +24.8 | +16.5 | +54.4 | +126.4 |
| Nan Ya PCB | benef3 | +110.6 | -46.7 | -0.5 | -38.3 | +188.2 | +125 |
| GlobalWafers | benef3 | +27.3 | -30.3 | +14.4 | -38.9 | +51 | +123.3 |
| Samsung Electronics | benef3 | -9 | -14.2 | +20.9 | -26.4 | +212.9 | +121 |
| KLA | benef3 | +68 | -11.2 | +56 | +9.4 | +94.5 | +114.2 |
| Entegris | benef3 | +44.6 | -52.5 | +83.5 | -17.1 | -14.6 | +112.5 |
| Coherent | benef3 | -10 | -48.6 | +24 | +117.6 | +94.8 | +111.1 |
| Simmtech | benef3 | +119.6 | -34.6 | +12.4 | -63.3 | +392.9 | +105.5 |
| BESI | benef3 | +33.7 | -6.1 | +121.8 | -9.3 | +35 | +89.5 |
| Tokyo Electron (TEL) | benef3 | +40.4 | -14.5 | +89.8 | -4.4 | +61.6 | +84.1 |
| ASML | benef3 | +64.1 | -30.5 | +39.9 | -7.7 | +55.8 | +81 |
| Monolithic Power | benef3 | +35.5 | -27.9 | +79.9 | -5.6 | +54.5 | +73.2 |
| SCREEN Holdings | benef3 | +40.7 | -13.2 | +220.1 | -24.1 | +84.9 | +71.9 |
| TSMC | benef3 | +12.1 | -36.8 | +42.5 | +92.9 | +56 | +53.3 |
| Tokyo Ohka Kogyo (TOK) | benef3 | -2.7 | -3 | +63.8 | +4.3 | +111.5 | +52.9 |
| Lasertec | benef3 | +78.9 | +2.4 | +57 | -60.6 | +138.9 | +51.4 |
| ASM International | benef3 | +44.2 | +2.9 | +68.7 | +10.3 | +25.9 | +50.8 |
| Shin-Etsu Chemical | benef3 | +6.3 | +3.1 | +57.8 | -15.6 | +7.6 | +43.8 |
| Advantest | benef3 | +17 | -2.3 | +155.7 | +49.5 | +196 | +24.6 |
| Cadence Design Systems | benef3 | +36.6 | -13.8 | +69.6 | +10.3 | +4 | +23.9 |
| HPSP | benef3 | · | · | +216.5 | -30.4 | +54.7 | +21.7 |
| Broadcom | benef3 | +56.8 | -13.4 | +104.9 | +110.9 | +50.7 | +19.3 |
| King Yuan Electronics (KYEC) | benef3 | +26.7 | -7.7 | +132.4 | +38.6 | +171 | +4.2 |
| Synopsys | benef3 | +42.1 | -13.4 | +61.3 | -5.7 | -3.2 | -3 |
Anthropic, with OpenAI, is one of the two apexes of AI demand. What sets it apart is that it does not rely on NVIDIA GPUs alone but spreads across Google TPUs and Amazon Trainium. Google’s $40B and Amazon’s $25B investments and ~10GW of committed compute flow down into cloud, custom silicon and memory. Private, so no stock validation, but as a top-of-chain demand source it is, with OpenAI, where capex begins.