Tomasz Tunguz 分析 Anthropic 是否更像一家云厂商

We all consider Anthropic an AI company.

But what if it’s a cloud?

Anthropic is growing much faster than the other clouds : 15x last year. GCP’s growth is far behind at 82%. But at current growth rates, Anthropic is closer to GCP’s pace, doubling next year.¹

In revenue terms, Anthropic is about 30% the size of AWS & Azure & nearly half the size of GCP.

Anthropic is the fastest growing but the smallest of the four. At current growth rates, it would pass Azure in about three years & AWS in about three & a half.⁴ Only GCP, now growing slightly faster than Anthropic’s current pace, would stay ahead. But Anthropic has the most concentrated business : 2 customers accounted for nearly 25% of revenue in 2025.⁵

Each started with a wedge to earn the cloud moniker : Amazon with ecommerce ; Google with search & ads ; Microsoft with the operating system. Anthropic’s angle : many of the best AI models in the world & the leading coding harness, Claude Code. 87% of businesses on Ramp that buy a coding agent buy Claude Code, up 56 points in a year.⁶

Anthropic exerts its pricing power in this market, charging more per task than others : Claude Opus 5.5 costs $5.98 per task on Artificial Analysis’ Intelligence Index, against $3.26 for OpenAI’s GPT-6 Astra & $1.99 for Google’s Gemini 4 Argon.⁷

It has the narrowest product offering, but it has made a record-breaking start with AI models, artifacts & design.

Growth needs capacity, & Anthropic has planned for it. It has made the largest data center commitments of the four : $518b, nearly 8x its $67b Q3 run rate, while AWS, Microsoft & Google have committed about 1x to 2x theirs.³

Each of these businesses is building its own chips, scaling data centers with debt, developing &/or reselling models, & managing infrastructure to support AI.

If it rains like a cloud & computes like a cloud… That’s how we should evaluate Anthropic.

Using Wall Street’s estimates, a standalone GCP would be worth roughly $1t to $2.3t.⁸ At $1.5t to $2t, Anthropic would be valued equally to GCP, on two-thirds the revenue & far faster growth over the past year.⁹

1. Anthropic’s annualized revenue rose from $65b at the end of July to about $71.5b at the end of September, 10% in two months or about 4.9% a month. Compounded over twelve months, 1.049^12 ≈ 1.77, a run rate near $127b by September 2027. The September figure is an estimate with a range of $69b to $74b; across that range the next twelve months imply 1.4x to 2.2x. Quarterly revenue grew faster, 45% from Q2 to Q3, but most of that gain came early in the quarter, so the exit pace is the better guide. Sources : CNBC, August 17, 2026; Syz Group, October 15, 2026. ↩︎

2. Microsoft’s Intelligent Cloud segment, which includes Azure plus server products & enterprise services. Microsoft first disclosed Azure alone for fiscal 2026 : $101.9b, up 41%. ↩︎

3. For AWS, Microsoft & Google, leases signed but not yet started, primarily for data centers, as of June 30, 2026, company-wide rather than for the cloud segment alone : Amazon $137.2b (10-Q), Microsoft $329.1b starting fiscal 2027 through 2033 (10-K), Alphabet $85.2b starting 2026 through 2031 (10-Q). These exclude the hyperscalers’ own construction, which runs separately at $116b of capex for Microsoft in fiscal 2026. Anthropic’s figure is broader : all cloud, compute & infrastructure obligations over roughly a decade, including at least $100b to AWS & $30b to Microsoft Azure, as reported by Reuters from a prospectus that has not been publicly filed. ↩︎ ↩︎

4. Q3 2026 run rates compounded at current growth : Anthropic $67b at 77% a year (4.9% a month, footnote 1), Azure $157b at 32%, AWS $169b at 37%, GCP $99b at 82%. Anthropic passes Azure in 2.9 years & AWS in 3.6. Because GCP is growing faster than Anthropic’s current pace, Anthropic would not catch it unless GCP slows; at 50% growth, Anthropic would pass it in about 2.4 years. ↩︎

5. As reported by Reuters from the prospectus, which does not name the 2 customers. ↩︎

6. Ramp, Claude Code vendor data, October 2026 : share of businesses with a coding-agent vendor that use Claude Code. Codex is second at 41% & Cursor third at 35% (Ramp, coding agents). ↩︎

7. Cost per Intelligence Index task at the effort settings Artificial Analysis lists, max for Opus 5.5 & GPT-6 Astra & high for Gemini 4 Argon : Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon, Artificial Analysis, October 2026. ↩︎

8. GCP’s standalone value, estimated two ways from published figures. Morgan Stanley expects GCP to earn 48% of Alphabet’s operating profit by 2028 & sets a $400 target, valuing Alphabet on one earnings multiple (TheStreet, August 26, 2026). At $400 × 12.23b shares (10-Q), Alphabet is worth $4.9t; GCP’s 40% to 48% share of profit implies $2.0t to $2.3t. Vulcan Value Partners puts GCP at roughly 40% of its value for Alphabet (Q2 2026 letter, via Insider Monkey). Alternatively, consensus 2027 GCP operating profit of $54b at 20x to 30x implies $1.1t to $1.6t. Neither firm publishes a standalone GCP value; these are our estimates from their numbers. ↩︎

9. Reuters reported from the prospectus that Anthropic seeks a valuation above $2t (CNBC). ↩︎