Anthropic's ARR Breaks $30 Billion — Tripled in One Quarter. How?

On April 6th, Anthropic dropped a bombshell: annualized recurring revenue (ARR) had exceeded $30 billion. Note: that’s annualized, not full-year revenue.

The number itself is staggering, but the growth rate is even more jaw-dropping. As recently as late 2025, that figure sat at just $9 billion. Anthropic tripled its annualized revenue in a single quarter.

I’m not a finance blogger, but $30 billion ARR in any context is exceptional.

Let me clarify what ARR means. Annual Recurring Revenue is the total contracted revenue from subscriptions, normalized to a one-year period. It’s a SaaS metric — it doesn’t equal actual revenue, but it tells you something important: customers are paying, and paying more over time.

So what does $30 billion ARR mean for Anthropic? Rough math: roughly $2.5 billion per month. Today Anthropic’s main revenue comes from Claude subscriptions — Claude Pro ($20/month), Claude Team ($25/month), and Claude Enterprise (per-seat pricing).

For perspective: hitting $2.5B/month from Pro users alone would require roughly 125 million subscribers. Enterprise pricing varies wildly, but you get the order of magnitude.

The real question: does Claude actually have that many paying subscribers?

I see three drivers. First, Claude Code’s explosive growth. Enterprise adoption of Claude Code as a programmer’s standard tool has been rapid — seat consumption is staggering. Several startup CTOs I know say the ROI of equipping their teams with Claude Code is straightforward.

Second, rising API call volumes. Anthropic’s API pricing sits at the premium end, but call volumes keep climbing. Especially after Claude 3.5 Sonnet and 3.7 dropped, plenty of companies shifted workloads originally running on GPT-4.

Third, the compute partnership. Same day as the ARR announcement, Anthropic revealed a three-way deal with Broadcom and Google for roughly 3.5 gigawatts of TPU compute. Not 3.5 gigahertz — 3.5 gigawatts. That’s a substantial number. The partnership comes online starting 2027.

Here’s what’s interesting: Anthropic is OpenAI’s fiercest competitor, yet it chose to partner with Google and Broadcom rather than build compute alone. There’s a pragmatic judgment here — custom silicon takes time and capital, but TPUs are available now. Almost like the Xiaomi approach: get the product working first, worry about self-development later.

The real question: can $30 billion ARR hold?

Honestly, I don’t doubt the number. But whether that growth rate is sustainable is another matter. SaaS has a “40% rule” — if your growth falls below 40%, valuations start compressing. At a $30 billion base, maintaining that growth rate requires enormous absolute additions.

One hidden risk: GPT-6 is expected later this year. Once it drops, the competitive landscape shifts. Whether Anthropic can keep the base model competitive determines the quality of this growth.

My take: $30 billion ARR is a milestone, but更像是一个起点. What’s next — more compute for bigger models, or focus on monetization and profit?

Do you see Anthropic’s growth as genuine capability, or a symptom of the AI bubble?