At the end of 2025, Anthropic was generating roughly $9 billion in annualized revenue.
Ninety days later, that number crossed 30 billion. CFO Krishna Rao announced the milestone on April 6, 2026, alongside an infrastructure deal that made the revenue figure look like the easy part.
The math is simple. The implications are not. A threefold revenue increase in a single quarter is the kind of growth curve that forces a company to rethink everything: its compute supply chain, its hiring pipeline, its infrastructure partnerships, and its entire relationship with the physical world of electricity and silicon.
Anthropic's answer came in a single announcement that rewrites the scale of what an AI startup can demand from the global chip industry.
For context: In late March, LDS reported on Anthropic's accelerating revenue trajectory and its plans for a potential IPO. The company has since blown past every projection in that article.
The Biggest Compute Deal an AI Startup Has Ever Signed
On April 6, Broadcom disclosed an expanded agreement to design and supply custom Tensor Processing Units for Google, paired with a networking and components deal extending through 2031. Anthropic is the anchor customer.
Starting in 2027, Anthropic will gain access to approximately 3.5 gigawatts of next-generation TPU capacity, built on Broadcom-designed chips and deployed through Google Cloud. That capacity builds on the 1 gigawatt already flowing to Anthropic in 2026.
Krishna Rao called it "our most significant compute commitment to date."
To put 3.5 gigawatts in perspective: the average U.S. nuclear power plant generates about 1 gigawatt. Anthropic is contracting for the equivalent output of three and a half nuclear reactors, dedicated entirely to running AI models. The majority of the new capacity will be built in the United States, extending Anthropic's November 2025 commitment to invest $50 billion in American AI computing infrastructure.
Revenue Growth That Has No Real Parallel
The infrastructure deal makes sense only in the context of Anthropic's revenue trajectory, which has few parallels in enterprise software history.
| Metric | End of 2025 | February 2026 | April 2026 |
|---|---|---|---|
| Annualized revenue | ~$9 billion | ~$16 billion | $30 billion+ |
| Enterprise customers ($1M+/year) | Not disclosed | 500+ | 1,000+ |
| Post-money valuation | ~$60 billion | $380 billion | $380 billion |
The acceleration began when Anthropic closed its Series G funding round on February 12, 2026. The round raised $30 billion and was led by GIC and Coatue, with co-investors including D.E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ, and MGX.
In the eight weeks since, enterprise spending on Claude has doubled. More than 1,000 businesses now spend over $1 million annually on Anthropic's products, up from roughly 500 in February. That doubling in paying enterprise customers is what drove the revenue from 16 billion to 30 billion in under two months.
Three Chip Vendors, One Insatiable Appetite
Anthropic's compute strategy is deliberately multi-vendor. The company runs workloads across three separate chip architectures simultaneously: Amazon Trainium processors, Google TPUs, and Nvidia GPUs.
The Amazon partnership remains Anthropic's largest. AWS invested $8 billion in the company, and the centerpiece of their relationship is Project Rainier, a cluster of approximately 500,000 Trainium 2 chips housed in Indiana.
The Google/Broadcom deal adds a second massive compute source. Broadcom CEO Hock Tan previously disclosed that Anthropic placed a $10 billion chip order in September 2025. A second order worth 11 billion dollars followed in December. The new deal extends that relationship through 2031.
Broadcom's SEC filing included a caveat that Wall Street noticed: "The consumption of such expanded AI compute capacity by Anthropic is dependent on Anthropic's continued commercial success."
This three-vendor approach is unusual in an industry where most companies bet heavily on a single hardware ecosystem. Anthropic's logic is pragmatic: no single chip supplier can scale fast enough to meet its demand, and diversification protects against supply chain disruption. When NVIDIA showcased its next-generation Vera Rubin architecture at GTC 2026, Anthropic was already a customer of all three competing chip pipelines.
Wall Street Is Pricing In a Power Problem
Analysts at Mizuho, led by Vijay Rakesh, estimated that Broadcom would record $21 billion in AI revenue from Anthropic alone in 2026. That figure is projected to roughly double the following year. Broadcom shares rose 3% on the announcement.
The scale of power consumption has become the central constraint for the entire AI industry. At 4.5 gigawatts total (1 GW current plus 3.5 GW incoming), Anthropic's compute footprint will consume more electricity than many small countries. The infrastructure required to deliver that power, including substations, cooling systems, and grid interconnections, takes years to build. That timeline is why the deal locks in capacity for 2027 rather than delivering it immediately.
The capital flowing into AI infrastructure is staggering. OpenAI raised $122 billion in the first quarter of 2026, with Amazon contributing 50 billion of that. Anthropic's own Series G brought in 30 billion. The total investment in AI compute infrastructure across just these two companies now exceeds 150 billion dollars in a single quarter.
The Risks Nobody in the Press Release Mentioned
Anthropic's growth numbers are extraordinary, but they come with caveats.
Run-rate revenue is not the same as confirmed annual revenue. It represents the current monthly revenue multiplied by 12, which means a single strong quarter can produce a dramatic headline number. Whether Anthropic sustains 30 billion dollars through the remainder of 2026 depends on enterprise retention rates that have not been publicly disclosed.
The compute commitments are front-loaded with risk. Broadcom's own filing acknowledged that the expanded deal is contingent on "Anthropic's continued commercial success." If revenue growth slows, the company could find itself locked into infrastructure contracts that exceed demand. That pattern has destroyed capital-intensive technology companies before.
Competitors are not standing still. OpenAI has surpassed 25 billion dollars in annualized revenue and is reportedly taking early steps toward a public listing. Google's Gemini 3.1 is competitive on key benchmarks. The question is not whether Anthropic can grow; it is whether it can grow fast enough to justify the infrastructure it is buying.
The Bottom Line
Anthropic went from 9 billion to 30 billion dollars in run-rate revenue in roughly three months, then committed to 3.5 gigawatts of new compute capacity to fuel the next phase. The company is betting that demand for Claude will not just continue but accelerate, and it is locking in the power and silicon to back that bet through 2031.
The deal with Google and Broadcom is the largest single compute commitment any AI startup has ever made. It is also the most exposed. If Anthropic's growth curve holds, the contract will look like a bargain. If it flattens, the company will be paying for nuclear-reactor-scale infrastructure it no longer needs.
Krishna Rao called it Anthropic's "most significant compute commitment to date." The word "commitment" is doing a lot of work in that sentence.
Sources
- Broadcom agrees to expanded chip deals with Google, Anthropic (CNBC, April 6, 2026)
- Anthropic ups compute deal with Google and Broadcom amid skyrocketing demand (TechCrunch, April 7, 2026)
- Anthropic reveals $30bn run rate, plan to use new Google TPU (The Register, April 7, 2026)
- Anthropic signs biggest compute deal yet with Google and Broadcom as run rate hits $30bn (The Next Web, April 7, 2026)
- Broadcom Inks Major AI Chip Deals With Google and Anthropic (Motley Fool, April 7, 2026)
- Anthropic, Google, Broadcom announce 3.5GW TPU deal (Silicon Republic, April 7, 2026)
- Anthropic Google Broadcom TPU deal 3.5 gigawatts and $30 billion revenue (Yahoo Finance, April 7, 2026)
- Anthropic's Revenue Nearly Doubled in Two Months. Now It's Targeting a $60 Billion IPO. (Let's Data Science, March 31, 2026)