Anthropic’s growth is outrunning its compute

Anthropic’s demand surge forced it to rent xAI’s Memphis data center and raise Claude limits.

Anthropic says its revenue and usage jumped 80-fold in one quarter on an annualized basis, a pace that turned a fast-growing AI company into an infrastructure headache. The company now says annualized revenue has reached $30 billion, while it also needs more GPUs, more megawatts, and more room for Claude Code users who keep hitting limits.

That creates a strange picture: a frontier AI company growing so quickly that it had to rent the full compute capacity of xAI’s Memphis data center, even as Elon Musk has spent years attacking Anthropic’s rival status. It also shows how far AI demand has moved beyond model quality alone. Capacity is now part of the product.

Anthropic’s growth is now a compute problem

At Anthropic’s developer conference in San Francisco, CEO Dario Amodei said the company planned for 10x growth, then got something much larger. He described the first-quarter surge as “just crazy” and “too hard to handle,” according to CNBC’s reporting from the event.

The numbers explain why the company is under pressure. Annualized revenue has climbed to $30 billion, up from a much smaller base a year ago. That growth has been driven by enterprise use, especially companies like Uber and Netflix, which are using Claude Code more heavily.

Anthropic says some customers prefer Claude Code over tools like Cursor because of its agentic behavior. That matters because agentic tools burn through more inference capacity than a simple chat interface. The better the model gets at taking actions, the more expensive it becomes to keep it online at scale.

• Enterprise demand is not just about chat prompts anymore.
• Agentic coding tools consume far more compute than light usage.
• Anthropic’s own limits became part of the customer experience.
• Infrastructure planning now decides how much product users actually get.

Claude Code users felt the strain first

The company’s rapid expansion came with visible friction. In late April, Anthropic published a postmortem saying three bugs had affected Claude Code since March 4. Its internal tests missed them, and users spent weeks dealing with degraded performance.

That gap between internal confidence and customer experience is ugly, especially for a company selling reliability to developers. Some users said on X that they were canceling subscriptions after weeks of stricter limits and weaker performance. One post summed up the frustration bluntly: “Whatever happened in the last 1-2 months is a significant regression.”

“That is the reason we have had difficulties with compute,” Amodei said at Anthropic’s developer conference, according to CNBC.

Amodei also said software engineers adopt new technology first, and that their behavior points to how AI will spread through the rest of the economy. He is probably right, but the more immediate lesson is less philosophical: if engineers hit limits first, everyone else will too once AI tools move deeper into daily work.

The Musk deal is weird, but the math makes sense

Anthropic’s new arrangement with xAI is one of those deals that only makes sense when the supply constraint is severe. Anthropic announced it would rent the entire compute capacity of Colossus 1, the Memphis data center built by Musk’s company, to raise usage limits for Claude Code and the Claude API.

The immediate payoff is concrete. Anthropic says it will double Claude Code’s five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans. It is also removing peak-hour limit reductions for Pro and Max accounts. For paying users, that means fewer interruptions and less waiting when demand spikes.

• 220,000 NVIDIA GPUs from the Memphis site.
• 300 megawatts of new compute available within a month.
• 5 gigawatts promised through the Google Cloud and Broadcom deal.
• Up to 5 gigawatts more through Amazon Web Services.

Anthropic’s broader compute strategy is clearly multi-supplier. The xAI deal buys short-term breathing room, while the Google, Broadcom, and AWS agreements are about future scale. The company is spending heavily to avoid being boxed in by its own success.

The irony is hard to miss. Musk has previously called Anthropic “misanthropic and evil,” yet his company’s data center is now helping power Anthropic’s growth. That is less a sign of friendship than a sign of how expensive AI infrastructure has become. When a rival can supply the only capacity you need fast enough, ideology loses to arithmetic.

What this says about the next phase of AI growth

Anthropic’s story is a useful snapshot of where the AI market is headed. Model quality still matters, but access, latency, and usage caps matter just as much once a company reaches real enterprise scale. Customers want tools that work under load, not demos that collapse at peak time.

The bigger question is whether this pace is sustainable without constant infrastructure deals. Anthropic is already talking about wanting “more normal” expansion, which is a polite way of saying the company would like growth to stop outrunning supply. That may be wishful thinking. If enterprise adoption keeps climbing, the next bottleneck will probably be energy, then chips, then the contracts that lock both in place.

For now, Anthropic has bought itself room to breathe. The real test is whether those extra GPUs and megawatts translate into better reliability before users decide the limits are the product. If it can make Claude Code feel dependable at scale, this quarter will look like a temporary strain. If not, the company may have to keep renting its way out of trouble.