Demand for Claude jumps 80-fold: why Anthropic is hunting for supercomputers

Demand for Anthropic's services grew 80-fold in the first quarter of 2025. That forced the company to urgently seek additional compute, including by considering offers from major providers of high-performance systems.

How demand for Claude is growing

The past few months have shown that the AI services market is far from saturated. Claude is quickly winning trust in the enterprise sector, from finance and consulting to software development and legal analysis. At the start of the year, Anthropic could serve orders with the compute it already had, but by the end of the quarter GPU consumption had grown 80-fold. This is not a forecast but a fact, acknowledged by the company's chief executive himself.

Several factors are driving the growth. Claude performs well on specialized tasks, and the company has landed large corporate clients, including in the public sector. The cost of serving requests through the Anthropic API has proved competitive with other solutions. Together, these created a wave of demand that nobody expected on this scale.

Anthropic's strategy fell behind reality

Unlike OpenAI, which aggressively bought up GPUs and signed capacity contracts early on, Anthropic took a conservative path. It invested gradually, based on current demand. From a financial standpoint that was sensible, but amid an unexpected boom it created a critical bottleneck.

When demand rose 80-fold, existing capacity turned out to be insufficient. The company cannot quickly bring new servers online: that takes time to buy chips, deploy infrastructure and reach agreements with providers. The GPU shortage is acute, as modern accelerators (NVIDIA H100, H200) are scarce across the entire global market.

The search for solutions

Anthropic is actively negotiating with providers of high-performance systems. Several strategies are on the table:

• Long-term contracts with cloud providers (AWS, Google Cloud): fast, but expensive
• Buying supercomputer systems for its own data centers: requires capital investment, but is cheaper in the long run
• Renting surplus capacity from companies that have already invested in GPUs
• Forming partnerships to deploy infrastructure jointly

Every path involves trade-offs. The cloud scales quickly but ties the company to a single provider. Owned infrastructure takes time but brings independence. Renting depends on what offers are available on the market.

What it means

Anthropic's story shows real demand that has swept across the AI market. The company bet on quality and caution in its investments. That won customers' trust, but now demands urgent scaling. Compute has turned out to be a real strategic asset, not an abstraction.

An 80-fold rise in demand points not to a bubble but to a healthy market that is ready to keep growing.