WSJ: AI Compute Power Shortage Deepens — H100 Rental Rates Up 60% in a Year

The Wall Street Journal reported on October 10, 2026, that demand for artificial intelligence compute power is growing at an unprecedented rate. The publication described the situation as a "desperate hunt" for AI compute: companies are struggling to find the graphics processors needed to train models and serve them. Citing data from research firm SemiAnalysis, the WSJ wrote that the rental price of Nvidia's H100 graphics processor under one-year contracts rose 60% over the past year. Meanwhile, Alphabet plans up to $200 billion in capital expenditure this year — much of it directed at AI infrastructure.

Rental Prices Are Climbing at a Record Pace

According to SemiAnalysis's analysis, the company's H100 one-year rental price index is compiled from a monthly survey of more than 100 market participants and confirmed with deal data. The index shows the price per GPU-hour rose from $1.70 in October 2025 to $2.35 by March 2026 — a gain of nearly 40%. The increase was especially rapid: the index's lowest point came in October 2025 itself, and contract prices surged over the following six months.

On-demand GPU capacity, meanwhile, is fully sold out across all GPU types. That means almost no spare capacity remains on the market — companies needing new capacity must wait in line or pay high prices on the contract market. According to the WSJ, some companies are trying to expand capacity under existing contracts, but supply is lagging behind demand.

As the WSJ writes, companies are in a "desperate" search for compute — the battle for capacity has peaked. The Journal, relying specifically on the SemiAnalysis index, notes that hourly rental rates under one-year contracts rose 60% in a year. (The Wall Street Journal report)

Anthropic Turned to Rivals for Capacity

The WSJ reported that Anthropic co-founder Tom Brown visited xAI's office at the end of March to negotiate a deal to rent compute capacity. The visit itself shows how serious the situation is: Anthropic went straight to its rival's office and offered to rent capacity from it.

The cause was the success of Claude Code. A sharp surge in demand for the service created a capacity crunch at the company, and users began facing frequent outages. According to the WSJ, those outages forced Anthropic to seek additional compute capacity from external sources.

In May, Anthropic announced an agreement with SpaceX to rent more than 300 megawatts of compute capacity. As the WSJ later found, the deal's total value reaches $45 billion. After the deal was announced, Elon Musk wrote on X that he was impressed by members of Anthropic's senior team.

"I am impressed by members of Anthropic's senior team," Elon Musk wrote on X (according to the WSJ).

Meta Refused to Share Chips

Anthropic CEO Dario Amodei spoke by phone early this year with Meta's chief AI officer, Alexandr Wang, to try to secure additional compute capacity. But the WSJ, citing people familiar with the matter, wrote that Meta discussed whether to give chips to Anthropic and has decided not to for now.

The decision shows how strained the market has become: even large technology companies are carefully rationing their GPU stockpiles and are unwilling to share capacity with competitors. Meta, mindful that it also needs capacity for its own internal AI projects, preferred to keep the chips. According to WSJ sources, the matter was seriously debated inside Meta — opinions split on whether to share chips with Anthropic, but the final decision was not to provide them for now.

Corporations Are Spending Billions on Infrastructure

The compute capacity shortage is also reflected in the investment plans of tech giants. Alphabet expects up to $200 billion in capital expenditure this year — much of it directed at artificial intelligence infrastructure. The figure shows the scale of the industry-wide capacity race: even the largest players are spending billions now to meet future demand. As the WSJ writes, most of these investments are going specifically into AI infrastructure — data centers and GPU capacity.

As the WSJ notes, the capacity shortage is felt across the industry. A 60% rise in rental rates in a year, on-demand capacity fully sold out, and leading AI labs asking rivals for capacity — all of it paints one picture: the compute infrastructure needed for artificial intelligence cannot yet meet demand.