The squeeze hands pricing power to anyone who already controls chips, data centres and electricity, favouring hyperscalers, neoclouds and chipmakers over the startups trying to rent from them. Rising rental prices and tougher contract terms point to stronger near-term revenue for compute providers, but also to growing concentration risk, as only the biggest and best-funded labs can lock in capacity. The rush to overbook capacity as a hedge raises the danger of a sharp reversal if demand disappoints, leaving firms stuck with costly lease commitments and a glut of resold capacity. That fragility echoes recent warnings that financing, power and data centre delays, not demand, are now the binding constraints on the AI build-out.---Earlier:Broadcom CEO likens AI boom to a casino even as he targets $230 billion in chip salesGoldman Sachs flags power and compute as two chokepoints in the AI boom---AI’s real shortage is not ideas or money but raw computing power, and the scramble for it is turning rivals into partners, executives into deal brokers and startups into casualties.Summary:A shortage of computing power is driving a frantic, industry-wide scramble across Silicon Valley, with power struggles inside companies and unlikely alliances between rivals, the Wall Street Journal (gated) reportedThe cost of renting Nvidia H100 chips on a one-year contract has risen 60% in the past year, according to research firm SemiAnalysisCloud providers now demand multiyear contracts with upfront payments often as high as 30%, sometimes backed by guarantors, shutting out many smaller playersTop executives are personally brokering deals, including Anthropic’s agreement to lease more than 300 megawatts from SpaceX, reported to be worth up to $45 billionAnnounced capacity often fails to materialise, with projects delayed, cancelled or scaled back, and some deals collapsing when data centres slipVenture firms are now reserving GPU clusters for their portfolio companies, as access to compute rivals capital as the key constraint for startupsThe race to secure computing power for artificial intelligence has turned into a desperate, industry-wide scramble that is upending Silicon Valley, forcing bitter rivals into alliances, pulling chief executives into deal negotiations and leaving a generation of startups struggling to compete, the Wall Street Journal reported.Compute, the industry’s shorthand for computing power, has become the AI boom’s most coveted resource, and there is not enough of it. The squeeze is showing up in prices. The hourly cost of renting computing power from Nvidia’s H100 chips on a one-year contract has jumped 60% in the past year, according to research firm SemiAnalysis. Alphabet expects capital spending of up to $200 billion this year, much of it on AI infrastructure.The shortage is reshaping how deals get done. Anthropic co-founder Tom Brown visited Elon Musk at xAI in late March to negotiate renting computing power, according to people familiar with the matter, after demand for Anthropic’s Claude Code created a capacity crunch that contributed to frequent outages. In May, Anthropic announced an agreement with Musk’s SpaceX to lease more than 300 megawatts of computing power, part of a deal later reported to be worth up to $45 billion. Anthropic chief executive Dario Amodei also approached Meta about capacity, but Meta decided against supplying chips for now, the Journal reported.Even inside the largest companies, the fight is fierce. At Google, which paid $2.7 billion in 2024 to bring back AI researcher Noam Shazeer, his team’s computing allocation was later cut in favour of another project, and he subsequently left for OpenAI. Google co-founder Sergey Brin sometimes overrides the formal allocation process, according to people familiar with the matter.For smaller players, the terms are becoming punishing. Where startups could once rent chips flexibly, cloud providers now require multiyear contracts with upfront payments often as high as 30%, sometimes backed by guarantors. Some are turning in desperation to providers with weaker security or infrastructure. Much of the capacity that has been promised is not yet available, with projects delayed, cancelled or scaled back. One open-source startup lost a capacity deal when its partner’s data centre was delayed.The stakes are high either way. Firms that buy too much capacity risk crippling lease commitments, while those that buy too little risk stalling their growth. Some venture capital firms are now reserving GPU clusters themselves or guaranteeing contracts on behalf of their startups.With data centre construction slowed by regulation and political backlash, the shortage shows little sign of easing soon. The danger is that AI’s next phase is decided less by who has the best ideas than by who can secure the hardware to run them.No, its not! This article was written by Eamonn Sheridan at investinglive.com.