GPU Shortage Reoccurs: Cloud Providers Tighten Supply, AI Startups Face 32% Chip Rental Price Increase and Long Waits Until Year-End

According to monitoring by Beating, several AI startups have reported that cloud providers like Microsoft and Amazon are concentrating GPU computing power towards internal teams and major clients (such as OpenAI and Anthropic), leaving small to medium-sized clients facing price increases, long wait times, and stricter contract terms. Recently, Microsoft’s Azure sales management informed employees that the GPU wait time for cloud customers is expected to last until the end of 2026. A specific case involves the image generation startup Krea (which raised $83 million, with investors including Andreessen and Bain Capital Ventures) that rented hundreds of Blackwell chips for $2.80 per card per hour six months ago under a six-month contract. When it came time to renew, several cloud providers did not answer calls, and the final deal was struck at $3.70/hour, a 32% increase, with the contract extended to one year. CEO Victor Perez stated that some providers did not respond at all, while others only engaged in discussions for three-year contracts. Will Falcon, CEO of GPU cloud provider Lightning AI, revealed that the company has 40,000 GPUs online, but the total demand from about 40 clients in the queue reaches 400,000 GPUs, with rental prices increasing by over 25% in six months. Microsoft implements tiered management for GPU access: approximately 1,000 largest clients (Tier 1) are prioritized; smaller clients wishing to rent Blackwell chips must commit to at least 1,000 units for a minimum of one year, with contracts starting at tens of millions of dollars. Pay-as-you-go customers may have their access revoked by Microsoft if their GPUs are idle for several hours. Startups participating in the ‘Microsoft for Startups’ free quota program have also been informed that insufficient usage will result in the withdrawal of GPU access. Hemant Taneja, a partner at General Catalyst, has surveyed portfolio companies about computing power bottlenecks, and the firm is planning to create a shared computing pool or collective bargaining for invested companies. Some startups are beginning to consider purchasing GPUs directly: AI startup Collide in the oil industry plans to spend about $500,000 to directly purchase NVIDIA GPUs and rent data center space to operate independently, in order to avoid queues and uncertainties.