GPU shortage repeats: OpenAI, Anthropic and other major players consume supply, AI startups wait until the end of the year

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1300 words

Cloud giants like Microsoft and Amazon are concentrating NVIDIA GPU computing power toward internal teams and top-tier clients such as OpenAI and Anthropic, leaving small and medium-sized AI startups facing a triple squeeze: a 32% surge in rental costs, waitlists extending to the end of the year, and contract thresholds skyrocketing to tens of millions of dollars. Some venture capital firms have begun planning shared computing pools for collective bargaining, while some startups are opting to purchase GPUs themselves to bypass the wait. (Previous coverage: NVIDIA stock hits all-time high with market cap exceeding $5 trillion: $5 billion investment in Intel more than doubled in less than a year, with $12.7 billion in paper gains.) (Background: Jensen Huang’s GTC2026 keynote: AI demand reaches trillions of dollars, computing power jumps 350x, OpenClaw turns every company into AaaS.) According to reports, Will Falcon, CEO of GPU cloud provider Lightning AI, provided figures that precisely describe the 2026 AI computing supply-demand ratio as 1 to 10. His platform has about 40 clients on a waitlist, and rental prices have risen by more than 25% over the past six months. With 40,000 GPUs online, there is a demand for 400,000 in the queue. According to The Information, image generation startup Krea (which has raised $83 million in total funding, with investors including Andreessen Horowitz and Bain Capital Ventures) rented hundreds of NVIDIA Blackwell chips six months ago at $2.80 per card per hour on a 6-month contract. When it came time to renew, many cloud providers simply stopped taking their calls. They eventually closed at $3.70, a 32% increase, with the contract term extended to one year. Krea CEO Victor Perez stated that some vendors would only negotiate if they signed three-year long-term contracts, while others simply did not respond. This is not an isolated case. The Ornn Compute Price Index shows that the hourly rental for Blackwell has soared to $4.08, a 48% increase within two months. The one-year lease price for H100s also jumped from $1.70 in October 2025 to $2.35 in March 2026, an increase of nearly 40%. Microsoft has implemented tiered management for GPU access, prioritizing computing power for its approximately 1,000 largest clients (Tier 1). Small clients wanting to rent Blackwell must commit to at least 1,000 units for at least one year, with contracts starting at tens of millions of dollars. For pay-as-you-go clients, if GPUs are left idle for a few hours, Microsoft may directly revoke access. Startups participating in the "Microsoft for Startups" free credit program have also been told that their GPU permissions will be revoked if they do not utilize them sufficiently. Azure sales management recently informed employees that GPU wait times for cloud clients are expected to last until the end of 2026. Microsoft is now concentrating computing power on major clients like OpenAI and Anthropic, leaving small and medium-sized AI startups waiting in line. Hemant Taneja, a partner at General Catalyst, has sent questionnaires to portfolio companies regarding computing bottlenecks, and the firm is planning shared computing pools or collective bargaining for its portfolio companies. A more extreme case is the oil industry AI startup Collide, which plans to spend about $500,000 to purchase NVIDIA GPUs directly and rent data center space to run them themselves, in order to avoid waitlists and price uncertainty. When an AI company in the oil industry decides to return to the old path of buying hardware and renting server space, everyone knows that the cloud is already too crowded.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-24

AI data centers become carbon bombs: annual greenhouse gas emissions exceed entire countries, OpenAI, Meta, Microsoft... decarbonization pledges collapse

Similarity 120%關鍵字 openai同分類 zh

2026-04-24

Jensen Huang sends internal memo: 10,000 NVIDIA employees mandated to switch to OpenAI Codex, GPT-5.5 to run on GB200 chips

Similarity 120%關鍵字 openai同分類 zh

2026-04-24

Jensen Huang sends all-hands email embracing OpenAI Codex: over 10,000 NVIDIA employees are already using it, with GPT-5.5 running on GB200

Similarity 120%關鍵字 openai同分類 zh

2026-04-23