Thinking Machines Must Secure GPU Capacity

Compute is the real bottleneck, not model ideas, because frontier AI labs only matter if they can turn talent and capital into sustained access to thousands of top tier GPUs. Thinking Machines was founded in 2024 and already carries an estimated $12B valuation with $2B of estimated funding, but the practical gate is whether it can lock in long term training and inference capacity from suppliers that either compete directly in AI or sell through intermediaries with their own pricing power.1 2 4 5

• Hyperscalers are not neutral landlords anymore. Amazon pairs cloud capacity with Bedrock and a deep Anthropic partnership, including an agreement for up to 5 GW of capacity and nearly 1 GW of Trainium2 and Trainium3 by the end of 2026. That shows how the biggest clouds reserve scarce compute for strategic model partners, not just any well funded newcomer.5 6
• Independent GPU clouds exist, but they are expensive and capacity constrained businesses themselves. CoreWeave built its position by becoming one of the largest private GPU operators in North America, while Lambda is expanding owned and leased data center capacity with new 2026 sites. In practice, this means Thinking Machines would likely need multi year commitments and very large prepayments to secure enough chips.7 8 9
• There is a clear split in the market between raw compute sellers and higher layer AI platforms. Some customers buy bare GPU clusters from CoreWeave or Lambda and manage training stacks themselves, while others use companies like Together that rent compute, add APIs and tooling, and resell it at a markup. Thinking Machines needs the first category at scale, because owning frontier model economics requires direct access to the underlying machines.8 9 10

The next phase of the market will reward labs that secure dedicated compute the way Anthropic and Mistral have, through long dated infrastructure deals, cloud alliances, and in some cases owned capacity. For Thinking Machines, GPU access is likely to become the main strategic axis around which fundraising, partnerships, and even product timing are organized.5 6 11