Analysis
Crusoe and Thinking Machines Lab announced a $65 million annual agreement for Crusoe to power inference workloads for the lab's models, according to Crusoe's own announcement and coverage from The Information. Under the deal, Thinking Machines Lab will serve production workloads -- including its Inkling models and fine-tuned GLM 5.2 and 5.3 variants -- through Crusoe Managed Inference, running on a dedicated deployment of Nvidia HGX B200 systems connected with Nvidia Quantum-2 InfiniBand networking.
Two Companies, Both Freshly Capitalized
Mira Murati founded Thinking Machines Lab in February 2025 after leaving her role as OpenAI's chief technology officer. The company closed a $2 billion seed round in mid-2025 led by Andreessen Horowitz, with Nvidia, Accel, ServiceNow, Cisco, AMD and Jane Street also participating, valuing the company at $12 billion -- one of the largest first funding rounds in Silicon Valley history. The lab has since attracted several former OpenAI colleagues of Murati's, including researchers John Schulman, Barret Zoph and Luke Metz, and separately closed a large Nvidia-backed compute deal in March.
“## Two Companies, Both Freshly Capitalized Mira Murati founded Thinking Machines Lab in February 2025 after leaving her role as OpenAI's chief technology officer.”
Pulse previously covered Crusoe's fundraising history. Crusoe, the AI-focused cloud infrastructure provider, closed its own $3.9 billion round at a $30 billion valuation earlier this month, according to TechCrunch. Both companies are now transacting with each other at meaningful scale within weeks of raising fresh capital -- Thinking Machines as a customer buying inference capacity rather than a peer competing for the same GPU allocation, and Crusoe as the vendor supplying it.
The Numbers In Context
$65 million a year is a modest sum next to either company's headline valuation, but the deal itself is the more informative number: it takes Crusoe Managed Inference past $100 million in contracted annual recurring revenue less than a year after the product line launched, a fast ramp that suggests real customer demand for managed inference specifically, not just raw GPU rental. Crusoe competes in this category against CoreWeave, Lambda, Together AI and the hyperscalers' own inference offerings from AWS, Google Cloud and Azure -- all chasing the same pool of AI labs and application companies that need to run inference at scale without building their own data centers.
What The Deal Structure Tells You
Thinking Machines Lab choosing Crusoe for inference, rather than running everything on the compute it already secured through its own Nvidia relationship, suggests frontier labs increasingly treat training and inference as separate procurement decisions with different vendors -- training capacity locked up through direct chip-and-capital deals, inference workloads shopped competitively among neoclouds on price and reliability. That's a more fragmented compute-buying pattern than the single-mega-partner structure that defined most 2024- and 2025-era AI infrastructure announcements.
The open question is durability: a $65 million annual contract is meaningful revenue for Crusoe today, but Thinking Machines Lab is still a 19-month-old company without disclosed product revenue of its own, and its own compute spending will scale or shrink with how quickly its models find paying customers. Crusoe's bet is that inference demand from labs like Thinking Machines keeps growing faster than any single lab's own infrastructure buildout can absorb.