“Crusoe Managed Inference took us from evaluation to production quickly and met the standards we hold our own systems to, giving us the scale and economics to reinvest in the research at the core of what we do,” said Myle Ott, ML infrastructure lead at Thinking Machines Lab.

Crusoe and Thinking Machines Lab announced on Wednesday a $65m annual agreement. Crusoe will run inference for the lab’s open models on Crusoe Cloud.

Former OpenAI chief technology officer Mira Murati founded Thinking Machines Lab. The lab will serve its production workloads through Crusoe Managed Inference. According to the companies, those include its Inkling models and GLM 5.2 and 5.3. Its own fine-tuned variants will run there too.

A dedicated Nvidia cluster

The workloads will run on a dedicated deployment of Nvidia HGX B200 systems. Nvidia Quantum-2 InfiniBand networking connects them. Crusoe will run and support the cluster directly, as what it calls a Tailored Deployment. That is a dedicated, benchmarked endpoint with a service-level agreement. That means Thinking Machines Lab does not have to manage its own inference stack.

The two companies said they are also looking at batch inference for large-scale synthetic data generation.

“Thinking Machines Lab has a very sophisticated engineering team that expects partners to meet their high bar as they grow,” said Erwan Menard, senior vice president of product management at Crusoe Cloud.

Crusoe’s inference business

Crusoe said the deal takes Crusoe Managed Inference past $100m in contracted annual recurring revenue. The service launched less than a year ago.

The agreement follows a busy month for both companies. Crusoe raised $3.9bn at a $30.9bn valuation last week. Earlier in September it signed a $13bn five-year deal with Jane Street. The Information reported that Thinking Machines Lab is in talks to raise $6bn.

China’s Z.ai develops the GLM open models. Z.ai recently open-sourced its ZCode tool.