Nvidia, Thinking Machines Lab partner on AI systems

Nvidia and Thinking Machines Lab announced a multiyear partnership to deploy at least one gigawatt of next-generation Nvidia Vera Rubin systems to support Thinking Machines Lab’s frontier model training and platforms, with deployment targeted for early next year.

Thinking Machines Lab builds frontier AI models and platforms.

The partnership includes work to design training and serving systems for NVIDIA architectures and to broaden access to frontier AI and open models for enterprises, research institutions, and the scientific community.

Nvidia has also made a significant investment in Thinking Machines Lab to support the company’s long-term growth.

The announcement did not disclose financial terms or additional deployment details.

🔗 Source: Nvidia

🧠 Food for thought

Implications, context, and why it matters.

A gigawatt commitment signals a new era of data center design

A one-gigawatt AI factory takes huge effort. It equals about one-tenth of the at-least-10-gigawatt buildout OpenAI and Nvidia outlined in a letter of intent for OpenAI’s next-generation AI infrastructure 1.
The build goes beyond GPUs. The NVIDIA Vera Rubin platform runs at rack scale. In the Vera Rubin NVL72 configuration, the rack acts as one accelerator, combining Rubin GPUs, Vera CPUs, plus NVLink 6 (Nvidia’s high-speed chip-to-chip interconnect) 2.
This size forces new power gear. Many buyers will move to 800-volt direct current (VDC) systems (a way of delivering power used in some electric vehicles) to move power efficiently and cut material costs such as copper busbars 3.

Nvidia’s annual cadence creates a high-stakes upgrade cycle

Nvidia’s faster release schedule raises the stakes. Rubin targets 2026, then Rubin Ultra follows in 2027, so AI labs feel pressure to keep up with new hardware 4.
Rubin NVL144 is expected to deliver more than three times the FP8 (8-bit floating point) training performance of Blackwell Ultra B300 NVL72. Nvidia has slated B300 NVL72 for the second half of 2025 4.
Open initiatives aim to make upgrades routine. The MGX rack architecture (a set of modular server design guidelines Nvidia promotes for building AI servers) has over 50 partners, which supports wider uptake while keeping Nvidia’s GPUs, CPUs, plus interconnects central to the ecosystem 3.