Compute Infrastructure

NVIDIA & Thinking Machines: The 1GW Infrastructure Era

Dillip Chowdary • Mar 10, 2026

In a historic move that redefines the scale of AI development, **NVIDIA and Thinking Machines Lab** have announced a strategic partnership to deploy at least **one gigawatt (1GW)** of next-generation **NVIDIA Vera Rubin** systems. This collaboration, involving the lab founded by former OpenAI CTO Mira Murati, marks the largest infrastructure commitment in the history of private AI labs.

Architecting the Vera Rubin Cluster

The Vera Rubin platform represents a paradigm shift from the Blackwell architecture. Key technical enhancements include:

The Power Challenge: Liquid Cooling at Scale

Operating a 1GW cluster requires more than just chips; it requires a revolutionary cooling and energy strategy. Thinking Machines Lab is designing bespoke data centers that utilize direct-to-chip liquid cooling and are paired with dedicated SMR (Small Modular Reactor) nuclear facilities to ensure "Carbon-Zero" training cycles.

Protect Your Infrastructure Data

As compute scales, so does the risk of data leakage. Use our M.A.N.A.V. compliant redactor to secure your proprietary datasets.

Data Masking Tool →

Strategic Implications

NVIDIA's investment in Thinking Machines is not merely a supplier-client relationship. It is a fundamental bet on the next generation of human-aligned frontier models. By controlling the entire stack—from the gigawatt of energy to the model architecture—Thinking Machines aims to build AI that is both more powerful and more deeply aligned than existing commercial offerings.