Nvidia Backs Mira Murati's Thinking Machines Lab With Gigawatt-Scale Vera Rubin Deal Worth Up to $50 Billion

The Biggest AI Compute Deal of 2026

Nvidia and Thinking Machines Lab have announced a multiyear strategic partnership that commits at least one gigawatt of next-generation Nvidia Vera Rubin systems to support Mira Murati's frontier model training ambitions. Nvidia CEO Jensen Huang has previously estimated that 1 gigawatt of AI computing capacity costs approximately $50 billion to build — making this one of the most significant AI infrastructure commitments ever announced.

The deal includes a "significant investment" from Nvidia into Thinking Machines Lab, though neither company disclosed the exact size of the stake. Deployment on the Vera Rubin platform is targeted for early 2027.

What the Partnership Includes

The collaboration goes beyond raw compute allocation:

Key partnership components:

1. Gigawatt-scale Vera Rubin deployment — At least 1 GW of next-gen compute for frontier model training
2. Significant Nvidia equity investment — Undisclosed stake ties Nvidia's financial interests to Thinking Machines' success
3. Joint system design — Co-engineering training and serving systems optimized for Nvidia architectures
4. Broad access mandate — Commitment to deliver frontier AI and open models to enterprises, research institutions, and the scientific community

The partnership positions Thinking Machines Lab alongside only a handful of organizations — OpenAI, Google DeepMind, Anthropic, and xAI — operating at gigawatt-scale compute.

From OpenAI Exodus to Frontier Competitor

The trajectory of Thinking Machines Lab has been anything but smooth. Mira Murati, who served as OpenAI's CTO and briefly as interim CEO during the Sam Altman board crisis in 2023, founded Thinking Machines Lab last year with the mission to make AI systems "more widely understood, customizable and generally capable."

The startup's early months were turbulent. As we reported in January, the company experienced significant co-founder departures, with key personnel leaving or being fired and some returning to OpenAI. The episode highlighted the intense instability in the AI talent market that has defined the sector.

This Nvidia partnership fundamentally changes the narrative. A gigawatt of compute is not a research lab allocation — it's a declaration of intent to build and train frontier models that compete with the largest AI organizations on Earth.

The Vera Rubin Platform

The deal is built on Nvidia's next-generation Vera Rubin architecture, which represents a significant leap beyond the current Blackwell platform that already dominates AI training workloads.

The fact that Thinking Machines Lab secured gigawatt-scale Vera Rubin access before most established labs suggests Nvidia sees strategic value in diversifying its frontier customer base beyond the current incumbents. As we covered in our CES 2026 analysis, the Vera Rubin platform represents Nvidia's next competitive moat against AMD and custom silicon from Google and Amazon.

What $50 Billion in Compute Means

To put the scale in perspective:

Gigawatts of AI compute (approximate, based on public reporting)

One gigawatt is enough electricity to power approximately 750,000 homes. The infrastructure to deploy that much AI compute — including data centers, cooling systems, networking, and power delivery — represents a $50 billion buildout by Nvidia's own estimates. This positions Thinking Machines Lab's compute capacity in the same tier as organizations that have raised tens of billions in capital.

The Broader AI Infrastructure Race

This partnership arrives as the AI infrastructure investment wave continues to accelerate. Nvidia's strategy of making equity investments in its largest compute customers — a model it has employed with CoreWeave, xAI, and others — effectively ties hardware dominance to financial returns from the AI labs themselves.

For Nvidia, the deal:

• Locks in revenue from gigawatt-scale Vera Rubin purchases
• Diversifies frontier customers beyond OpenAI and Google
• Creates financial upside through the equity investment if Thinking Machines succeeds

For Thinking Machines Lab, the deal:

• Provides compute parity with established frontier labs
• Validates the startup after a rocky early period
• Enables frontier model training at a scale that would otherwise require tens of billions in independent fundraising

What to Watch

The key question is whether Thinking Machines Lab can convert gigawatt-scale compute into competitive frontier models. Compute alone doesn't guarantee breakthroughs — algorithmic innovation, data quality, and team execution matter as much or more. But without frontier-scale compute, competing at the frontier is simply impossible.

With Vera Rubin deployment targeted for early 2027, Thinking Machines Lab has roughly a year to build the team, develop the training infrastructure, and design the model architectures that will put this compute to work. The co-founder departures from earlier this year make that timeline even more challenging.

The deal also raises questions about the sustainability of the current AI infrastructure buildout. If Jensen Huang's $50 billion estimate is accurate, the total capital being committed to gigawatt-scale AI compute across all labs now exceeds $200 billion — a figure that will eventually need to be justified by revenue from AI products and services.

---

This is a developing story. We will update as additional details emerge about the partnership terms and Thinking Machines Lab's model development plans.

Sources:

• Nvidia Blog — Official Announcement
• CNBC
• TechCrunch
• Axios
• The Information
• Bloomberg