Meta Signs $27 Billion AI Infrastructure Deal with Nebius — Nvidia's Vera Rubin Chips Incoming

In one of the largest AI infrastructure commitments ever made, Meta has signed a five-year agreement with Dutch data center company Nebius for up to $27 billion in GPU capacity. The deal was announced on March 16, 2026, and includes one of the first large-scale deployments of Nvidia's next-generation Vera Rubin architecture.

Deal Structure

The agreement is split into two parts:

• $12 billion in dedicated GPU capacity — predictable baseline access for Meta over five years
• Up to $15 billion in elastic compute — flexible scaling capacity on demand

Nebius will primarily sell the elastic capacity to third-party customers, with Meta acting as a backstop buyer for remaining allocation. This gives Nebius an attractive business model where they can sell at market prices, with Meta as a safety net.

Vera Rubin clusters are expected to be operational from early 2027.

Context: Meta's AI Capital Expenditures Explode

This deal is not a one-off event — it's part of a massive scale-up:

• Meta's total AI-related capital expenditures are expected to reach $135 billion in 2026
• External cloud commitments exceed $40 billion since October 2025
• Nebius signed a $3 billion deal with Meta in November 2025 — this new agreement is a 9x expansion in just five months

Markets reacted: Nebius shares surged 14% on the announcement day, and are up 400% since its New York listing in 2024.

Background: Meta's AI Struggles and Infrastructure Needs

Meta is in a challenging position: the company plans massive AI scale-up, but internal data centers can't keep pace with demand. Meanwhile, the company has faced challenges with its next-generation model (codenamed Avocado), delayed to at least May, after the largest Llama 4 variant (Behemoth) was scrapped.

The result is that Meta is instead investing heavily in infrastructure — securing GPU access to compete with OpenAI, Google, and Anthropic when the models are ready.

What Does This Mean?

For the industry, this sends a clear signal: the AI arms race on the infrastructure side is not slowing down. On the contrary, it's accelerating, despite — or perhaps because of — geopolitical turbulence and market volatility. Nebius, formerly known as Yandex Cloud, is positioning itself as a key player in the global AI infrastructure market.