Meta seals multibillion-dollar Nvidia deal for millions of AI chips

Meta and Nvidia strike huge AI chip deal

Meta Platforms is deepening its relationship with Nvidia in a multibillion-dollar agreement to buy millions of AI processors over several years. The partnership spans Nvidia’s GPUs, CPUs, and networking technology, aimed at powering everything from recommendation engines to new generative features.

People familiar with the deal say it is worth tens of billions of dollars, though no official figure was disclosed. Meta is positioning Nvidia as the backbone of its AI infrastructure even as it pursues in-house chips and evaluates alternative suppliers.

Laying the foundation for the next AI frontier

Nvidia CEO Jensen Huang argues that no one deploys AI at Meta’s scale, combining frontier research with industrial infrastructure for billions of users. The deal involves close co-design across CPUs, GPUs, networking, and software to put the full Nvidia platform into the hands of Meta’s engineers.

Meta CEO Mark Zuckerberg presents the partnership as central to delivering “personal superintelligence” to people around the world. The company will use Nvidia’s Vera Rubin platform to build cutting-edge clusters with a unified architecture linking its own data centers and cloud partners.

What’s inside the partnership

Meta plans to roll out millions of Blackwell GPUs now ramping up, along with next-generation Rubin chips expected from 2027 onwards [Unverified]. Beyond accelerators, the package includes Grace and Vera CPUs, marking the first major hyperscale deployment of Nvidia’s Arm-based Grace processors.

On the networking side, Meta is adopting Nvidia’s Spectrum-X Ethernet platform to provide low-latency, high-throughput links between AI clusters. The company has also embraced Nvidia’s Confidential Computing technology for WhatsApp, enabling new AI-driven features while keeping private messages protected.

Massive AI and data center investment push

The expanded deal comes as Meta prepares a major spending surge to stay competitive in the AI race against other tech giants. For 2026, the company is projecting capital expenditures between 115 billion and 135 billion dollars, nearly double the 72 billion it spent the previous year.

Much of that budget will go into data center expansion and Meta Superintelligence Labs, which focuses on training cutting-edge AI models. Meta has also created a Meta Compute unit to manage its fast-growing infrastructure footprint and supplier ecosystem [Unverified].

Meta hedges on chips but sticks with Nvidia at the core

Even with this huge commitment, Meta says it wants to keep flexibility by matching different workloads with different chips. Its Andromeda ads retrieval engine already runs across Nvidia and AMD hardware as well as its own MTIA accelerators, nearly tripling compute efficiency in some cases [Unverified].

Meta has explored additional options with other cloud and chip providers, including potential use of rival custom accelerators [Unverified]. Still, the new agreement signals that Nvidia remains the central pillar of Meta’s most demanding AI workloads and its long-term “personal superintelligence” roadmap.

Why it matters

• Deal reinforces Nvidia’s dominance in AI chips despite rising custom silicon from rivals.
• Meta is building vast infrastructure to deliver AI features across apps used by billions.
• Surging data center capex shows AI is becoming big tech’s main investment driver.
• Users could see more powerful AI tools inside Facebook, Instagram, and WhatsApp over time.