Meta on Tuesday pulled the curtain back on a massively scaled AI hardware pact with Nvidia, locking in millions of chips that will feed its U.S. data centers and, crucially for Ohio, supercharge the Prometheus cluster rising just outside Columbus. The expanded multiyear partnership covers standalone Nvidia Grace CPUs along with current Blackwell and upcoming Rubin GPUs, supplying the raw compute for Meta’s multi‑gigawatt buildouts at Prometheus near New Albany, Ohio and the Hyperion campus in Richland Parish, Louisiana as the company races to train and serve new AI models and services.
What Meta Is Installing
In a detailed announcement, Nvidia said the multiyear, multigenerational deal will bring millions of NVIDIA Blackwell and Rubin GPUs into Meta data centers, tied together with Spectrum‑X Ethernet networking and supported by the first large‑scale rollout of Grace CPUs as standalone servers. The company also flagged joint engineering efforts and what it called "deep codesign" work to tune Meta’s AI workloads, along with a roadmap that includes deploying Vera CPUs in 2027. NVIDIA framed the effort as spanning both Meta’s on‑premises infrastructure and its use of cloud platforms.
Where It Fits In Meta’s Spending Plans
Meta is treating the Nvidia pact as a core part of its planned spending surge. In its Jan. 28 earnings release, the company guided investors to expect $115–$135 billion in capital expenditures for 2026, with AI infrastructure a central focus. One thing Meta and Nvidia did not spell out is how much this particular hardware tranche will cost, a missing detail highlighted in outside coverage. Meta laid out the 2026 capex range, while reporting from Reuters-syndicated outlets underscored that the companies declined to disclose the deal’s dollar value.
Local Footprint: New Albany And Richland Parish
On the ground, those chips are headed for some of the largest construction projects in their regions. Meta is already building out the Prometheus data center complex in New Albany and the Hyperion campus in Richland Parish, and the fresh Nvidia gear is destined for those halls of servers. Industry reporting has tied a recent land purchase outside New Albany to a Meta‑linked contractor, suggesting further expansion of the Ohio campus, while Meta has set up a joint venture with Blue Owl Capital to push Hyperion forward. Data Center Dynamics detailed the Ohio land buy, and the City of New Albany has issued public statements clarifying how local energy arrangements are expected to work around the Prometheus site.
What Analysts Say
Industry watchers see the move as another strong signal that Nvidia remains the default choice for hyperscale AI infrastructure, at least for now, and that scarce next‑generation GPUs will be spoken for well before they roll off the line. "The deal is certainly in the tens of billions of dollars," analyst Ben Bajarin told Axios, a ballpark that helps explain why shares of both companies reacted once the news hit. Commentators also say the pact complicates life for rival chipmakers and raises questions about how far Meta can really push its own in‑house silicon while committing so heavily to Nvidia’s roadmap.
Why Nvidia's CPUs Matter
GPUs get the headlines, but the CPU choice is doing a lot of quiet work in this story. Meta’s decision to roll out stand‑alone Arm‑based Grace CPUs now, with Vera successors on the horizon, points to a performance‑per‑watt strategy that shifts certain inference and back‑end tasks away from more expensive GPU racks. Coverage of the deal has emphasized the companies’ close co‑design work and the potential efficiency gains for large‑context models that can better exploit this division of labor. The Verge and other outlets have laid out the Vera and Rubin roadmap and the technical logic behind the overall architecture.
What To Watch Next
From here, the big questions are mostly about timing and execution. Observers are tracking when Blackwell and Rubin systems actually start landing in Meta racks, how quickly the company scales out Grace‑only servers, and whether Vera CPUs really begin appearing in production deployments in 2027. Local communities near the New Albany and Richland Parish sites, along with suppliers and investors, will be paying close attention to how fast the hardware shows up and how Meta mixes Nvidia gear with its own chips as that 2026 capex ramp kicks in. For those following the silicon itself, technical outlets such as Tom's Hardware are closely watching Rubin and Vera milestone dates and what they signal about the next wave of AI infrastructure.
