Meta Platforms has committed roughly $21 billion to CoreWeave for specialized GPU capacity through December 2032, a 50 percent increase over the $14 billion deal the two companies signed just six months earlier. The expanded agreement, finalized on March 31, 2026, makes CoreWeave one of the largest external infrastructure suppliers to the company behind Facebook, Instagram, and WhatsApp.
The scale of the commitment points to a blunt reality: even a company spending tens of billions a year on its own data centers cannot build fast enough to keep pace with the computing power its AI products demand.
What the contract covers
CoreWeave disclosed the expansion in a Form 8-K filed with the SEC in early April 2026. According to the filing, the new order form falls under a Master Services Agreement the two companies originally signed on December 10, 2023. Meta’s initial commitment totals approximately $21 billion, with the contract running through December 20, 2032. An existing option within the agreement, exercisable through April 2032, gives Meta room to adjust capacity over the life of the deal.
A press release accompanying the filing fills in operational details. The infrastructure will be deployed across multiple locations and will include initial rollouts of NVIDIA’s Vera Rubin platform, the chipmaker’s next-generation data center architecture built for large-scale AI workloads. CoreWeave said the capacity is intended to support Meta’s inference operations, the computationally intensive process of running trained AI models to generate real-time responses for users across Meta’s family of apps.
The leap from $14 billion to $21 billion in roughly half a year is notable. When the original deal was reported by Bloomberg in September 2025, it ranked among the largest cloud computing commitments on record. The expansion suggests Meta’s internal forecasts for AI compute demand have outpaced even its own planning assumptions.
Why Meta is buying rather than building
Meta has invested heavily in its own data center network. During its fourth-quarter 2025 earnings call, the company reported capital expenditures of $39 billion for the year and projected spending between $60 billion and $65 billion in 2026, with much of that directed toward AI infrastructure. Yet even at that pace, Meta is turning to CoreWeave for supplemental capacity.
Speed is a major factor. Building a new data center from scratch takes years once permitting, power procurement, and construction are factored in. Contracting with CoreWeave lets Meta tap GPU clusters that are either already running or further along in the buildout pipeline. Diversifying infrastructure sources also serves as a hedge: if construction at any single Meta-owned facility hits delays, the company’s AI product roadmap does not grind to a halt.
“This expanded partnership underscores the growing demand for purpose-built AI infrastructure at scale,” CoreWeave CEO Michael Intrator said in the press release. As of mid-April 2026, Meta has not issued its own public statement about the expanded deal, leaving the company’s internal rationale and how it views this spending alongside its direct capital budget largely unaddressed.
What it means for CoreWeave
For CoreWeave, which went public in March 2025, a single contract worth $21 billion provides extraordinary revenue visibility. The company’s SEC filings show that large, long-duration contracts with hyperscale customers form the backbone of its business model.
That model carries concentration risk. Dedicating a significant share of its GPU fleet to one buyer ties CoreWeave’s financial health closely to Meta’s willingness and ability to follow through over six-plus years. CoreWeave also carries substantial debt, having borrowed billions to finance the GPU purchases that underpin its cloud offerings. A long-term, high-value contract like this one helps service that debt, but any disruption in payments or a renegotiation could ripple through the company’s balance sheet.
Investors responded positively. CoreWeave shares climbed in the trading sessions following the disclosure, reflecting optimism about the revenue certainty a deal of this duration provides. Analysts, however, have flagged that heavy reliance on a small number of large customers introduces vulnerability if spending plans shift over the contract’s life.
The agreement also positions CoreWeave as an early deployment partner for NVIDIA’s Vera Rubin platform. That could sharpen its competitive edge against Amazon Web Services, Microsoft Azure, and Google Cloud. If CoreWeave can demonstrate production-scale Vera Rubin deployments before those rivals, it gains a tangible selling point with enterprise customers seeking next-generation AI hardware. Microsoft, notably, is both a CoreWeave investor and customer, making the competitive dynamics in this market unusually layered.
Open questions
Several important details sit beyond the reach of the current disclosure. The SEC filing states Meta “initially committed” to pay approximately $21 billion, but the word “initially” leaves open whether the total could shift upward or downward as the contract unfolds. Payment schedules, annual minimums, and termination penalties are not detailed in the public portions of the 8-K.
The geographic scope is similarly vague. The press release references “multiple locations” but does not name specific data center sites or regions. For a deal of this size, the physical buildout carries real implications for local power grids, real estate markets, and permitting pipelines, none of which can be assessed without location-level detail.
NVIDIA’s Vera Rubin platform, while named in the press release, comes without performance benchmarks, delivery timelines, or volume commitments. The platform has not yet been widely validated in production environments, making the reference more of a directional signal than a firm technical specification.
The broadest uncertainty may be whether AI workload growth will justify this level of spending through the end of 2032. If inference costs drop sharply because of hardware improvements or algorithmic breakthroughs, a fixed commitment of this size could look expensive in hindsight. If demand for AI-powered services keeps accelerating, Meta may find the locked-in capacity to be a significant competitive advantage. The filing does not address scenarios in which Meta might seek to renegotiate or reduce its obligation.
The bigger picture
The Meta-CoreWeave expansion is the latest sign that access to top-tier GPU capacity is being locked up in long-term contracts by the industry’s largest buyers. Microsoft has committed billions to infrastructure supporting OpenAI. Google and Amazon have each announced massive capital spending programs for AI data centers. The cumulative effect is a tightening market for specialized compute, one that could squeeze smaller AI startups and mid-tier enterprises that depend on cloud providers for on-demand access.
For anyone tracking the AI infrastructure arms race, the evidence trail on this deal is unusually clean. The primary document is a legally binding SEC disclosure. The counterparties are named, the dollar figure is explicit, and the timeline is defined. What the evidence does not yet reveal is whether $21 billion over six-plus years represents a fair price, an aggressive bet, or a defensive move by a company worried the computing power it needs will become scarce. That judgment will depend on disclosures still to come from Meta, CoreWeave, and NVIDIA as the Vera Rubin platform moves toward full-scale deployment.
