Meta signs $60 billion AMD chip deal, gaining a 10% stake in NVIDIA's biggest rival

Meta and AMD announced a five-year agreement on February 24, 2026, to deploy up to six gigawatts of AMD Instinct GPUs across Meta's AI data centers.^[1][2] The deal — valued at approximately $60 billion and potentially exceeding $100 billion if fully executed — includes a performance-based warrant granting Meta the option to acquire up to 160 million AMD shares, roughly 10% of the chipmaker, at an exercise price of one cent per share.^[3][4]

The announcement came exactly one week after Meta struck a separate multiyear partnership with NVIDIA to deploy millions of Blackwell and Rubin GPUs, Grace and Vera CPUs, and Spectrum-X Ethernet networking.^[7] Together, the two deals represent Meta's clearest articulation yet of a "portfolio approach" to AI infrastructure: NVIDIA for training, AMD for inference, and its own in-house MTIA silicon program running alongside both.^[2]

AMD stock jumped nearly 9% on the news. Meta shares traded slightly lower. NVIDIA was largely flat.^[3]

The deal

Under the agreement, Meta will deploy custom AMD Instinct GPUs based on the MI450 architecture — AMD's first custom GPU effort at this scale — paired with 6th-generation AMD EPYC "Venice" CPUs and running on AMD's ROCm software stack.^[1][2] The hardware will be built on the Helios rack-scale architecture, which AMD and Meta co-developed through the Open Compute Project and unveiled at the OCP Global Summit in 2025.^[1][2]

Shipments supporting the first one-gigawatt deployment are scheduled to begin in the second half of 2026.^[1] The agreement spans five years, with the first gigawatt committed and subsequent tranches scaling to the full six gigawatts.^[5][6]

AMD CEO Lisa Su said the agreement covers future generations as well, noting that AMD is already working on "MI500 and beyond."^[5] Meta will also be a lead customer for AMD's next-generation EPYC processor, codenamed "Verano," which includes workload-specific optimizations aimed at performance per watt.^[5][6]

The equity structure

The financial structure is what separates this from a conventional procurement deal. AMD issued Meta a performance-based warrant for up to 160 million shares of AMD common stock — about 10% of the company — at an exercise price of $0.01 per share.^[3][4]

The warrants vest in tranches tied to deployment milestones: the first tranche upon delivery of the initial one gigawatt of Instinct GPUs, with additional tranches as Meta's purchases scale toward six gigawatts.^[1] Vesting is further conditioned on AMD's stock reaching certain price thresholds, with the final tranche requiring AMD shares to hit $600 — roughly three times the pre-announcement price of $196.60.^[4][5]

Exercise of the warrants is also tied to Meta achieving key technical and commercial milestones, creating what AMD CFO Jean Hu described as a structure that is "accretive to our non-GAAP earnings per share" and "tightly aligns AMD and Meta around execution and long-term value creation."^[1]

The structure mirrors an almost identical deal AMD struck with OpenAI in October 2025: the same six-gigawatt capacity, the same 160 million share warrant, the same milestone-based vesting.^[8] Between the two agreements, AMD has committed to deploying 12 gigawatts of AI compute for its two largest customers, with both ramps beginning in the second half of this year.

The numbers in context

The Wall Street Journal reported the deal's total value could top $100 billion.^[4] Jon Peddie Research estimated the base agreement at approximately $60 billion over five years.^[3][4] AMD has said each gigawatt generates "significant double-digit billions of dollars" in data center AI revenue, with revenue beginning in the second half of 2026.^[5]

JPMorgan analyst Harlan Sur estimated that Meta's planned six-gigawatt deployment could generate $23 billion to $25 billion in annual revenue from 2027 to 2030, or roughly $90 billion to $100 billion over four years.^[9]

Meta's 2026 capital expenditure guidance stands at $115 billion to $135 billion, up from $72.2 billion in 2025 — a roughly 75% year-over-year increase.^[3][4] The company has pledged to invest at least $600 billion in U.S. data centers and AI infrastructure through 2028, with plans for 30 data centers, 26 of which will be based in the United States.^[3]

Why this deal matters for the chip market

NVIDIA controls approximately 90% of the AI GPU market, with a valuation of roughly $4.7 trillion.^[3] AMD, valued at around $320 billion pre-deal, has historically been positioned as a distant second. This agreement changes that framing.

As we reported in our coverage of NVIDIA's Q4 FY2026 earnings, the broader AI infrastructure cycle is entering a new phase. Hyperscalers are no longer simply buying whatever GPUs they can get — they are diversifying supply chains, co-designing custom silicon, and using equity structures to bind chip suppliers into long-term partnerships. NVIDIA's own results showed record revenue of $68.1 billion, but the market punished the stock 5.5% the following day, suggesting investors are looking beyond quarterly beats and interrogating the durability of the entire AI spending cycle.

Meta's two-deal strategy illustrates the shift. The NVIDIA partnership, announced February 17, covers training workloads with millions of Blackwell and Rubin GPUs, plus the first large-scale NVIDIA Grace-only CPU deployment.^[7] The AMD partnership, one week later, targets inference workloads with custom MI450 silicon optimized for Meta's specific models.^[1][2] Meta CEO Mark Zuckerberg framed both deals around the same goal: delivering "personal superintelligence to everyone in the world."^[2][7]

Analyst Ben Bajarin of Creative Strategies noted that the AMD deal includes customized GPUs from the start, distinguishing it from Meta's NVIDIA arrangement.^[3] Matt Kimball of Moor Insights & Strategy added that the deal validates AMD's Instinct GPU line in the inference market, calling Meta "one of the largest inference environments in the world."^[6]

The skeptic's view

Not everyone sees the equity structure as a sign of strength. BofA Securities analyst Vivek Arya raised the question directly during AMD's conference call: if demand for AMD's AI accelerators is so strong, why is the company effectively giving away equity to secure orders?^[4]

Su's response was that each deal at this scale is unique, and that the warrant structure creates alignment around execution, ecosystem maturation, and software maturity.^[4] Goldman Sachs analyst James Schneider maintained a Neutral rating on AMD despite raising his price target, citing the company's significant exposure to OpenAI and elevated operating expenses.^[9]

There is also the question of circular ownership. Both Meta and OpenAI now hold warrants for approximately 10% of AMD — their primary alternative to NVIDIA. AMD is simultaneously dependent on both companies for its largest revenue commitments, while both companies are counting on AMD to diversify them away from NVIDIA dependency. If either deal underperforms on deployment milestones, the other becomes proportionally more important to AMD's financial model.

What happens next

Meta already runs millions of AMD EPYC CPUs and significant quantities of Instinct MI300 and MI350 GPUs across its global infrastructure.^[1] The AMD deal is part of what Meta calls its "Meta Compute" initiative — an effort to massively scale infrastructure for what it calls the era of personal superintelligence.^[2]

The company is simultaneously building a $27 billion data center campus in Louisiana through a joint venture with Blue Owl, and a one-gigawatt Prometheus facility in New Albany, Ohio, alongside a five-gigawatt Hyperion site in Richland Parish, Louisiana.^[3]

The first real test comes in the second half of 2026, when both Meta and OpenAI are scheduled to begin receiving their initial one-gigawatt MI450 deployments. AMD confirmed during its most recent earnings call that the OpenAI ramp is "on track."^[6] Whether AMD can execute two simultaneous gigawatt-scale custom deployments — for two of the most demanding AI infrastructure customers in the world — will determine whether this deal represents a genuine inflection point for the AI chip market or an ambitious bet that outran its execution capacity.