AMD’s $60B Meta Deal: Why Big Tech Is Finally Cutting Nvidia’s Chain

On February 24, 2026, AMD and Meta announced a five-year strategic partnership to deploy 6 gigawatts of custom AMD Instinct GPUs across Meta’s AI data centers. Reuters estimates the AMD Meta chip deal at roughly $60 billion, making it one of the largest non-Nvidia AI infrastructure commitments ever signed. It sent AMD shares surging 8.77% to close at $213.84, with trading volume hitting 79.8 million shares — 120% above the three-month average.

But buried inside the headline number is a detail that tells you everything about who really holds the power in AI chips: AMD gave Meta a warrant for up to 10% of the company. Nvidia, which signed its own $50 billion+ Meta deal just one week earlier, gave up nothing.

The AMD Meta Chip Deal Structure: Estimated at $60B+

The partnership centers on a custom AMD Instinct GPU based on the MI450 architecture, co-designed with Meta and optimized for inference workloads. According to AMD’s official announcement, the deal also includes 6th Gen EPYC “Venice” CPUs — Zen 6 architecture on a 2nm process with up to 256 cores — and AMD’s Helios rack-scale system. The first 1-gigawatt deployment ships in the second half of 2026, scaling to 6 gigawatts over five years across multiple GPU generations.

The Helios rack packs serious hardware: 72 MI455X accelerators, 31 TB of HBM4 memory, and up to 2.9 FP4 exaFLOPS for inference. This is not AMD recycling off-the-shelf silicon — Meta helped shape the MI450 specifically for the inference demands of serving AI assistants across Facebook, Instagram, and WhatsApp at billions of queries per day.

Then there is the equity. AMD issued Meta a performance-based warrant for up to 160 million shares of common stock at $0.01 per share — approximately 10% of the company. Vesting is tied to two conditions: shipment tranches aligned to the 1GW-to-6GW deployment ramp, and AMD stock price hurdles with the final tranche vesting at $600 per share. The warrant expires on February 23, 2031. This deal fits squarely into Big Tech’s $650 billion AI infrastructure spending plans for 2026.

The Nvidia Comparison: Equity vs. Dominance

Here is the number that matters most for competitive positioning: zero. That is how much equity Nvidia gave Meta when it expanded its own partnership on February 17 — just one week before the AMD deal. Meta’s expanded Nvidia partnership includes millions of Blackwell and Rubin GPUs, Nvidia’s first standalone Grace CPU deployment, and Spectrum-X Ethernet networking. No disclosed equity component.

CNBC’s analysis put it bluntly: AMD needed to sweeten the deal with equity because its technology does not yet command Nvidia-level pricing power on merit alone. That asymmetry is uncomfortable for AMD bulls but impossible to ignore. When Jim Cramer asked Lisa Su directly why Meta was taking a stake in AMD when Nvidia did not offer equity, Su called the structure “a win for AMD, a win for Meta, and a win for our shareholders,” projecting a $1 trillion AI accelerator market over five years.

The market’s verdict was telling. AMD surged 8.77%. Nvidia moved 0.68%. Wall Street treated this as a transformative event for AMD — not a threat to Nvidia. And the combined dilution from both the Meta and October 2025 OpenAI warrants could reach roughly 20% of AMD’s 1.63 billion outstanding shares if all milestones are hit. That is a real cost. The question is whether guaranteed, multi-year revenue visibility justifies it. For context on why inference hardware is diverging from Nvidia, the trend predates this deal by months.

Meta’s Dual-Supplier Strategy: Not AMD vs. Nvidia

Framing this as AMD versus Nvidia misses the point. Meta guided $115 to $135 billion in 2026 capital expenditure — up from the $72.2 billion it spent in 2025. There is more than enough money for both vendors. Meta’s official announcement framed the AMD partnership as “an important step” in diversifying compute as the company pursues “personal superintelligence.”

The MI450 is optimized for inference — responding to user queries at scale — while Nvidia’s Blackwell and Rubin GPUs remain the standard for training frontier models. Meta is splitting workloads across architectures, not replacing one vendor with another. This mirrors a broader pattern: Google is ramping Ironwood TPUs, Amazon is pushing Trainium, Microsoft is developing Maia, and OpenAI is building custom ASICs with Broadcom.

The combined Big Tech AI capital expenditure for 2026 is approximately $650 billion — roughly double what the same companies spent a year earlier. Every major hyperscaler is building toward multi-vendor AI infrastructure. Meta just put the biggest dollar figure behind it.

AMD’s Inflection Point: From Challenger to Pillar

The financial trajectory backs up the ambition. AMD reported full-year 2025 revenue of $34.6 billion, up 34% year-over-year, with Q4 2025 hitting a record $10.3 billion. Data center revenue reached $4.3 billion in Q3 2025 alone, driven primarily by 5th Gen EPYC processors and Instinct MI350 GPUs. AMD projects an 80%+ compound annual growth rate in AI revenue over the next three to five years, targeting tens of billions annually by FY2027.

The Meta deal, paired with the structurally identical OpenAI partnership from October 2025, provides unprecedented revenue visibility. Lisa Su told Fortune that “we’re early in the cycle of seeing what the ultimate payoff can be.” Combined, the two deals create multi-year revenue commitments that analysts estimate could exceed $12 billion — a floor, not a ceiling, assuming deployment milestones are met.

AMD still commands less than 10% of the AI chip market. But these are not speculative design wins — they are signed, multi-year commitments from two of the most aggressive AI infrastructure spenders on the planet. The gap between current market share and contracted pipeline is where the investment thesis lives. Nvidia’s $20 billion Groq licensing deal shows how seriously the incumbent takes the inference threat.

Wall Street Verdict: 33-67% Upside, But With Caveats

Analyst reaction was broadly bullish. Here is where the major price targets landed after the announcement:

Firm	Analyst	Price Target	Implied Upside
Evercore ISI	Mark Lipacis	$358	67.4%
Rosenblatt	Kevin Cassidy	$300	40.3%
Jefferies	Blayne Curtis	$300	40.3%
Consensus Average	—	$285.07	33.3%

Benzinga reported one analyst calling the deal “receipts” for AMD’s AI roadmap — meaning the company finally has signed commitments backing up what had previously been a forward-looking narrative. The bull case writes itself: guaranteed volume, custom silicon, and a massive inference opportunity as AI deployment scales beyond training.

The bear case is equally straightforward. A potential 20% dilution from combined OpenAI and Meta warrants is not trivial. AMD needed to give away equity to secure volume that Nvidia commands on merit. If execution slips — delayed shipments, software stack issues with ROCm, or Meta shifting workloads back to Nvidia — the warrants create downside asymmetry. AMD’s stock closed at $213.84 on February 24 after the 8.77% surge. The final warrant tranche vests at $600. That is a lot of ground to cover.

What to Watch Next

The AMD-Meta deal is a genuine inflection point for AMD, but it is also a referendum on competitive structure. Nvidia’s ability to command premium deals without equity sweeteners reveals something uncomfortable about where the two companies sit in the market hierarchy — today. Whether that gap narrows depends on execution over the next 18 months.

Four things will determine whether this deal is remembered as AMD’s breakout moment or an expensive bid for relevance:

• Do the H2 2026 first-gigawatt shipments hit schedule? Custom silicon co-developed with a hyperscaler has a history of delays.
• What does AMD’s FY2027 guidance say about AI revenue run-rate?
• Do other hyperscalers — Google, Microsoft, Amazon — announce similar diversification deals? If Meta stays alone, that is a weaker signal.
• Do the warrant milestones get hit, or does the $600 stock price hurdle prove unreachable?

Inference is the true battleground. Training remains Nvidia’s fortress, but as AI shifts from model building toward deployment at scale, AMD’s co-designed MI450 becomes defensible. The era of single-vendor dominance in AI infrastructure is ending. The question is not whether AMD gains market share — the signed deals confirm that. It is whether the 10% dilution proves a bargain for guaranteed revenue or a warning that AMD still cannot compete on silicon alone.