Meta just signed a massive AMD chip deal as it diversifies beyond Nvidia

Meta $META is buying itself options. The company said Tuesday that it has struck a multiyear, multi-generation agreement with AMD to deploy up to six gigawatts of AMD Instinct GPUs to power its next wave of AI infrastructure, with shipments supporting the first gigawatt expected to begin in the second half of 2026. 

On paper, this is a supply deal. In practice, this is a statement about leverage.

“We’re excited to form a long-term partnership with AMD to deploy efficient inference compute and deliver personal superintelligence,” Meta CEO Mark Zuckerberg said in a press release. “This is an important step for Meta as we diversify our compute.” That word — “diversify” — is doing the heavy lifting. Last week, Meta inked a massive multiyear pact with Nvidia $NVDA. Meta is building its own MTIA silicon. Now, Meta is locking in AMD at gigawatt scale. That’s portfolio construction.

AMD, for its part, is pitching more than just a chip.

The initial deployment pairs the MI450-based GPU with 6th Gen EPYC CPUs, codenamed “Venice,” running ROCm software and built into the Helios rack architecture. Meta will also be a lead customer for “Verano,” a next-generation EPYC processor tuned for workload-specific performance. Lisa Su, AMD’s chair and CEO, framed this as roadmap alignment across “Instinct GPUs, EPYC CPUs and rack-scale AI systems,” placing AMD “at the center of the global AI buildout.”

Translation? AMD wants hyperscaler credibility at the highest level. Meta wants capacity, bargaining power, and a second supplier that can scale.

AMD shares surged as much as 14% on the news as the market digested what a hyperscaler-scale customer does for a company that has been trying to graduate from “credible alternative” to “default second lane.” This partnership leans into AMD’s preferred framing: not a chip drop, a full-stack relationship — silicon, systems, and software — designed around a hyperscaler’s workloads at datacenter scale.

Then, there’s the clause that has Wall Street leaning forward — way, way forward. As part of the agreement, AMD issued Meta a performance-based warrant for up to 160 million shares of AMD common stock. The tranches vest as shipment milestones are achieved — the first at one gigawatt, more as purchases scale toward six — and are tied to stock-price thresholds and Meta hitting technical and commercial benchmarks.

That’s Meta turning a supplier relationship into an execution contract with equity attached. If AMD delivers, Meta gets a cheap path to meaningful ownership. If AMD slips, Meta doesn’t just lose time; AMD loses upside. The structure makes one thing clear: In the AI hardware economy, reliability is now a feature you can put a price on.

It’s an unusual structure for a chip supply deal and a clear signal that this is about alignment as much as allocation. Jean Hu, AMD’s CFO, said the partnership is expected to drive “substantial multi-year revenue growth” and be accretive to non-GAAP earnings per share, adding that the performance-based structure “tightly aligns AMD and Meta around execution and long-term value creation.” The warrant mechanics immediately drag a second conversation into the room: dilution, optics, and whether “alignment” is also shorthand for “this took extra incentive to close.” Both things can be true at once. 

Estimates floating through coverage put the deal’s value anywhere from about $60 billion over five years to more than $100 billion, depending on whose math you trust and how aggressively you translate “gigawatts” into revenue. Meta and AMD didn’t print a single clean total in the announcement, which leaves the market to do what it always does: argue about magnitude while the infrastructure gets ordered anyway. 

Regardless, that up-to-$100 billion looks like a drop in the bucket for a company that is throwing money at its AI projects. Meta is planning AI infrastructure spending that could reach as much as $135 billion this year. And between Meta’s multiyear agreement for millions of Nvidia AI chips days ago and Tuesday’s AMD news, the message is less romance than risk management: Meta wants more than one lane into the future, and it wants those lanes paved early.

Right now, compute is scarce. Power is constrained. Nvidia’s dominance is real. So Meta is hedging, booking, and building all at once. The six-gigawatt figure lands like infrastructure language because that’s what it is: data centers, racks, electricity, long procurement cycles.

Meta is building its AI empire under real-world constraints: power, delivery schedules, and the inconvenient fact that everyone wants the same hardware at the same time. This AMD partnership doesn’t solve that problem. But it does formalize Meta’s answer to it — with a gigawatt target and a warrant-shaped enforcement mechanism. So AMD gets a marquee customer and a shot at becoming a default alternative. Meta gets more supply — and a deal structure that tries to make late shipments someone else’s problem, too.