AMD just traded gigawatts of GPUs — and stock options — to lock Meta into its future

Why this matters: AMD is effectively selling the future of its data‑center business to bankroll Meta’s AGI race

Think of this as a marriage of convenience – and risk. AMD has agreed to supply roughly 6 gigawatts of inference GPU capacity in custom Helios rack systems to Meta across the latter half of the decade, and it didn’t stop at hardware. The deal comes with performance‑based equity warrants that could give Meta as much as ~10% of AMD if milestones and stratospheric share‑price targets are met. For Meta, it buys prioritized capacity and influence over chip design; for AMD, it nails down long‑term revenue in a market where compute is the choke point. Both sides are betting big on AI’s insatiable demand for inference compute.

Key takeaways

• AMD will deliver up to ~6GW of Instinct GPU inference capacity to Meta, starting with a 1GW tranche in H2 2026 (sources vary on MI450 vs MI540 naming for the semi‑custom parts).
• The agreement includes performance‑based warrants – up to 160M shares – that could amount to roughly 10% of AMD if aggressive milestones and a high AMD share price are hit (TechCrunch first reported the warrant structure).
• This is a strategic move by Meta to diversify away from Nvidia and secure prioritized, energy‑efficient inference racks for its “personal superintelligence” ambitions; it’s also a bet against the compute bottleneck outlined by industry insiders.
• Numbers reported vary ($51B-$100B quoted across outlets). The exact dollar figure depends on scope, FX and whether you count total hardware, services and multi‑year options — treat headline dollar totals cautiously.

What actually changed

Multiple outlets — TechCrunch, 3DJuegos and others — describe a multiyear pact that threads together AMD silicon (custom MI4xx Instinct inference GPUs and next‑gen EPYC CPUs), ROCm software, and a jointly developed rack design through the Open Compute Project. That Helios rack architecture and the semi‑custom MI450/MI540 chips are tuned for Meta’s inference workloads — lower power per token, high throughput for agentic AIs — and ship at scale: 1GW of capacity in H2 2026, ramping toward the full 6GW by around 2030.

The equity side is the uncomfortable bit. AMD has issued performance‑tied warrants to Meta that vest as delivery and commercial milestones are hit, with final tranches conditioned on AMD reaching an eye‑watering share price (TechCrunch says $600 per share for full vesting). That structure guarantees Meta preferential access without an immediate cash equity purchase, and it lets AMD monetize future production in exchange for potential dilution later.

The angle nobody’s soft‑selling

This isn’t a simple supplier contract. It’s a strategic swap: capacity now, equity later. AMD is locking in demand in an overheated market where supply — from fabs to DRAM to packaging — is the real bottleneck. As Google AI Studio lead Logan Kilpatrick warned, the gap between AI compute demand and supply is widening daily, and big buyers are willing to guarantee future orders (and take equity) to secure lanes in the traffic jam. Meta gets priority fabrication, co‑engineered chips and racks; AMD gets predictable order books and a big revenue runway. The uncomfortable truth: AMD is voluntarily concentrating a large slice of its future revenue on a single partner while also ceding upside if its stock rockets — a sensible strategy if you believe device‑agnostic scale wins, risky if market concentration or execution misfires.

The question I’d ask the PR rep

“Why structure warrants to vest at a share price so far above today’s market — is that genuinely a stretch target, or a way to ensure Meta pays almost nothing for a substantial long‑term stake unless AMD has an extraordinary run?” That matters because the economics decide whether this was a fair price for prioritized capacity or a covert transfer of future ownership for a discount.

What to watch next — specific, dateable signals

• H2 2026: first 1GW shipment and early performance reports from Meta’s Helios racks. This is the earliest direct proof the build works at scale.
• Warrant vesting triggers: delivery milestones and AMD share‑price moves. Track quarterly disclosures from AMD for tranche conditions and any early partial vesting.
• AMD’s revenue mix in earnings — look for double‑digit billions from the first GW tranche in the next couple of reporting cycles, as Lisa Su suggested.
• Nvidia’s response and customer chatter: if other hyperscalers follow Meta in equity+capacity deals, we’re watching a structural shift in how compute is procured.

TL;DR

Meta secured prioritized, co‑engineered GPU capacity from AMD — about 6GW over several years — and took performance‑based warrants that could become a large equity stake. The deal is a clear sign that compute scarcity is reshaping supplier relationships: big buyers will trade capital, preferred access and even ownership to guarantee capacity. Watch the H2 2026 1GW delivery and the warrant vesting mechanics — they’ll tell you whether this was a clever lock‑in or AMD signing away future upside for short‑term certainty.