This is the largest GPU supply agreement ever announced. AMD and Meta have signed a definitive multi-year, multi-generation partnership to deploy up to 6 gigawatts of AMD Instinct GPUs across Meta’s global AI infrastructure — with first shipments supporting the initial 1GW deployment expected in the second half of 2026. The deal includes a custom AMD Instinct GPU derived from the MI450 architecture, 6th Gen EPYC CPUs codenamed “Venice,” and the AMD Helios rack-scale architecture co-developed with Meta through the Open Compute Project.
AMD × Meta 6GW Partnership: Full Breakdown
AMD has also issued Meta a performance-based warrant for up to 160 million shares of AMD stock — vesting as shipment milestones and stock price thresholds are hit, with exercise tied to Meta achieving specific technical and commercial milestones. It’s a rare structure that directly ties AMD’s equity upside to Meta’s purchasing execution.
| Detail | Info |
|---|---|
| Total Deployment | Up to 6 gigawatts of AMD Instinct GPUs |
| Custom GPU | AMD Instinct MI450-based — co-designed for Meta’s workloads |
| First Shipments | 2H 2026 — supporting initial 1GW deployment |
| CPU Partner | 6th Gen AMD EPYC “Venice” — lead customer; “Verano” next-gen also confirmed |
| Rack Architecture | AMD Helios — co-developed by AMD + Meta via Open Compute Project |
| Software Stack | AMD ROCm |
| GPU Already Deployed | AMD Instinct MI300 + MI350 series — millions of units in Meta infra |
| EPYC CPUs | Millions deployed across Meta’s global data center fleet |
| Performance Warrant | 160 million AMD shares — vests with 1GW milestone; scales to 6GW |
| Warrant Stock Price Condition | AMD must hit specific price thresholds |
| Financial Impact | AMD CFO: “Substantial multi-year revenue growth; accretive to non-GAAP EPS” |
| Zuckerberg’s Goal | “Deliver personal superintelligence” — Meta’s stated AI end game |
| Lisa Su Quote | Partnership “places AMD at the center of the global AI buildout” |
Why 6 Gigawatts Is a Number That Changes the Industry
For context: Nvidia’s entire revenue from data centre GPUs in 2024 was roughly $47 billion — fuelled by demand measured in hundreds of thousands of H100s and B200s. A 6GW AMD Instinct commitment from a single customer, spanning multiple GPU generations, is the kind of order volume that determines foundry capacity planning, HBM allocation, and packaging supply chains for years ahead.
The MI450-derived custom GPU is the most strategically significant detail. Meta isn’t buying an off-the-shelf Instinct GPU — it’s getting silicon co-designed for Meta’s inference and training workloads specifically. This mirrors what Google did with its TPUs and what Amazon did with Trainium: vertically-integrated AI silicon that extracts more useful compute per watt from workloads than general-purpose GPU architectures.
The AMD Helios rack-scale architecture — co-developed with Meta through the Open Compute Project and announced at the 2025 OCP Global Summit — means the entire power delivery, cooling, networking, and compute stack is being built as an integrated system rather than assembled from commodity components. At gigawatt scale, rack-level efficiency gains compound enormously: a 5% improvement in power efficiency across 6GW saves 300MW of continuous power draw — more than many mid-sized countries consume for their entire digital infrastructure.
The 160 million share warrant is unusual in its structure. AMD is incentivised by its own stock price to ensure Meta’s shipments arrive on time — and Meta is incentivised by technical and commercial milestones to actually absorb 6GW rather than renegotiate. It’s mutual accountability baked into the financial instrument.
For Nvidia, this is the clearest signal yet that hyperscale buyers are diversifying away from single-vendor GPU dependence. Meta joining Google, Microsoft, and Amazon in building non-Nvidia AI silicon relationships is structural, not opportunistic.
For the full official announcement, visit AMD’s newsroom and Meta’s blog.
Check our AMD Instinct and AI GPU news on TechnoSports and our Meta AI infrastructure and data centre coverage 2026.
