Why it matters
The strategic read
Adds a second publicly disclosed 6 GW AMD hyperscaler lane beside OpenAI. Meta is not just buying accelerators; it is aligning silicon, CPU, rack-scale systems, software and shipment milestones around AMD Helios, while AMD issued a performance-based warrant tied to GPU shipment milestones up to the full 6 GW scale.
Source-supported terms
What was disclosed
- 01
AMD and Meta announced a definitive multi-year, multi-generation partnership on February 24, 2026.
- 02
The agreement is designed to deploy up to 6 GW of AMD Instinct GPUs for Meta next-generation AI infrastructure.
- 03
AMD said shipments supporting the first 1 GW deployment are scheduled to begin in the second half of 2026.
- 04
The first deployment will use a custom AMD Instinct GPU based on the MI450 architecture and optimized for Meta workloads.
- 05
The first deployment also uses 6th Gen AMD EPYC CPUs, codenamed Venice, ROCm software, and AMD Helios rack-scale architecture.
- 06
AMD said Helios was developed jointly by AMD and Meta through the Open Compute Project to enable scalable rack-level AI infrastructure.
- 07
Meta will be a lead customer for 6th Gen AMD EPYC Venice CPUs and Verano, a next-generation EPYC processor with workload-specific optimizations.
- 08
AMD said it issued Meta a performance-based warrant for up to 160M AMD common shares.
- 09
The first warrant tranche vests with the initial 1 GW of shipments, with additional tranches vesting as Meta purchases scale to 6 GW.
- 10
Meta framed the agreement as part of compute diversification for AI infrastructure and personal superintelligence workloads.
Disclosure boundary
What is known—and what is not
Primary AMD announcement linking to Meta primary announcement. Public terms disclose the definitive partnership, up-to-6-GW scale, first-1-GW shipment timing, MI450-based custom GPU, Venice EPYC CPU, ROCm, Helios rack architecture, roadmap alignment and performance-based warrant. Public materials do not disclose total dollar value, unit pricing, delivery schedule beyond initial shipment timing, GPU counts, data-center site allocation, power procurement, liquid-cooling or electrical packages, cancellation rights, warranty economics, Meta capex accounting, or full warrant agreement terms.