Meta Platforms has entered into a sweeping long-term infrastructure partnership with Advanced Micro Devices (AMD), securing up to 6 gigawatts of AI and CPU compute capacity. The agreement marks one of the largest AI hardware commitments disclosed to date.
Initial Deployment Scheduled for Late 2026
The first phase of the rollout—representing roughly one gigawatt of compute—will begin shipping in the second half of the year. This tranche includes:
-
A customized version of AMD’s Instinct MI450 AI accelerators
-
Sixth-generation EPYC server processors, internally codenamed “Venice” and “Verano”
Meta will act as the launch customer for both chip platforms, positioning the company at the forefront of AMD’s newest data center architecture.
Inside AMD’s Zen 6 Architecture
The Venice and Verano processors are powered by AMD’s next-generation Zen 6 CPU microarchitecture. Unlike incremental updates, Zen 6 represents a fresh design approach focused on maximizing throughput for hyperscale data centers and high-performance computing environments.
AMD reports performance improvements of up to 70% over previous EPYC generations under select workloads.
Two Core Configurations
Zen 6 will be available in two distinct implementations:
-
Standard Zen 6
-
Up to 96 cores / 192 threads
-
Built using 12-core chiplets
-
Designed for balanced enterprise workloads
-
-
Zen 6c (Dense Variant)
-
Up to 256 cores / 512 threads
-
Optimized for maximum core density
-
Reduced feature set to enable higher scaling
-
In addition to core increases, Zen 6 is projected to more than double per-socket memory bandwidth compared to fifth-generation EPYC processors, reaching approximately 1.6 TB/s.
Rack-Scale AI: The Helios Platform
The processors and accelerators will operate within AMD’s Helios rack-scale system architecture and leverage the company’s ROCm software stack.
Helios was co-developed by AMD and Meta under the umbrella of the Open Compute Project. The initiative focuses on scalable, open infrastructure designs tailored for hyperscale AI deployments.
Financial Structure: An Unconventional Approach
While official financial terms were not disclosed, reporting from the The Wall Street Journal suggests the agreement exceeds $100 billion in total value. Each gigawatt of compute capacity could represent tens of billions in potential revenue for AMD.
Rather than relying on a traditional cash transaction, the deal includes an equity-based incentive structure. AMD has granted Meta warrants to purchase up to 160 million shares at a nominal price of $0.01 per share.
With approximately 1.6 billion shares outstanding, full exercise of the warrants would equate to nearly a 10% ownership stake. However, the structure is performance-based:
-
Shares vest progressively as Meta commits to additional compute capacity.
-
The final allocation depends on AMD’s stock reaching $600 per share.
- AMD currently trades at just above $200.
Parallels to AMD’s OpenAI Agreement
This structure closely mirrors the partnership AMD signed with OpenAI in October of last year. That arrangement also covered 6 gigawatts of GPU compute and included warrants tied to milestone achievements.
Meta Maintains Multi-Vendor Strategy
Despite the scale of its AMD commitment, Meta continues to diversify its hardware suppliers. Recently, the company confirmed it will deploy standalone Grace CPUs from Nvidia in its production data centers, citing significant gains in performance per watt.
Strategic Implications
This agreement underscores several industry trends:
-
Hyperscalers are securing AI compute capacity years in advance.
-
Chipmakers are increasingly structuring mega-deals around equity incentives.
-
Rack-scale AI design is becoming central to next-generation infrastructure.
If fully executed, the Meta–AMD partnership could reshape competitive dynamics across the AI semiconductor landscape.