Meta has entered into a long-term partnership with AMD that will see the chipmaker supply Instinct GPUs for Meta’s expanding artificial intelligence infrastructure.

The arrangement builds on an established working relationship between the two firms and introduces closer coordination across silicon design, system architecture and software development. Rather than operating as a straightforward purchasing contract, the structure creates shared planning across future product cycles, allowing Meta to integrate more directly with AMD’s hardware roadmap.

Initial rollouts will be based on the Helios rack-scale platform, a high-density rack system jointly engineered by Meta and AMD and first revealed at the Open Compute Project Global Summit last year. By grounding the new agreement in infrastructure that has already been co-developed, both companies are accelerating deployment using a framework designed specifically for hyperscale AI environments.

The commitment, valued at approximately A$9.2 billion based on a 6GW capacity allocation, positions AMD as a credible counterweight to Nvidia within the large-scale AI compute market. Nvidia’s H100 and Blackwell processors have accounted for the majority of enterprise and hyperscale AI installations over the past two years, but Meta’s expanded reliance on AMD signals a more competitive supplier landscape.

Meta has presented the agreement as part of its wider Meta Compute strategy, which centres on diversifying hardware sources while continuing to advance its in-house Meta Training and Inference Accelerator programme. By blending third-party silicon with proprietary development, the company aims to limit reliance on any single vendor and increase resilience across its supply chain.

This multi-vendor model mirrors a broader industry trend. Technology groups such as Google, Microsoft and Amazon have increasingly invested in custom-designed chips and distributed GPU sourcing strategies rather than depending solely on external accelerator providers as AI workloads grow at scale.

Under the expanded collaboration, roadmap coordination extends beyond Instinct GPUs to include AMD EPYC processors and rack-level AI systems. The partnership effectively enables joint optimisation of hardware and software stacks ahead of commercial launches, providing opportunities to refine performance, efficiency and integration before products reach the wider market.

Meta has linked the infrastructure expansion to its ambition to develop what it describes as personal superintelligence, a class of applications requiring continuous scaling of both training and inference workloads across global data centres. Supporting these simultaneous compute demands necessitates substantial capacity growth and architectural flexibility.

The first Helios-based GPU deployments are scheduled for the second half of 2026. Additional milestones, including further capacity figures and implementation phases, are expected to be disclosed as Meta’s AI infrastructure programme advances over the course of the year.