Meta Platforms to deploy 6 gigawatts of AMD Instinct GPUs
Updated
Updated · The Motley Fool · May 2
Meta Platforms to deploy 6 gigawatts of AMD Instinct GPUs
13 articles · Updated · The Motley Fool · May 2
The rollout over several years adds to an October OpenAI deal and underpins AMD's push into AI infrastructure and inference computing.
AMD says annual adjusted earnings per share can exceed $20 within three to five years, with AI revenue reaching tens of billions of dollars by 2027.
Revenue rose 34% to $34 billion in 2025, while upcoming MI450 GPUs and Helios systems are expected to help AMD challenge Nvidia more strongly in data-centre AI.
AMD has secured massive AI deals, but can its supply chain handle the unprecedented gigawatt-scale demand?
Can AMD's cheaper hardware break Nvidia's software stranglehold on the AI industry?
As AI demands gigawatts of power, what is the unspoken environmental cost of this hardware arms race?
The Largest GPU Procurement in History: Meta’s 6GW AMD-Powered AI Data Center Buildout
Overview
In 2026, Meta and AMD launched a historic $60 billion, 5-year partnership to deploy up to 6 gigawatts of AMD-powered AI infrastructure, starting with 1GW in late 2026. This includes the custom MI450X GPU and Helios rack system, delivering a major performance leap and requiring around 3 million GPUs. Meta’s strategy diversifies AI workloads across AMD for inference, Nvidia for training, and potential Google TPUs, while also developing in-house chips. The massive power demand raises environmental challenges, prompting Meta’s investment in nuclear energy. Meanwhile, reliance on TSMC for chip manufacturing creates geopolitical risks, driving Meta’s vertical integration and supply diversification to secure its AI future amid growing regulatory scrutiny.