AMD projects explosive data center growth driven by AI inference by 2027

AMD just turned in a quarter that would have seemed like science fiction two years ago. The company’s data center division generated $6.7 billion in Q2 2026 revenue, a 107% jump from the same period a year earlier. Total revenue hit $11.54 billion, up 50% year-over-year, with the data center segment now accounting for 58% of everything AMD sells.

CEO Lisa Su isn’t content to let those numbers speak for themselves. She’s projecting data center sales will exceed $13 billion in Q3 2026 alone, with the full segment more than doubling again by 2027. The engine behind that forecast: AI inference workloads, which AMD expects to surpass training demands for the first time this year.

The inference shift changes everything

AMD’s server CPU revenue grew over 70% year-over-year in Q2 2026, and its Instinct GPU lineup is gaining traction with hyperscalers who need cost-effective inference solutions alongside Nvidia’s offerings. Inference workloads tend to be more price-sensitive than training, which creates an opening for AMD’s value proposition.

New silicon and big-name customers

AMD is stacking its product roadmap accordingly. The Instinct MI350, MI400, and MI450 GPUs are all in various stages of development and deployment, with the MI400-series expected to ramp volume in late 2026 into 2027. On the CPU side, the EPYC “Venice” and “Verano” processors are designed to complement those GPUs in AI-optimized server configurations. The company also unveiled the Helios rack-scale platform, a complete system architecture built specifically for AI inference and training at data center scale.

The customer list reads like a who’s who of AI infrastructure. Meta signed a multi-year GPU deal with AMD involving up to 6 gigawatts of compute capacity. Anthropic, the company behind Claude, committed to 2 gigawatts of MI450 systems beginning in 2027.

Wall Street’s mixed feelings

Despite numbers that most companies would frame in gold, AMD shares dropped 8-9% in after-hours trading following the earnings release. The primary concern appears to be supply chain constraints. Doubling data center revenue requires doubling production capacity, and AMD relies on TSMC for its most advanced chip manufacturing. Any bottleneck in wafer allocation, packaging, or high-bandwidth memory could create a gap between what AMD promises and what it delivers.

There’s also the Nvidia factor. Jensen Huang’s company still commands the lion’s share of the AI accelerator market, and its upcoming Blackwell and Rubin architectures are designed to maintain that lead. AMD isn’t trying to dethrone Nvidia so much as it’s positioning itself as the essential second supplier, a role that hyperscalers actively want filled to avoid single-vendor dependency.

What to watch from here

Lisa Su has laid out a roadmap where data center revenue more than doubles in 2027. The partnerships with Meta and Anthropic provide contractual underpinning, but converting commitments to shipped silicon at scale is where chipmakers succeed or stumble.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.