AMD is in active discussions with top clients about Instinct GPUs that are still years from shipping, as hyperscalers plan AI infrastructure far ahead

Tue 3 Nov 2026 · 18:19 EST · 3 min read

In Technology

Share

AMD CEO Lisa Su says the company’s biggest customers are already asking about chips that won’t exist for years. According to Su, AMD is in active talks with its largest clients about the MI500, the MI600 and products beyond them.

“We’re in very active dialogs with our largest customers on MI500 and MI600 and beyond.” – Lisa Su, CEO, AMD

“We’re in very active dialogs with our largest customers on MI500 and MI600 and beyond.” – Lisa Su, CEO, AMD

What AMD is actually selling

The near-term piece is the MI400 series, built on the CDNA 5 architecture. AMD’s roadmap has it shipping in volume by 2026.

Next up is the MI500 series, scheduled for a 2027 launch. It is built on CDNA 6 architecture, uses an advanced 2nm manufacturing process, and pairs with HBM4E memory.

The MI500’s stated ambition is large. AMD is targeting more than 2,000× higher inference throughput than the MI300X within four years.

After the MI500 comes the MI600 series, which is expected in 2028. Su’s comments suggest customers are already factoring that generation into their planning too.

Helios and the rack-scale pitch

The company plans annual releases of new Helios rack-scale AI platforms through at least 2030. Each Helios system integrates Instinct GPUs, EPYC CPUs and Pensando networking into one package.

Meta is the showcase customer here. The company has already begun deploying Helios systems and has contributed to the development of AMD’s latest GPUs.

AI, tech, and the markets they move—in one daily briefing.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

We respect your privacy. Unsubscribe anytime.

Meta has also deployed millions of EPYC processors from AMD. It is described as an integral partner in custom Instinct GPU deployment, which makes it less of a customer and more of a co-designer.

The numbers behind the confidence

Su’s comments come with financial backing. AMD reported Q1 AI-related revenue of $5.8 billion in 2026, a 57% year-over-year increase.

MI300-series accelerators made up approximately 73% of that AI revenue.

What this means

Talks are not orders, though. Su described dialogs, not signed commitments, and the gap between a roadmap slide and a shipped product is where chip companies historically stumble.

The MI500 depends on a 2nm process and HBM4E memory, both advanced technologies, and any delay ripples through every customer’s build-out plan.

The Meta relationship is the template to watch. If AMD can replicate that co-development model with other large customers, the MI500 and MI600 conversations could turn into the kind of multi-generation commitments that reshape market share.

The near-term checkpoint is the MI400 volume ramp in 2026. If that goes smoothly, Su’s talk of MI500, MI600 and beyond starts to look less like ambition and more like a schedule.

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

More in Technology