AMD introduced Helios and said Anthropic plans up to 2 gigawatts of GPUs through it.
Why it matters: Helios moves AMD from selling chips to pitching full AI racks, where buyers commit to much larger systems and budgets. Anthropic's planned deployment gives AMD a concrete demand signal as it tries to win more data-center spend.
- AMD unveiled Helios at its Advancing AI conference in San Francisco on July 23, 2026.
- AMD said Helios will start shipping in the second half of 2026, with broader deployment expected in 2027.
- Helios combines Instinct MI455X GPUs, EPYC 'Venice' CPUs, Pensando networking and ROCm software.
- AMD said Anthropic plans to deploy up to 2 gigawatts through Helios, while OpenAI, Meta and Oracle are among expected deployers.
AMD unveiled Helios at its Advancing AI conference in San Francisco and said the system is designed as a rack-scale platform for frontier-model training and large-scale inference. The company is pitching the system as an integrated alternative to buying separate components. AMD's announcement says Helios is built around a full rack architecture rather than a standalone accelerator.
The company said Helios combines Instinct MI455X GPUs, EPYC 'Venice' CPUs, Pensando networking and ROCm software. AMD described the system as its "highest-performance AI rack" in the announcement, but that is AMD's own characterization rather than an independently verified benchmark.
AMD said Helios is scheduled to begin shipping in the second half of 2026, with wider deployment expected in 2027. Reuters reported the launch and AMD's timing, while noting the system is meant to compete more directly with Nvidia's AI infrastructure offerings.
The clearest named demand signal came from Anthropic. AMD said the AI company plans to deploy up to 2 gigawatts of GPUs via Helios. Anthropic's planned deployment is a company-stated commitment, not a completed installation, but it underscores how large AI buyers are now thinking in rack and cluster-scale power terms. Reuters also reported the Anthropic plan.
AMD said OpenAI, Meta and Oracle are among expected deployers of the platform. The company also cited Microsoft in related coverage, but its July 20 partnership announcement with Microsoft only confirmed plans for Microsoft to deploy next-generation AMD AI infrastructure on Azure, not a specific Helios rollout. Microsoft's partnership announcement does not name Helios.
AMD is trying to turn a product launch into evidence of real customer pull. That matters because rack-scale systems require buyers to commit far more capital, power and infrastructure than a single chip purchase.
By the numbers
- 2 gigawatts - Anthropic's planned GPU deployment through Helios
- Second half of 2026 - when AMD said Helios will start shipping
- 2027 - when AMD expects broader deployment
Yes, but: AMD's customer list mixes confirmed plans with expected deployers, so the scale of actual Helios revenue still depends on follow-through.
What's next: AMD said Helios shipments begin in the second half of 2026, with broader deployment expected in 2027.