Skip to content
TECH CEO Daily
AILaunch

AMD Helios AI racks enter production at Advancing AI 2026; MI500 GPUs are due in 2027

AMD claimed up to 30% more tokens per dollar than Nvidia's Vera Rubin NVL72, and said OpenAI expects to bring Helios online from the fourth quarter of 2026.

By · Editor

Added to archive · Covers events of July 23, 2026 · 3 min read · Fact-checked

The 60-second brief

  • 1AMD said Helios, its first rack-scale AI system with 72 MI455X GPUs per rack, was in production on July 23, 2026.
  • 2AMD claimed up to 30% more tokens per dollar than Nvidia's Vera Rubin NVL72, based on its own estimates.
  • 3OpenAI expects Helios online from the fourth quarter of 2026; AMD's MI500 GPUs follow in 2027 and MI600 in 2028.

The news

On July 23, 2026, Advanced Micro Devices (AMD) said AMD Helios, its first rack-scale AI system, was in production, and launched Instinct MI400 Series GPUs and new EPYC server CPUs at its Advancing AI 2026 event in San Francisco, setting dates buyers can plan against.

Each Helios rack combines 72 Instinct MI455X graphics processing units (GPUs) with 18 6th Gen AMD EPYC central processing units (CPUs) and AMD Pensando networking, according to AMD. The company claimed Helios delivers up to 30% more tokens per dollar, a measure of AI output per dollar spent, than the leading competitive system. A footnote says that figure is an AMD Performance Labs estimate against Nvidia's Vera Rubin NVL72 rack, using one model workload, Kimi K2 Thinking, and projected GPU pricing.

AMD also made chip-level claims based on its own testing. It said the MI455X delivers 34 times the token throughput of its predecessor, the MI355X, and that the new MI350P GPU delivers up to 4.2 times more tokens per second per dollar than Nvidia's H200 NVL. It positioned the MI430X for high-performance computing (HPC) and sovereign AI. The top 6th Gen EPYC "Venice" part, the EPYC 9996, has 256 cores and 512 threads.

AMD said AI labs and cloud providers choosing Helios include OpenAI, Anthropic, Meta, Microsoft, Oracle, HUMAIN, Tensorwave, Vultr and Cirrascale. OpenAI expects to bring Helios online beginning in the fourth quarter of 2026, with deployments accelerating through 2027, according to AMD, while Meta has begun testing workloads on Helios racks. AMD named Bull, HPE (HPE), Lenovo and Supermicro (SMCI) as system makers.

On the roadmap, AMD said MI500 Series GPUs are coming in 2027, paired with future EPYC "Verano" CPUs in a Helios 500 rack, followed by an MI600 Series in 2028. AMD put its total addressable market at about $2 trillion in 2030 and released ROCm.ai, an AI-driven platform for building and optimizing GPU software.

"The next phase of AI will span frontier models, agents and physical AI," AMD Chair and CEO Lisa Su said in the announcement.

The numbers

MI455X GPUs per Helios rack
72 (plus 18 EPYC CPUs)
Tokens per dollar vs Nvidia Vera Rubin NVL72
Up to 30% more (AMD estimate)
MI455X token throughput vs MI355X
34x (AMD claim)
EPYC 9996 cores / threads
256 / 512
OpenAI Helios deployments begin
Q4 2026 (OpenAI expectation, per AMD)
MI500 Series GPUs
2027
AMD addressable market estimate for 2030
About $2 trillion

Why CEOs should care

For infrastructure buyers, the value of the event is the calendar: Helios in production in July 2026, OpenAI deployments from the fourth quarter, MI500 in 2027 and MI600 in 2028. Teams planning 2027 AI capacity should ask cloud providers and system makers for firm Helios delivery slots and pricing, and decide whether a 2026 or 2027 purchase fits their model roadmap. Treat AMD's 30% tokens-per-dollar figure as a vendor estimate based on one model and projected pricing, and benchmark your own models before committing.

For CFOs, tokens per dollar is the right lens for inference budgets, but only when measured on your workloads. Ask vendors to quote cost per million tokens for your models and service levels. With AMD promising a new GPU generation each year, finance teams should also revisit useful-life and depreciation assumptions for accelerators bought now.

For boards and CISOs, a production-ready second rack-scale supplier strengthens negotiating positions and reduces single-vendor dependence. Security leaders adopting ROCm.ai or other AI-assisted GPU development tools should apply the same code review and software supply chain controls they use for other open-source components.

The bigger picture

Advancing AI 2026 capped a week in which AMD announced Helios deals with Microsoft on July 20 and Anthropic on July 22. Tech Wire Asia reported that OpenAI's rollout forms the first phase of a six-gigawatt agreement AMD and OpenAI announced in October 2025, and that Meta plans to deploy up to six gigawatts of AMD GPU infrastructure. By benchmarking Helios directly against Nvidia's (NVDA) Vera Rubin NVL72, AMD signaled that it now competes on complete racks rather than individual chips.

What happened next

On August 4, 2026, AMD's second-quarter results listed the launches of Helios and the MI400 Series and the introduction of 6th Gen EPYC among its highlights. Lisa Su said Helios was beginning to ramp, and AMD guided third-quarter revenue to about $13 billion, plus or minus $300 million. On August 31, 2026, AMD said MI355X-based systems had gone live for HUMAIN in Saudi Arabia and that AMD, Cisco (CSCO) and HUMAIN plan up to 250 megawatts of MI400 Series infrastructure, with deployment planned to begin in 2027 and capacity expected to start coming online in the second half of that year.

The next checkpoints are OpenAI's planned Helios deployments starting in the fourth quarter of 2026 and the MI500 launch in 2027.

What “Fact-checked” means

Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.

What we checked
Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
How
A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
Who
The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, . A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
Archive story
It was written after the event. Its dates were checked against the date of the event, including the dates of any later developments it reports.
If something is wrong
“Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error

How we fact-check →

Companies in this story

AMDAI infrastructureInstinct GPUsOpenAI

Earlier coverage of AMD

All AMD coverage →

Written by

Editor · Technology & Business Writer

Hussein is a writer and business technology enthusiast focused on the intersection of technology, entrepreneurship, finance, artificial intelligence, and digital innovation.

CoversAICybersecurityBig TechSaaSStartupsFintech

About this story. Researched from primary sources whenever they are available and fact-checked before publication.

Published by Tech CEO Daily, an independent publication. Masthead · Editorial standards

Follow Tech CEO Daily on Facebook for the day’s top stories in your feed.

Free newsletters

The technology briefing for people running businesses.

Daily, weekly, bi-weekly or monthly. You choose.

How often

The Daily Brief · Monday to Saturday, 7 a.m. ET

Free forever. One click to unsubscribe. We never sell your email.