AMD Helios AI Servers Enter Full Production, OpenAI Commits to Deployment
AMD CEO Lisa Su confirms Helios AI servers are in full production and will ship by end of Q3 2026, with OpenAI and Anthropic both committed to the platform.
AMD CEO Lisa Su announced on July 23, 2026, at the company's Advancing AI conference in San Francisco that its second-generation Helios AI server racks are in full production and will begin shipping by end of Q3 2026. The Helios platform pairs AMD's MI455X AI accelerator with its new Venice CPU, both made by TSMC. OpenAI's compute VP confirmed plans to deploy Helios at scale before year-end, and AMD separately disclosed a deal to supply Anthropic with MI450 chips starting in the first half of 2027, backed by up to $5 billion in investment.
What happened
| Detail | Fact |
|---|---|
| Product | Helios AI server rack (2nd generation) |
| Key chips | MI455X AI accelerator + Venice CPU, both from TSMC |
| Production status | Full production as of July 23, 2026 |
| Shipping timeline | End of Q3 2026 |
| OpenAI deployment | Massive scale from late 2026, accelerating through 2027 |
| Anthropic chip deal | MI450 chips, up to $5B investment, starting H1 2027 |
| AMD’s 2030 market forecast | $2 trillion total compute; $1.4 trillion AI accelerators; $220 billion CPUs |
| 2025 total compute market | ~$365 billion (AMD estimate) |
Su made the announcement during a keynote at AMD’s Advancing AI 2026 event, where Sachin Katti, OpenAI’s vice president of compute strategy, joined her on stage. Katti said OpenAI expects to deploy Helios “at massive scale” starting toward the end of 2026 and accelerating through 2027. He added that OpenAI plans to use AMD’s next-generation MI500 chips as well.
The OpenAI relationship is not new. In October, AMD announced a multiyear deal with the ChatGPT maker that includes tens of billions in potential annual revenue and gives OpenAI the option to acquire roughly 10% of AMD.
On the Anthropic side, AMD announced plans the day before the keynote to sell up to two gigawatts of MI450 chips to the Claude maker beginning in H1 2027. Tom Brown, an Anthropic co-founder, appeared on stage to say that Claude had autonomously configured AMD’s AI servers over a single weekend.
Cerebras Systems CEO Andrew Feldman also joined Su to announce a partnership combining AMD and Cerebras AI servers, starting inside Cerebras data centers. Cerebras stock rose about 5% on the news; AMD shares fell roughly 2%, a drop analysts at Emarketer attributed to broader semiconductor sector weakness rather than AMD-specific concerns.
Why does AMD’s Helios matter for the AI infrastructure market?
Nvidia currently dominates AI server hardware. AMD is targeting the inference segment specifically: the compute work that happens every time someone sends a query to a model like ChatGPT. Inference is where demand is growing fastest as AI products scale to millions of daily users.
Su did not shy away from the competitive framing. When asked whether AMD was comfortable being second to Nvidia, she said the company believes it will have “leadership in the scale-up compute domain.” That is a direct challenge in the highest-margin part of the market.
Nvidia, for its part, released technical details this week about its Vera CPU, positioning it alongside the Rubin GPU as the most power-efficient option for AI agents. The two companies are now competing not just on raw chip performance but on entire rack-level systems and the software stacks that run on them.
Our earlier coverage of AMD’s Helios rack system and its competitive positioning against Nvidia has more background on the hardware architecture.
Our take
The Helios announcements carry more weight than a typical hardware keynote because the customer commitments are specific and named. OpenAI and Anthropic are not vague “major partners.” They are the two most-watched AI labs right now, and both are putting real chips on the table with AMD.
That said, AMD has made bold timeline promises before. “Shipping at the end of Q3” gives a narrow window, and the gap between full production and at-scale deployment at OpenAI is not trivial. Watch whether Helios racks actually show up in OpenAI infrastructure by December 2026. If they do, AMD’s push into Nvidia’s territory becomes a real story. If the timeline slips, these announcements will look like stage management.
For businesses making decisions about cloud infrastructure or AI workloads, the practical effect is more competition in the market for GPU compute. That tends to put downward pressure on pricing over time. If you are currently planning AI workloads and working with a provider that uses AMD hardware, such as Vultr or TensorWave (both displayed at the AMD event), it is worth asking what Helios availability looks like for your use case. If your team needs help mapping AI capabilities to the right infrastructure, our AI integration services cover exactly that kind of evaluation.
What to do about it
- If you run AI workloads on cloud GPU instances, ask your provider now about Helios availability and pricing timelines.
- Check whether your current AMD-based cloud provider (Vultr, TensorWave) will offer MI455X instances and when.
- If you are planning inference-heavy applications for late 2026 or 2027, factor in potential AMD pricing pressure when budgeting compute costs.
- Follow Anthropic’s Claude API pricing closely: AMD’s MI450 deal could reduce Anthropic’s infrastructure costs by H1 2027, which may flow through to API rates.
The market is moving fast: AMD’s own forecast puts the AI accelerator segment growing from a slice of $365 billion today to $1.4 trillion by 2030, so infrastructure decisions made now will look very different in 18 months.
Frequently asked questions
When will AMD Helios AI servers start shipping?
AMD CEO Lisa Su said Helios servers are in full production and shipments are scheduled to begin at the end of Q3 2026.
What chips are inside the AMD Helios server?
The Helios rack contains AMD's MI455X AI accelerator and the new Venice CPU. Both chips are manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC).
Is OpenAI switching from Nvidia to AMD?
OpenAI has not announced a full switch. Its VP of compute strategy said OpenAI plans to deploy Helios racks at scale from late 2026 through 2027 and will also use AMD's next-generation MI500 chips. The two companies signed a multiyear deal in October giving OpenAI the option to buy roughly 10% of AMD.
What is the AMD and Anthropic chip deal?
AMD announced plans to supply up to two gigawatts of MI450 AI chips to Anthropic starting in the first half of 2027. The deal includes an AMD investment of up to $5 billion in Anthropic.
