Advancing AI 2026 keynote
Lisa Su · Jul 23, 2026 · 2h 14m
Summary
AMD's 2026 keynote was about Helios, its rack-scale system built on the MI455 GPU, Venice EPYC CPUs and Pensando networking, now in full production with Anthropic and OpenAI on stage as partners. Su also showed a 256-core Venice CPU aimed at agent workloads and an AI-driven ROCm aimed at removing software friction.
Key themes: AI infrastructure, Agents · AI-written summary from the published transcript
- 1Helios combines the MI455 GPU, Venice EPYC CPUs and Pensando DPUs in one rack-scale system16:10
- 2Helios is in full production, with shipments starting at the end of the third quarter and ramping in the fourth20:51
- 3Anthropic joined as AMD's newest Instinct partner21:49
- 4OpenAI is one of the deepest and earliest partners deploying Helios30:46
- 5A 256-core Venice EPYC CPU is built for agent workloads, "more agents per rack"44:07
- 6AMD is bringing AI-driven development to ROCm to automate optimization and bring-up1:07:38
Announced
1 available the same day| Announcement | Status |
|---|---|
| Helios rack-scale systemIn full production, shipments from end of Q3 | GA |
| Anthropic Instinct partnershipAnthropic adopts AMD Instinct | Announced |
| Venice 256-core EPYCCPU for agent workloads | Announced |
| AI-driven ROCmAutomated optimization and workload bring-up | Announced |
Related keynotes
| Date | Keynote | |||
|---|---|---|---|---|
| Jun 12, 2025 | AI infrastructure, Open source | Demand for inference has grown significantly, driven by more capable models and new uses | 3 | |
| Oct 10, 2024 | AI infrastructure | MI300, launched less than a year before, was deployed by Microsoft and Meta for major workloads | 2 |
About this keynote
Sources: official keynote videos, their captions and the hosts' recaps, summarized by AI; watch the source for exact wording. Logos via logo.dev; trademarks belong to their owners.