OpenAI has chosen AMD’s EPYC server processors to run alongside its first in-house AI chips, Lisa Su said on AMD’s third-quarter earnings call on 3 November: “As one example, OpenAI selected Turin EPYC processors as the host CPUs for its first custom accelerator platforms”.
Su did not name the accelerator or say how many processors are involved. She presented the win as part of a wider CPU push into AI systems and agents, saying AMD’s hardware partners have committed more than 240 platforms to its next CPU generation, Venice, 20% more than Turin. By our arithmetic, that puts Turin at about 200 platforms.
Business Pill · Distribution Advantage
An evergreen Business Engineer explainer on distribution advantage. It teaches the general idea only and does not discuss AMD or OpenAI.
The key insight: By our arithmetic from Su’s numbers, partners committed about 200 Turin platforms and more than 240 for Venice; OpenAI picked Turin to host its first custom accelerators, putting AMD CPUs beside chips that compete with AMD GPUs.
About 22:06 UTC, in her prepared remarks, Su said: “EPYC’s leadershipperformance and mature ecosystem, proven at scale, have made it the host CPU of choice for custom accelerators.” The OpenAI line followed. Just before it, she said: “In head nodes, we won major deployments with several of the largest model developers and cloud providers in the quarter, which will ramp over the coming quarters.”
A host CPU, or head node, is the server processor that sits next to the accelerators in a system and feeds them work. OpenAI unveiled its first custom AI chip, Jalapeño, built with Broadcom, in June; we covered the chip and its self-reported benchmarks. Su did not name Broadcom or the chip. OpenAI is also an AMD GPU customer under the 2025 Instinct agreement, and Su named it among the first labs running Helios racks.
EPYC server platforms committed by AMD’s OEM and ODM partners: more than 240 for Venice, which Lisa Su said is 20% more than Turin, on the 3 November earnings call; the Turin figure of about 200 is our arithmetic.
What Was Posted on X
$AMD CEO Lisa Su on EPYC Venice demand:
โWe began revenue shipments of EPYC Venice processors in the third quarter and the first OEM systems are on track for availability this quarter.โ
โEarly demand is the strongest we have seen for any new EPYC generation.โ
Su added that OEM and ODM partners have committed to more than 240 Venice platforms, 20% more than Turin, with cloud providers beginning deployments this quarter ahead of instance rollouts in early 2027.
OpenAI selected EPYC Turin as the host CPU for its first custom AI accelerator platforms.
Wall St Engine posted Lisa Su’s call remarks on EPYC Venice demand at 22:16 UTC, including the line that OpenAI selected EPYC Turin as the host CPU for its first custom AI accelerator platforms. The quotations in this piece come from our own transcript of the call audio.
Agents as the CPU Story
Su framed the CPU business around agents. She said cloud providers added EPYC capacity “for personal and enterprise agents, which are quickly becoming the next killer app for CPUs in the cloud”, and that “Meta’s Muse and OpenAI’s DOTS illustrate how quickly personal agents are growing, with more than 10 million people activating Muse agents in the first six weeks”. That Muse figure is Su’s; Meta did not give it on the call.
On Venice, she said: “We began revenue shipments of EPYC Venice processors in the third quarter, and the first OEM systems are on track for availability this quarter.” She added: “Early demand is the strongest we have seen for any new EPYC generation.” Asked in the Q&A about 2027, she said “For the data center CPU business, we will about double”.
The Arithmetic
More than 240 Venice platforms at 20% more than Turin implies about 200 Turin platforms (240 divided by 1.2), by our arithmetic. Su also said a single all-CPU Helion rack holds up to 144 Venice processors with nearly 74,000 threads, which is about 512 threads per processor (73,728 divided by 144), by our arithmetic.
How Each Side Reads These Figures
For readers who see a broader AMD footprint at OpenAI, Su placed AMD in two parts of OpenAI’s systems: Instinct GPUs in Helios racks, and EPYC host CPUs next to OpenAI’s own accelerators.
For readers who see limits, the host CPU win comes alongside an accelerator that OpenAI is building itself, and Su gave no unit, revenue or timing figure for it. We cite no analysts’ views here.
As we read it, the host CPU role lets AMD sell into OpenAI systems whichever accelerator OpenAI uses, its own chips or AMD Instinct.
As we read it, Su used agents to explain CPU demand: the Muse figure she cited and the Helion all-CPU rack both point at CPU-heavy agent workloads.
As we read it, the gap in the disclosure is size: Su gave no units, revenue or dates for the OpenAI host CPU win.
Lisa Su, AMD earnings call, 3 November 2026 (our transcript)
“As one example, OpenAI selected Turin EPYC processors as the host CPUs for its first custom accelerator platforms”
Three Implications
FOR OPENAI WATCHERS Su named OpenAI both as a Helios GPU customer and as the buyer of Turin host CPUs for its own accelerators.
FOR AMD SHAREHOLDERS Su said the data center CPU business will about double in 2027 and that early Venice demand is the strongest of any EPYC generation.
FOR BROADCOM AND NVIDIA WATCHERS Su called EPYC the host CPU of choice for custom accelerators and said AMD won major head-node deployments with several of the largest model developers.
The framework puts it this way: “That is no longer a person using a tool; it’s a human managing a team of agents.”
As we read it, Su described the CPU side of that shift: agents that, in her words, work continuously and in parallel, adding to accelerator demand and to CPU demand. The framework was published in July 2026 and does not discuss AMD.
Quotations come from our own machine transcript of the call audio, as rebroadcast live; it is not an official AMD transcript and times are approximate. Su did not name OpenAI’s accelerator, its supplier, volumes or dates; the Jalapeño and Broadcom details come from our earlier coverage, not from the call. The Muse activation figure is Su’s statement, not a Meta disclosure.
The 200 and 512 figures are our arithmetic from Su’s numbers. We did not contact AMD or OpenAI.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Su said OpenAI picked AMD’s Turin EPYC processors as host CPUs for its first custom accelerator platforms, adding a CPU role at OpenAI to AMD’s Helios GPU deployments. She tied CPU demand to agents and said partners have committed more than 240 Venice platforms, against about 200 for Turin by our arithmetic.
A note on sourcing. Quotations from Lisa Su come from our own machine transcript (faster-whisper) of AMD’s third-quarter 2026 earnings call of 3 November, as rebroadcast live; it is not an official transcript and times are approximate. The call line was also posted on X by Wall St Engine. The Turin platform count and threads per processor are our arithmetic. We did not contact AMD.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
