At its Advancing AI 2026 event in San Francisco, Advanced Micro Devices NASDAQ: AMD laid out an expansive AI roadmap spanning data center racks, server CPUs, enterprise accelerators, developer software, personal AI systems and robotics platforms, while highlighting partnerships with major AI and enterprise customers.
AMD said AI demand is accelerating as usage shifts from training models to running them. The company said more than 35 quadrillion tokens are consumed each month, up nearly 160 times over two years, and estimated that roughly 60% of global AI compute capacity in 2026 will be used for inference. AMD also said agentic AI is driving a “step change” in compute demand because agents reason through multiple steps, call tools, access data and coordinate workloads.
The company updated its market outlook, saying it now expects the AI accelerator market to reach about $1.4 trillion by 2030. AMD also said it expects the server CPU market to grow from about $25 billion today to more than $200 billion by 2030, citing agentic AI as a new growth driver for CPUs.
AMD Launches Helios AI Rack
AMD announced Helios, which it described as the “industry’s highest performance AI rack,” built around Instinct GPUs, EPYC CPUs and Pensando networking. The company said the rack includes MI455 accelerators, Venice EPYC processors, Pensando DPUs and Vulcano AI NICs using the open Ultra Ethernet standard.
AMD said Helios is in full production, with shipments on track to begin at the end of the third quarter and ramp in the fourth quarter. The company claimed Helios delivers 15% more compute, 50% more HBM4 memory capacity and bandwidth, and 50% more scale-out bandwidth than competing systems. It also said the rack delivers up to 30% more tokens per dollar than the competition.
Anthropic Co-founder and Chief Compute Officer Tom Brown joined the event and said Anthropic will deploy up to 2 gigawatts of Helios. Brown said Anthropic’s evaluation of MI355X was faster than expected, saying one engineer connected the system to Claude and had it producing performance data over a weekend. Brown also said Claude is increasingly useful in software engineering and adjacent technical workloads such as design and layout.
OpenAI Head of Infrastructure Sachin Katti said OpenAI expects to deploy Helios at massive scale starting toward the end of the year and accelerating through 2027. Katti said OpenAI has been working side by side with AMD engineers to optimize software and run GPT-class workloads on Helios. He also said future AI infrastructure must be treated as a data-center-scale systems problem involving CPUs, GPUs, memory, networking, storage, power distribution and cooling.
Venice EPYC CPUs Target Agentic AI
AMD also detailed Venice, its next-generation EPYC server CPU family built on Zen 6. The company said Venice is designed for the agentic era and delivers up to 1.8 times more performance than Turin, with support for up to 512 threads per socket. AMD said Venice is in full production, with major server OEMs and cloud providers expected to begin rolling out systems in the fourth quarter.
The company described several Venice variants, including Venice HF for AI host nodes, a 256-core EPYC Venice chip for agent sandboxes and a 128-core version for enterprise and general-purpose servers. AMD said it also plans Verano for AI host nodes and Venice-X for high-performance and technical computing workloads.
Meta Head of Infrastructure Santosh Janardhan said demand across inference, training, recommendation systems and content creation is growing rapidly. He said Meta is moving from optimizing individual servers to treating the data center as one integrated system involving servers, networking, cooling and power. Janardhan said Meta has worked with AMD across multiple EPYC generations and is also collaborating on accelerator deployments, including MI450.
ROCm AI and Enterprise Products
Vamsi Boppana, AMD senior vice president of AI, introduced ROCm AI, an agentic AI platform intended to help developers build and optimize workloads for AMD GPUs through coding agents such as Cursor, Claude and Codex. Boppana said ROCm releases now go out every six weeks, compared with every four months previously, and said AMD has expanded support across open-source AI ecosystems including Hugging Face, PyTorch, JAX, vLLM and SGLang.
Boppana said ROCm AI includes HyperLoom, an AI-assisted optimization layer that can analyze workloads, tune configurations and iterate toward performance targets. In one example, he said ROCm AI produced a 38% tokens-per-second improvement for MiniMax M3 with vLLM on MI355s. Philippe Tillet, a researcher at OpenAI and creator of Triton, said OpenAI and AMD are collaborating across the software stack, including LLVM code generation, and said AMD’s open software approach has enabled performance gains that would be harder to achieve in a closed stack.
For enterprise AI, Dan McNamara, AMD senior vice president and general manager of compute and enterprise AI, announced the Instinct MI350P, an air-cooled GPU designed to fit within existing enterprise server power and cooling envelopes. AMD said a single MI350P can support up to 260 billion parameters and delivers more than four times the tokens per second per dollar than the competition.
AT&T Chief Technology Officer Jeremy Legg said AT&T is consuming more than 1 trillion tokens per month and has more than 100 GenAI models in production. He said AT&T has used AMD technology to train and post-train models, manage token costs and support open-source telecom AI models. Legg also announced OTel 2.0, an updated Open Telco AI model trained on AMD and made available through open source.
Personal AI, Robotics and Roadmap
Jack Huynh, AMD senior vice president and general manager of the computing and graphics group, outlined AMD’s personal AI strategy, including Ryzen AI Halo and a new Gorgon Halo system with 192 gigabytes of unified memory and support for models up to 300 billion parameters. Huynh said AMD is expanding its partnership with Hugging Face and that later this year every Ryzen AI Halo box will include a full year of Hugging Face Pro.
Cisco President and Chief Product Officer Jeetu Patel said enterprises will need management, security, observability and cost controls as inference spreads beyond data centers to desk-side systems. Patel said Cisco’s management apparatus for AMD Halo devices is in early availability for select customers and is expected to reach general availability in the U.S. in early fall.
AMD also introduced the Kria AI System-on-Module, powered by Ryzen AI Embedded X100, and a Kria AI robotics developer platform built on ROCm and ROS2. Huynh said the platform brings CPU, GPU, NPU and unified memory together for robotics workloads involving perception, reasoning and real-time control.
Looking ahead, AMD said it plans Florence EPYC CPUs with Zen 7 cores in 2028 and Ravenna with Zen 8 for 2030. On GPUs, AMD said MI500 will bring next-generation HBM, a larger scale-up domain and new copper and optical interconnects, while MI600 is in development for 2028. The company said customers should expect a new Helios system every year.
About Advanced Micro Devices (NASDAQ:AMD)
Advanced Micro Devices, Inc NASDAQ: AMD is a global semiconductor company that designs and sells microprocessors, graphics processors, chipsets and adaptive computing solutions for a broad set of markets. The company's product portfolio includes consumer and commercial CPUs under the Ryzen and Threadripper brands, data center processors under the EPYC brand, and Radeon graphics processing units for gaming and professional visualization. AMD also offers semi-custom system-on-chip (SoC) products for gaming consoles and other specialized applications, and provides supporting software and platform technologies for OEMs, cloud service providers and end users.
Founded in 1969, AMD has evolved from a supplier of logic chips into a diversified, fabless semiconductor designer.
This instant news alert was generated by narrative science technology and financial data from MarketBeat in order to provide readers with the fastest reporting and unbiased coverage. Please send any questions or comments about this story to contact@marketbeat.com.
Before you consider Advanced Micro Devices, you'll want to hear this.
MarketBeat keeps track of Wall Street's top-rated and best performing research analysts and the stocks they recommend to their clients on a daily basis. MarketBeat has identified the five stocks that top analysts are quietly whispering to their clients to buy now before the broader market catches on... and Advanced Micro Devices wasn't on the list.
While Advanced Micro Devices currently has a Moderate Buy rating among analysts, top-rated analysts believe these five stocks are better buys.
View The Five Stocks Here
MarketBeat just released its list of the 7 hottest IPOs expected to hit Wall Street in 2026. See which companies are preparing to go public and why investors are watching closely.
Get This Free Report
