AMD CEO Lisa Su expressed strong confidence in the CPU market outlook and characterized 2026 as the "year of inference." She forecasts that the CPU market will experience explosive growth over the next several years, driven by agentic AI workloads.

During a keynote speech at the OCP Global Summit held on October 12, Su revealed that Advanced Micro Devices (AMD) has raised its CPU market forecast three times over the past nine months. The company now expects the market to grow at a compound annual growth rate of approximately 50% through 2030, reaching a total size exceeding $200 billion. By comparison, the overall CPU market was worth roughly $20 billion a year ago, when consensus expectations pointed to only single-digit annual growth.

At the same time, Su announced that AMD's next-generation AI accelerator product Helios began volume shipments in September, with the first systems deployed to leading customers including OpenAI, Meta, Microsoft, Oracle, and Anthropic. AMD also unveiled the Helion rack system for agentic AI computing at the summit, with plans to launch it through ecosystem partners starting November 26.

CPU Market Forecast Sharply Revised Upward, Agentic AI Emerges as Core Driver

Speaking at the summit, Su said AMD expects the data center AI accelerator market to approach $1.5 trillion by 2030, but she pointed to even greater growth potential in the CPU business.

"We have raised this market forecast three times in the past nine months because we are seeing such strong growth in overall CPU demand. We now believe the compound annual growth rate over the next four to five years will be approximately 50%, and the market will exceed $200 billion by 2030," Su said.

The core logic behind this forecast lies in the rise of agentic AI workloads. Su noted that the continued evolution of AI applications from training to inference and then to agentic tasks is generating a new wave of demand for CPU compute power, a trend that has surpassed previous market expectations.

2026: Inference Workloads to Dominate the Market

Su characterized 2026 as "the true year of inference" and projected that inference workloads will capture approximately 60% of the market in the future.

"This year, we believe inference now accounts for about 60% of the market. And it's not just inference — what's really exploding is agentic workloads," she said.

This assessment carries significant implications for product strategy across the semiconductor industry — as AI deployment shifts from model training to large-scale inference and agentic tasks, the competitive focus of chipmakers and the structure of customer demand are undergoing a structural transformation.

Helios Enters Mass Production, Leading Cloud Providers Deploy First

Su disclosed that AMD began volume shipments of its Helios series products in September, with OpenAI, Meta, Microsoft, Oracle, and Anthropic among the first customers to install them.

"Our volume production will continue to ramp through the fourth quarter and into the first half of next year," Su said.

The smooth mass production of Helios marks a milestone for AMD in the high-performance AI acceleration space. Its deep ties with leading technology and AI companies further solidify its competitive position in the data center market.

Helion Rack Unveiled, Betting on the Agentic AI Computing Race

Following Helios, AMD further unveiled the Helion Agentic AI rack system at the summit. Su described Helion as a "close relative of Helios," positioned specifically to serve agentic AI computing needs.

According to Su, Helion features a standard 21-inch rack design supporting both liquid cooling and air cooling, and packs over 32,000 Zen 6 cores — 40% more than competing racks — delivering the industry's highest compute density. Additionally, the rack can accommodate up to 32 Instinct MI350 GPUs and integrates high-capacity networking capabilities, all built on open compute standards (OCP specifications).

Su said market response to the Helion rack has been "very positive," with ecosystem partners set to make the product available starting November 26. The related design contributions will also be shared with the OCP open standards community.