AMD Bets on Rack-Scale AI Systems and Open Software to Rival Nvidia

Quick Facts

• AMD reported record 2025 revenue of $34.6 billion, up 34% year-over-year, with data center revenue reaching $16.6 billion.
• Q1 2026 revenue hit $10.3 billion, up 38%, with Q2 2026 guidance set at $11.2 billion, representing 46% year-over-year growth.
• CEO Lisa Su projects data center GPU revenue will reach $15 billion in 2026, a 114% increase year-over-year.

AMD is rewriting its growth strategy. The company that clawed back share from Intel in the CPU market is now building a full-stack AI infrastructure platform, and its financial results suggest the bet is paying off early.

For full-year 2025, AMD posted record revenue of $34.6 billion, with gross margin at 50%, operating income up 94% to $3.7 billion, and net income rising 164% to $4.3 billion. Data center revenue hit $16.6 billion for the year, up 32%, driven by EPYC CPU and Instinct GPU demand.

The momentum carried into 2026. Q1 revenue reached $10.3 billion, up 38% from the prior year. Data center revenue for Q1 alone was $5.8 billion, a 57% year-over-year jump. Gross margin expanded to 53%, up from 50% a year earlier.

The Helios Platform

AMD’s new strategy centers on rack-scale systems rather than discrete chips. The company acquired ZT Systems and used that expertise to develop AMD Helios, a rack-scale AI infrastructure platform that integrates Instinct MI455X GPUs, EPYC Venice CPUs, and Pensando networking into a unified system.

Helios is built on Meta’s Open Wide Rack specification, signaling AMD’s push toward open standards. The platform targets large-scale AI training and inference deployments. AMD expects Helios to launch in 2026.

The shift reflects a broader view inside AMD about where competition in AI infrastructure is headed. Dave Vellante of SiliconAngle wrote that AMD’s objective is not to beat Nvidia outright but to carve a durable position in an expanding market where Nvidia remains the dominant supplier.

Software as a Strategic Gap

AMD SVP Salil Raje acknowledged the company’s historical weakness directly: AMD was not traditionally a software company. The company has since turned its attention to ROCm, its open-source GPU software platform, and is actively courting developer communities that have built workflows around Nvidia’s CUDA.

The open-source positioning is deliberate. By lowering the barrier for developers to build on AMD hardware, the company aims to create an alternative software ecosystem without requiring a wholesale abandonment of existing tools.

SVP of AI Vamsi Boppana noted that inference token demand has grown 20 times year over year, and that workload profiles are shifting from single-turn inference toward long-context, multi-step agentic pipelines. That shift creates new performance requirements that AMD is positioning Helios to address.

CPU Market Confidence

AMD also doubled its 2030 server CPU market forecast to $120 billion, a sign that the company sees sustained demand growth beyond the current AI spending cycle. Fifth-generation EPYC processors were cited as a primary driver of Q1 2026 data center gains alongside Instinct GPU shipments.

CEO Lisa Su described 2025 as a defining year for the company. At CES 2026, she introduced the term yotta-scale computing to describe the next phase of AI infrastructure buildout, framing AMD as a foundational supplier for that transition.

Forrest Norrod, AMD’s EVP and GM of Data Center Solutions, called a recent OpenAI deal transformative for the industry’s competitive structure. The company has not disclosed full terms of that arrangement.

Q2 2026 guidance of $11.2 billion, if realized, would mark six consecutive quarters of accelerating data center growth. AMD’s ability to sustain that pace depends heavily on Helios adoption, ROCm developer uptake, and continued EPYC share gains in enterprise server deployments.

Read more: AMD’s next reinvention: A new playbook for the AI era