AMD Ships Helios Rack to Scale AI Inference
The co-engineered system arrives as industry demand for inference-capable AI infrastructure continues to surge.
Updated on Oct. 22, 2026 in Data Centers
Live Poll
Do you believe the rapid expansion of artificial intelligence infrastructure will improve our society?
AMD has begun shipping its Helios rack-scale AI solution, a hardware system co-engineered with Meta. The rack, which weighs nearly 7,000 pounds, aims to address the shifting demands of AI infrastructure.
Why it matters
The shift toward agentic AI workflows and inference-heavy models is driving massive infrastructure growth. This collaborative hardware approach seeks to normalize rack-scale architectures to capture a projected $1.4 trillion AI accelerator market by 2030.
The Helios rack weighs nearly 7,000 pounds and integrates Meta-developed OCP power shelves with AMD open rack components. It arrives as the industry faces a 5x annual growth rate in training flops and a 95x increase in monthly tokens over the past two years.
The players
AMD
A semiconductor company that designs high-performance CPUs and AI accelerators for data centers and consumer PCs.
Meta
A technology conglomerate that develops large-scale AI models and contributes heavily to open-source hardware standards.
The details
The Helios system utilizes an open ecosystem design where AMD provides open rack wide components that interface directly with Meta's custom power distribution shelves. This co-engineered approach allows for modular scaling of AI compute density, which is critical as agentic traffic—automated systems that execute multi-step tasks—has surged 14x this year. The physical rack architecture is designed to manage the high power and thermal requirements of modern inference workloads, which now account for 60% of the total AI market.
Timeline
2020: Training flops growth trajectory began.
2026: Keynote address occurred in San Jose.
Q4 2026: Helios production ramp begins.
Q1 2027: Helios production ramp continues.
2030: Projected TAM milestones for CPUs and accelerators.
The Tech Race
The Helios system marks a milestone in the 15-year collaboration between AMD and the Open Compute Project. This release signals a strategic push to capture the projected $1.4 trillion AI accelerator market by aligning hardware production with open-source infrastructure standards.
Enterprises and cloud providers can begin integrating the Helios rack into their data centers starting in Q4 2026. This hardware transition specifically targets users managing high-volume inference tasks, which now represent the majority of the current AI workload.
The takeaway
The rapid 14x growth in agentic traffic indicates that inference capabilities are becoming the primary constraint for data center operators. Watch the production ramp-up through Q1 2027 to see if this open-ecosystem rack design achieves adoption among hyperscalers.
Further reading
For more on evolving infrastructure requirements, visit the Data Centers section.
Source note: This article includes information reported by DQ.
Live Poll
Do you believe the rapid expansion of artificial intelligence infrastructure will improve our society?
