AMD ships first Helios racks with 72 MI455X GPUs to Meta, Microsoft in late 2026
The letter grade, factuality score, political-lean rating, and social-media sentiment for this report unlock with a free CladFacts account — no card, no trial clock. Already have one? Sign in. The full report below is free to read.
Disagree with this grade or political lean?
Flagging is open to every reader with a free account. Sign in or create one to dispute this report.
Topics in this report
Summary
CNBC visits AMD facilities in Texas for an exclusive look at the Helios rack-scale AI system, its 72-GPU configuration, liquid cooling, and integration of GPUs, CPUs, and networking. Segments cover customer commitments from Meta, Microsoft, OpenAI, Oracle, and TCS, plus AMD's data center growth and challenges in supply chain and software.
Editorial Assessment
The segment accurately reports verified hardware specs, revenue numbers, and announced deployments while highlighting AMD's open-standards pitch against Nvidia's ecosystem. Viewers may miss independent benchmarks comparing Helios to Vera Rubin at scale or details on ROCm adoption hurdles. Framing presents Helios as a credible first-gen challenger without overstating near-term market share gains. Supply constraints at TSMC and power/water demands receive appropriate attention.
Key Moments
Helios rack has 72 MI455X GPUs and 18 EPYC CPUs
Confirmed in AMD product specs and multiple partner announcements
Meta to deploy one gigawatt of AMD GPUs on Helios starting 2026
Aligned with Meta's earlier announcements and AMD updates
AMD Q1 2026 data center revenue $5.8B, up 57% YoY and now majority of total revenue
Matches AMD's official earnings release
Helios priced $5-5.5M vs Nvidia Vera Rubin $3.5-4M
Futurum estimate cited; other analyst ranges show Helios potentially competitive or lower
Nvidia holds over 95% data center GPU market share
Consistent with industry reports; AMD targets 20-25% long-term
Notable Concerns
- Limited independent performance data on Helios vs Nvidia systems
- Pricing estimates vary across analysts