Ben Pouladian
Founder of BEP Research, AI Compute Economics
AMD risks missing large advance rack orders because Helios is still unproven at volume manufacturing. It looked credible on paper at July's Advancing AI, but this is AMD's first rack-scale system; meaningful volume is expected only in 2H27, two quarters past AMD's window. Hyperscalers order racks a year or two before seeing them, and custom inference chips are squeezing AMD from the other side. The photographed Helios rack's cabling underscores how hard it is to build. x.com x.comÂ
Cerebras looks attractive because the short report's two-year wafer-life thesis rests on unvalidated reliability assumptions, not representative failure data. Energy per useful token, lower warranty utilization and ordinary repair accounting also undercut its financial conclusions. OpenAI has integrated Cerebras into production; AMD and AWS have announced integrations, while Oracle plans 2027 availability for ultra-fast inference, adding commercial validation despite the report. x.com x.comÂ
AMD's accelerator partnerships are a weak signal of true demand because it gives customers huge chunks of ownership to secure purchases. Other examples look toothless for now. Single sourcing is a real problem, and AMD likely keeps its data-center accelerator program long term. But a distant second place is not protected: relative competitiveness matters, and an upstart could take share, though there is no evidence of that today. x.com x.comÂ
Many neoclouds favor AMD, making it potentially the semiconductor to play. Helios still needs to mature into a complete rack-scale platform for AMD to become a true second AI platform, rather than just the No. 2 GPU vendor. x.comÂ
AMD's strong Q2 set up a classic sell-the-news reaction against a stock that had priced in perfection. Capex hit $808M, nearly 3x what many modeled, compressing free cash flow. The roughly $13B Q3 guide beat the $12.5B consensus but not whispers near $14B, with growth slowing to about 41%. Management guided no near-term margin expansion, server CPU supply is constrained, and Helios and the multi-gigawatt Anthropic, OpenAI and Meta deals are mainly a 2027 story. x.comÂ
Rackspace is AMD's key neocloud partner, building governed enterprise AI cloud infrastructure at scale on AMD Instinct GPUs and EPYC CPUs, so AMD's accelerating AI momentum flows directly to it. Per a SemiAnalysis (@SemiAnalysis_) thread, an AMD senior director's public GitHub lists Anthropic as a high-priority customer, still in evaluation. AMD's Advancing AI event this week could bring big announcements and ecosystem updates. x.comÂ
Agentic AI demand should drive AMD's 2027 data-center growth well over 100% year on year. Server CPU revenue is forecast to double, with over 5 million Venice units shipping in 2027; Helios racks should ramp mostly in Q1 2027, with over 20,000 expected that year, while MI450/455X volume remains limited by end-2026. Projected 2027 sales are $103bn, with 2027 EPS revised up 15% and a new $720 target at 40x 2027 earnings. x.comÂ
Inference's move toward edge devices and dense, memory-hungry racks should drive demand for specialized chips, links, power delivery and cooling. McKinsey projects inference at 60% of AI computing demand by 2030. Ambarella makes on-device AI chips; Astera Labs links GPUs, memory and networks; Credo supplies short copper connections; Rambus earns from memory interfaces; Navitas reduces power waste; Modine cools dense racks; and Super Micro builds inference-specific server racks at volume. x.com … Show more
AI's physical buildout looks set to last to 2030 and beyond because financing, memory supply and hyperscaler spending still point to sustained demand. Nebius has $3.5B in financing, IREN $2.4B at 9% and SPCX $40B for GPU equipment. SK hynix expects shortages through at least 2030 and Samsung says HBM4 is sold out; memory companies are making money with no sign of slowing. Google expects 2027 capex to rise significantly. Pullbacks can happen; be cautious and buy quality. x.comÂ
AMD's Helion rack design could benefit from agentic-AI demand beyond GPUs as more advanced agents require more CPU compute. Built for agentic AI, it can support up to 32,768 EPYC CPU cores in a rack, with configurations that also integrate Instinct GPUs. Helion configurations are planned for November, making this a very bullish development for AMD's long-term data center opportunity. x.comÂ
Insufficient substrate supply may cap AMD's growth because the company has not secured enough capacity. The author says limited substrate availability drove AMD's lowered guidance, making supply-chain execution the constraint on how far it can grow. This adds an execution risk to the earlier concern that AMD is too expensive to buy despite strong results. x.com x.comÂ
