AMD warns CPU supply tightness

Lisa Su said on May 22 that AI inference demand and the rise of agentic AI are tightening the server CPU market, adding a new supply constraint to a hardware stack better known for GPU shortages. The AMD chief executive said in Taipei that customer demand for CPUs tied to AI systems has exceeded the company’s earlier expectations, prompting AMD to work with partners in Taiwan to raise capacity. AMD this week also said it would invest more than $10 billion across Taiwan’s semiconductor ecosystem and that its next-generation EPYC “Venice” processor is already in production on TSMC’s 2-nanometer process. ### Why are CPUs suddenly part of the AI bottleneck? Lisa Su said Friday that inference demand is no longer falling only on accelerators. She said CPUs are seeing stronger demand because AI systems still need host compute, orchestration and general-purpose processing around model serving, and because agentic AI increases the amount of software and system work wrapped around each model call. AMD said earlier this month that it now expects the server CPU market to grow more than 35% annually through 2030, roughly double its prior long-term view. CNBC reported after AMD’s earnings that Su had tied that higher forecast to stronger CPU demand from AI workloads, especially as inference expands beyond training clusters. ### What exactly did AMD say it is doing in Taiwan? AMD said on May 21 that it would invest more than $10 billion across Taiwan’s ecosystem to expand strategic partnerships and scale advanced packaging for AI infrastructure. The company said the spending is aimed at manufacturing and assembly capacity for next-generation systems, including packaging technology for 6th-generation EPYC processors codenamed “Venice.” The company’s statement named advanced packaging as a focus area and said its EFB-based 2.5D approach is intended to raise interconnect bandwidth and improve power efficiency. Reuters reported that Su, speaking in Taipei, said AMD was working with Taiwan partners to ramp production as stronger-than-expected demand squeezed the global CPU market. ### Where does “Venice” fit into this? AMD said this week that “Venice,” its 6th-generation EPYC processor, is in production on TSMC’s 2-nanometer technology. The company linked that chip to its broader AI infrastructure push and said the processor will be part of its Helios rack-scale platform alongside Instinct MI450X GPUs. Taipei Times reported that AMD plans to produce “Venice” in Taiwan and later at TSMC’s Arizona fab. AMD’s own announcement said Helios deployments are on track to begin in the second half of 2026. ### Why would startups care about a CPU squeeze if GPUs still dominate headlines? AI startups that plan on-premises inference clusters or custom ML hardware purchases still need server CPUs, memory, networking and packaging capacity alongside accelerators. A tighter CPU market can affect delivery schedules for full systems even when the headline constraint appears to be GPUs. Reuters reported that AMD is asking partners to increase production because demand has been stronger than expected. That matters for buyers planning hardware procurement cycles, especially if they are relying on bundled server platforms rather than buying only standalone accelerators. ### Is AMD saying CPUs will matter as much as GPUs? Lisa Su has argued in recent interviews that AI infrastructure is becoming more balanced between CPUs and GPUs as inference scales. That is a forecast, not a company disclosure on current mix, but AMD’s updated market outlook and Taiwan capacity push show the company is positioning for higher CPU demand inside AI systems. AMD’s next concrete milestones are already set out. The company said Helios systems with “Venice” and MI450X are due for multi-gigawatt deployments beginning in the second half of 2026, and Su said on May 22 that AMD is working with Taiwan manufacturing partners now to expand supply.