Server CPUs forecast $120B by 2030

Server CPUs are suddenly back at the center of the AI story. Not because GPUs stopped mattering — they absolutely did not — but because the next wave(ir.amd.com)al, planning, and verification. On May 5, AMD used its first-quarter 2026 earnings call to say that changes the market enough to push the server CPU opportunity above $120 billion by 2030, more than double its prior view. (ir.amd.com) ### Why are CPUs back in the conversation? A GPU is gr(money.usnews.com)ontext, calling tools, checking outputs, moving data, and keeping the whole workflow stitched together. That control layer leans heavily on CPUs, which is why chip companies and analysts are now talking about AI infrastructure as a broader stack instead of a pure accelerator story. (money.usnews.com) ### What exactly did AMD change? AMD did not anno(ir.amd.com)le market can grow at more than 35% annually and exceed $120 billion by 2030. The notable part is the jump from the company’s previous framework, which had implied something closer to an 18% annual growth rate and a market around $60 billion by 2030. (msn.com) ### Why did that land with investors? Because the company paired the forecast with numbers that already look like the shift is showing up (money.usnews.com)n Q2 revenue. So this was not a vague “someday AI will help CPUs” pitch — it came attached to a business already growing fast in servers. (ir.amd.com) ### What does “agentic AI” change in a data center? Basically, cluster design. Older AI buildouts often got described as a few CPUs feeding a lot of GPUs. But ana(msn.com)t as a shift from roughly 1 CPU for 4 GPUs toward 1-for-2 or even 1-for-1 over time. That matters because a richer CPU mix lifts both unit demand and the average selling price of the CPUs that do get deployed. (finance.yahoo.com) ### Why does memory keep showing up here too? Because agents are memory-hungry in the plain-English sense and in the(ir.amd.com)nley argued that agentic AI could add $32.5 billion to $60 billion of incremental data-center CPU demand by 2030, while also driving 15 to 45 exabytes of additional DRAM demand. That is why memory names keep getting pulled into what used to look like a GPU-only trade. (money.usnews.com) ### Does this mean GPUs matter less? Not really. Th(finance.yahoo.com) the surrounding compute needed to make AI systems useful in production looks bigger than many people modeled a year ago. Turns out the hard part is not just generating tokens — it is coordinating the whole machine around them. (wccftech.com) ### Who benefits if this view is right? AMD obviously wants investors to think EPYC benefits. But the spillover is br(money.usnews.com) more orchestration and more DRAM. The winners are the companies selling the plumbing, not just the engine. (money.usnews.com) ### Bottom line The new claim is simple: AI is no longer just a GPU scaling story. If agentic workloads become real produ(wccftech.com)n of travel is the important part — AI spend is widening. (ir.amd.com)