Now Playing
AMD's Google Win, Intel's Parity Bet & the Inference CPU Shift
Audio is available on Spreaker — see link below.
AMD lands a landmark Google Cloud EPYC deal as Intel bets on CPUs closing the gap with GPUs in AI inference workloads. CoreWeave's $132B backlog and Nvidia leadership fallout round out a pivotal week for semiconductor infrastructure.
Audio is available on Spreaker — see link below.
Intel's CEO is making a bold call: the ratio of GPUs to CPUs in AI infrastructure is shifting from eight-to-one toward parity. If he's right, it's one of the more significant demand shifts in the server chip market in years.
Lip-Bu Tan's argument runs like this: training workloads are GPU-heavy, but inference is different. Serving an AI agent in real time, handling live requests at scale, is a task where CPUs can compete on efficiency and cost.
AMD isn't waiting on theory. This week, AMD's fifth-generation EPYC processors were embedded into Google Cloud's TPU infrastructure, a deal concrete enough to move the stock.
The demand context behind all of this: AI cloud provider CoreWeave posted three-point-six-six billion dollars in quarterly revenue, up one-hundred-sixty-eight percent year on year, with full-year guidance raised to thirteen-point-one to thirteen-point-three billion and a backlog of one-hundred-thirty-two-point-five billion dollars. That backlog is the key figure.
Two forces are complicating that picture from the outside. China is signaling rare earth export retaliation if the US tightens chip controls further.
The signal to track from here is whether Intel can point to actual signed inference deployments, not forecasted ratios. AMD's execution on Azure and Oracle will be an early proof point.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.