Guest Post: Rebalancing
By Matt Swayne · · 2 min read
Guest Post by Ian Foley
Have we become so accustomed to Nvidia dominating AI, with stories of infrastructure buildout, that we’re overlooking what consumers are clamouring for?
Nvidia’s GPUs earned their prominence by handling the vast parallel calculations involved in training and running AI models. As assistants move from answering questions to carrying out instructions, however, they also create more work for CPUs, the general-purpose processors running applications and coordinating activity between model calls.
Meta’s Muse reached an estimated 10M US downloads in just 40 days. Meta’s substantial advertising helped, but the figures suggest consumer interest in assistants that can take on everyday tasks. For example, rearranging a holiday to Mexico after a cancelled flight, requires software to retrieve reservations and check alternatives against your calendar before processing changes through booking systems. Much of that surrounding work runs on CPUs, alongside the model’s reasoning.
That demand is beginning to reshape hardware requirements. TrendForce expects planned server configurations to move from roughly one CPU per eight GPUs towards one per two, potentially reaching parity. It forecasts server CPU shipments growing more than 40% in 2027. Against that backdrop, the semiconductor selloff was perhaps overdone, particularly for AMD and Intel, the leading CPU suppliers. As Intel’s finance chief David Zinsner observed “the CPU has become cool again this year.”
About Ian Foley
Ian Foley is an entrepreneur, investor, and operator with over 20 years of experience across Silicon Valley and global tech markets. His work sits at the intersection of AI, fintech, and infrastructure, where he focuses on building and scaling companies from early stage through exit.
IanToons — Tech Cartoons by Ion Foley
