The current hypergrowth in GPUs and other accelerator chips for AI in the cloud and the data center will eventually slow down but not until it has transformed the industry, finds the latest edition of Omdia’s AI Processors for Cloud and the Data Center Forecast. The market has grown from just under $10bn in 2022 to $78bn this year, and Omdia expects it will eventually reach $151bn in 2029. However, there is likely to be a distinct inflection point in 2026, with a shift from technology adoption to change in demand for AI applications as the driver of growth.
Omdia’s Principal Analyst for Advanced Computing, Alexander Harrowell said: “The most significant change is the upward revision of the underlying data center semiconductor market driven by growth in AI. Beyond that, we’re noticing hyperscalers’ custom chips, especially Google’s TPUs, starting to capture market share from GPUs. We expect them to be increasingly important.”
“The size of the biggest AI models has not advanced since January 2021, and the focus of innovation has shifted to a much greater variety of smaller ones. There are also signs that models of similar size are becoming more efficient generation on generation. This is a headwind for investment in AI infrastructure, although it also permits more organizations to get involved, which tends to make up in breadth what is lost in depth. Over time, inference is going to become much more important. We expect a growing user base and fleet of applications to require more inference infrastructure."
