New Omdia forecast: AI data center chip market to hit $286bn, growth likely peaking as custom ASICs gain ground

LONDON, August 28, 2025: Omdia’s latest AI Processors for Cloud and the Data Center Forecast show the AI data center chip market continuing its rapid rise, but signs indicate growth is starting to slow. 

The boom in AI infrastructure drove a substantial upward revision of the forecast: $123bn in GPUs and AI accelerators shipped in 2024, $207bn is expected in 2025, and $286bn by 2030. While the market grew over 250% year-on-year between 2022 to 2024, growth from 2024 to 2025 is projected at around 67%. AI infrastructure spending is expected to peak as a share of the data center in 2026, when essentially all incremental spending will be driven by AI, before gradually tapering toward 2030.

“The key drivers of growth are the adoption of AI applications, the increasing democratization of AI fine-tuning, and the use of reasoning models, which generate large numbers of tokens that are not used in the final output,” says Alexander Harrowell, Principal Analyst for Advanced Computing. “On the other hand, the shift towards smaller, specialized models is slowing  demand for AI computing, as are generation-on-generation improvements in AI model efficiency. These improvements include better curation of training datasets, advances in fundamental model design, and operational optimizations in inference.”

Although NVIDIA remains the dominant vendor, Omdia observes that alternatives to GPUs are gaining traction in 2025. These include both custom ASIC chips such as Google’s TPUs, and merchant ASSPs including Huawei’s Ascend series, Groq, or Cerebras. AMD has also made significant progress with its Instinct line of GPUs, boosted by major software investments in 2024.

The newly available Omdia forecast covers GPUs, AI ASICs and ASSPs, FPGAs for AI applications, and CPUs with integral AI acceleration hardware, and provides estimates by market segment, global region, vertical industries, applications, memory content, compute performance, and power consumption bands.