Experts predict cooling of AI server market in 2025

US technology sanctions and the supply chain’s preparation for the arrival of NVIDIA’s latest hardware will likely lead to a decline in AI server sales in 2025, The Register reports, citing independent experts. TrendForce reports that AI server shipments grew by 46% last year, primarily due to orders from cloud service providers (CSPs). In particular, server manufacturer Foxconn attributed its recent record revenue growth to AI server sales.

At the same time, TrendForce is considering several scenarios for the development of the server market in 2025 due to the uncertainty that reigns in the AI ​​solutions niche. The most likely is further market expansion, but with lower annual growth. However, even in such conditions, it can exceed 30% – Microsoft, Meta✴, Amazon and Google intend to increase capital expenditures on AI infrastructure. This is expected to help support demand for AI servers.

The worst-case scenario, according to TrendForce, with the next most likely, is an increase in AI server shipments to “only” 20–25%. This forecast takes into account the US tightening its AI chip exports to China, which creates great uncertainty in the market. In addition, delays in the shipment of NVIDIA rack-mount solutions based on Grace Blackwell superchips cannot be ruled out — their structure is quite complex, so large-scale deployment is postponed until the second half of the year.

Image source: Foxconn

A more optimistic scenario suggests that large-scale projects in China and the US (such as Stargate) will help accelerate the deployment of AI servers. Additionally, while DeepSeek’s triumph may reduce the need for large numbers of AI accelerators, it could also expand the use of AI, spurring the growth of AI at the edge. If all goes well, AI server shipments could grow by nearly 35% in 2025.

Recently, IBM CEO Arvind Krishna has already predicted that the use of cost-effective and efficient technologies similar to DeepSeek solutions will not only not lead to a fall in the market, but, on the contrary, will significantly increase the use of equipment after the “entry threshold” for the use of AI models is lowered. A similar position is held by investors in AI infrastructure such as Blackstone and Brookfield, who expect that the demand for data centers and equipment will not decrease.

TrendForce expects DeepSeek’s impact to drive cloud operators to adopt low-cost, in-house chips as the focus shifts from AI training to inference. As a result, servers optimized for running models will take over the market. The server market is likely to become more segmented as large cloud players continue to invest in high-performance solutions and enterprise customers look to lower-cost alternatives.