US tech sanctions and supply chain readiness for racks of Nvidia’s latest gear will likely cause AI server sales to cool-off in 2025.
Taiwan-based market watcher TrendForce says shipments of AI-configured systems grew 46 percent last year on the back of orders from cloud service providers (CSPs).
This trend was also noted by Synergy Research Group, which published stats last month showing that anticipated customer demand for AI drove a 48 percent surge in spending on public cloud infrastructure during 2024, and server maker Foxconn attributed its recent record revenue rise to sales of AI boxes.
TrendForce, however, sees some uncertainty ahead for AI servers, leading it to forecast several alternative outlooks for 2025.
Considered most likely is that the market will continue to expand, but at a lower level of annual growth – albeit one that might exceed 30 percent. Microsoft, Meta, Amazon, and Google are planinng increased capital expenditure in cloud and AI infrastructure, and this is projected to sustain some momentum for AI server demand.
The worst-case scenario, which TrendForce considers the next most probable, is that the growth of AI server shipments will slow to “only” 20-25 percent.
Factors shaping this outlook include potentially stricter US regulations on AI chip exports to China, creating greater market uncertainty. On top of this is possible delays of Nvidia rack systems based on Grace-Blackwell superchips as their complex design pushes large-scale deployment to the second half of the year.
A more optimistic view assumes that proactive policies in both the US and China, such as the Stargate project, could accelerate AI server deployment. It also considers that while the revelations about DeepSeek may have an impact on high-end on GPU adoption, the big cloud companies are procuring AI servers with proprietary silicon faster than expected. DeepSeek could well expand AI adoption, said TrendForce, driving edge AI growth. Should these factors be realized, the analyst forecast AI server shipments to inflate by nearly 35 percent during 2025.
IBM chief Arvind Krishna this month predicted that “a reckoning” over the costs of developing AI models following the release of DeepSeek will help fuel adoption of the technology.
“We will find that the usage will explode as costs come down,” Krishna said. “I think it is a validation – we have been on the point that you do not have to spend so much money to get these models.”
TrendForce said it expects DeepSeek’s influence to drive cloud operators toward lower-cost proprietary chips as the focus shifts from AI training to inferencing, leading to servers optimized for inferencing making up a larger slice of the market.
The AI server customer market is also likely to become more segmented, with the big cloud firms continuing to invest in high-performance solutions, while enterprises will prioritize more economical alternatives. ®
