Global shipments of AI servers in 2025 are projected to grow by nearly 28% year over year, according to research firm TrendForce. That would be a slowdown from a 46% increase in 2024 but Nvidia’s pricey GB200 NV liquid-cooled rack systems are expected to be in high demand as the most advanced AI infrastructure. 

Analysts currently expect Nvidia to report $192.44 billion in revenue for 2025 overall, up from $123.52 billion in 2024. 

TrendForce’s estimates factor in a more than 30% increase in spending on cloud and AI infrastructure from Amazon.com, Google-parent Alphabet, Microsoft, and Facebook-owner Meta Platforms collectively this year, despite Chinese firm DeepSeek’s apparent demonstration of low-cost AI development. However, big U.S. technology companies are likely to expand their use of in-house chips.

“TrendForce highlights that DeepSeek’s influence will drive CSPs [cloud service providers] toward lower-cost proprietary ASIC [application-specific integrated circuit] solutions, shifting focus from AI training to AI inference. This shift is expected to gradually increase the share of AI inference servers to nearly 50%,” the firm wrote in a research note. 

Nvidia has pushed back against suggestions that a shift to inference—the process of generating answers from AI models—from training makes it vulnerable to losing market share. It has previously said inference makes up around 40% of its data-center revenue and is growing fast. It says that its NVL72 server system delivers a fourfold improvement in AI model training but up to a 30 times improvement in inference compared with previous systems.

Among other chip makers, Advanced Micro Devices was rising 0.4% and Broadcom was falling 0.5% in premarket trading. 

Write to Adam Clark at adam.clark@barrons.com