Global server shipment forecasts expect 15 million units, indicating sustained datacenter buildout.

Global server shipments are projected to surge to 15 million units next year, up from 4 million units this year, according to the Market Intelligence and Consulting Institute (MIC). Artificial intelligence servers will account for approximately 30 percent of that total, driven by massive capital spending from major cloud service providers. Google's parent company Alphabet Inc, Microsoft Corp, Amazon.com Inc and Meta Platforms Inc are projected to budget US$450 million for capital expenditure next year, up from US$400 million this year, MIC ICT Industry Research Center director Edward Lin told a news conference in Taipei. Cloud operators Oracle Corp and CoreWeave Inc are also expected to ramp up purchases of AI and general-purpose servers for their data centers.

The institute projects servers powered by Nvidia Corp's graphics processing units will account for about 80 percent of total AI servers next year, with application-specific integrated circuit-based servers comprising the remaining 20 percent. Nvidia is expected to launch its next-generation AI platform, Vera Rubin, in the fourth quarter of next year and will begin standardizing its heat-dissipation specifications—a move that could weigh on the profit margins of Taiwanese cooling solution providers, according to MIC ICT Industry Research Center deputy director Chris Wei.

Market trends indicate Nvidia plans to require suppliers to provide higher-level system products, including complete compute trays, while continuing to verify server racks, cooling distribution units and cold plates based on its recommended vendor list. Leading cooling solution providers with strong production capacity and cost efficiency, such as Delta Electronics Inc and Lite-On Technology Corp, are expected to have a better chance of participating in the Vera Rubin supply chain. As Nvidia pushes the adoption of liquid cooling over air cooling, Taiwanese makers capable of offering both air and liquid-cooling solutions are positioned to benefit from this transition.

For server and related component makers, those who can integrate power, cooling, server racks and infrastructure are expected to secure more critical positions in the next wave of data center expansion, while those only selling components risk being commoditized, Wei noted. Whether Nvidia's new platform could boost sales for Taiwanese manufacturers depends on whether they are included in the US company's ecosystem and can provide higher levels of system integration.

In personnel developments, Google DeepMind vice president John Jumper, who won the 2024 Nobel Prize in chemistry for his work on artificial intelligence, is leaving the company to join Anthropic PBC. Jumper has been a key member of Google's AI coding development team, and his departure strains the tech giant's efforts to compete with Anthropic, OpenAI and SpaceX in the race to build the most powerful AI models.