CONNECT WITH US

Grok 3 launch drives AI server demand spike

Ninelu Tu, Taipei; Jingyue Hsiao, DIGITIMES Asia 0

Credit: DIGITIMES

Recent developments in artificial intelligence have intensified the competition among tech giants, as Elon Musk's xAI unveiled Grok 3, which the company claims to be the most powerful AI model to date. This launch is expected to catalyze growth in AI server sales.

The xAI development team emphasized that robust inference models require equally powerful training infrastructure. Their ongoing development of new language models, designed to run on both GB200 and GB300 platforms, underscores these systems' critical role in the evolving AI landscape. The GB300 is scheduled for release in late 2025.

According to Foxconn (Hon Hai Technologies) Chairman Young Liu, DeepSeek's emergence has democratized large-model training. This has expanded AI server demand beyond traditional cloud service providers and high-performance computing operators to include mid-sized enterprises, driving increased hardware requirements.

Quanta reported limited GB200 shipments in the fourth quarter of 2024, with full-scale production expected to commence by the end of the first quarter of 2025. The company anticipates triple-digit growth in AI server sales for 2025, maintaining its characteristically conservative outlook.

Similarly, Wistron forecasts AI server sales to maintain triple-digit year-over-year growth in 2025, matching their 2024 expectations. The company supplies server racks to Dell and motherboards to Super Micro Computer, Inc., both of which provide servers to xAI. Wistron declined to comment on specific customer relationships.

Industry sources report steady demand for advanced AI server racks, including the GB200, alongside robust H100 series shipments. The market impact of the GB300's late 2025 launch remains uncertain.

While ASIC servers have garnered attention following DeepSeek's introduction, industry experts clarify that they complement rather than compete with GPU servers. ASICs offer enhanced customization capabilities compared to GPU servers, enabling them to address specific customer requirements more effectively.