The demand for AI servers is exploding: Nvidia and Supermicro are leading the way, H100/H200 GPUs are experiencing premium pricing and shortages, and the RTX 5090Ti may be withdrawn from the consumer

At the beginning of 2026, global demand for AI computing infrastructure continued to grow explosively, with the deep collaboration between Nvidia and Supermicro becoming a key highlight in the industry. Following Foxconn's announcement of a 35.5% year-on-year surge in January revenue, the underlying surge in shipments of customized AI servers, primarily driven by Nvidia and manufactured by Supermicro, became apparent.  Simultaneously, the supply-demand imbalance for supporting AI graphics cards intensified, with H100/H200 inference cards experiencing price premiums exceeding 80% and some models facing shortages.  There were even reports that the RTX 5090Ti would be discontinued for the consumer market, only remaining available through enterprise channels. The entire AI hardware supply chain experienced significant fluctuations.

On February 5th, Foxconn's monthly revenue report showed that the company's consolidated sales in January reached NT$730.04 billion (approximately RMB 165.2 billion), a substantial 35.5% year-on-year increase, marking the highest growth rate for the same period in the past three years and far exceeding institutional forecasts of 28%-30%. As a leading global electronics contract manufacturer, the core driver of Foxconn's revenue surge is its AI server business, which is deeply intertwined with Nvidia and Supermicro. Foxconn's subsidiary, Foxconn Industrial Internet, secured 70% of the Nvidia GB300 GPU system assembly orders and is also responsible for the core manufacturing of Supermicro's customized Nvidia AI servers.

Reportedly, Supermicro, as a core partner of Nvidia, focuses on creating highly compatible customized AI servers.  Leveraging the powerful computing capabilities of Nvidia GPUs and its own hardware integration advantages, it has become the preferred partner for global AI giants and cloud service providers in procuring AI infrastructure. Driven by the continuous increase in global demand for large-scale model training and AI inference, Supermicro's customized Nvidia AI server shipments have experienced exponential growth.  Orders for the first quarter of 2026 are already scheduled until the end of the month, and Foxconn's production capacity has provided solid support for this growth, with its GB300 models in continuous mass production, yielding a net profit of approximately RMB 420,000 per unit.

The surge in AI server shipments is accompanied by the persistent supply-demand imbalance for Nvidia's supporting AI graphics cards. Currently, Nvidia's H100 and H200 core inference cards have become essential hardware for AI data centers.  Due to factors such as TSMC's 3nm production capacity being largely locked up by Nvidia and export control approvals, supply remains constrained.  Meanwhile, demand continues to surge due to the accelerated construction of global intelligent computing centers and the booming computing power leasing market, directly leading to price premiums of over 80% for these two graphics cards. According to industry data, the H100/H800 graphics cards are selling for over 250,000 yuan each after the premium, with only 80,000-120,000 units available domestically.  There is virtually no stock available through official channels, resulting in a supply-demand gap of over 30%. The H200 is priced at approximately 190,000 yuan per unit, with limited quantities flowing through unofficial channels and significantly extended delivery times through official channels. Some models are even experiencing a "money but no goods" situation.

More notably, the consumer-grade high-end graphics card market is also being severely impacted by AI demand. Industry sources indicate that Nvidia plans to discontinue the RTX 5090Ti consumer-grade graphics card, shifting all future production of this model to enterprise channels to prioritize computing power needs in areas such as AI servers and professional design. The core reason for this decision is that the RTX 5090Ti, with its 32GB of large video memory, has become a popular choice for unofficial AI modifications and lightweight inference, leading to a large influx into the computing power leasing market. This further exacerbates the graphics card supply shortage and aligns with Nvidia's strategic focus on the enterprise-level AI computing power market. Previous reports have indicated that several models in the RTX 50 series have been gradually discontinued, with tight video memory supply chains becoming a core bottleneck limiting production capacity.

Industry analysts point out that the increased collaboration between Nvidia and Supermicro, and the surge in Foxconn's revenue, are essentially a direct reflection of the global AI computing power infrastructure construction entering an accelerated phase. Goldman Sachs predicts that the global server market will reach $606 billion in 2026, a 40% year-on-year increase, with AI servers accounting for 67% of the market, becoming the core engine of industry growth. High-power, full-rack AI servers will be the main drivers of this growth. As the complexity of large AI models continues to increase and cloud service providers significantly expand their capital expenditures, Nvidia's GPU advantage and Supermicro's hardware integration capabilities will be further amplified.  Their collaboration is expected to continue leading the AI ​​server market. The imbalance between supply and demand for graphics cards, and the divergence between consumer and enterprise markets, may persist into the first half of 2026.

For the supply chain, contract manufacturers like Foxconn will continue to benefit from the growth in AI server shipments, and the demand for supporting hardware such as memory, hard drives, and optical modules will also increase. For consumer users, the discontinuation of the RTX 5090Ti may lead to increased hardware procurement costs in high-end gaming and professional design fields, further pushing the market towards mid-range models or alternative products. In the future, as Nvidia expands its production capacity and the supply chain gradually eases, the supply and demand dynamics of the AI ​​hardware market are expected to stabilize. However, the explosive growth in demand for AI computing power will continue to dominate the development pace of the entire hardware industry in the short term.