According to the latest AI server industry analysis from TrendForce, significant changes are anticipated in the shipment structure of NVIDIA's high-end AI chips by 2026. Influenced by shifts in the international landscape and the time required for supply chain optimization, the proportion of Hopper and Rubin series within NVIDIA's total high-end GPU shipments is projected to decline. This decline is expected to drive a substantial increase in the share of the Blackwell series, rising from 61% to 71%, thereby solidifying its dominant position in the market.
TrendForce indicates that, despite strong AI demand and NVIDIA's active promotion of integrated GB/VR cabinet solutions that utilize a high volume of chips, the projected annual growth rate for NVIDIA's high-end GPU shipments in 2026 has been slightly revised downward from an initial estimate of approximately 26.8% to nearly 26%. This adjustment is primarily due to potential delays in the Rubin series' launch schedule. Challenges include the time-consuming certification process for core components like HBM4, the need to adapt to network transmission upgrades from CX8 to CX9, managing significantly increased power consumption, and optimizing overall performance with more advanced liquid cooling solutions.
Consequently, TrendForce forecasts that the Rubin series' share of NVIDIA's high-end GPU shipments will decrease from an initially projected 29% to 22%. Furthermore, due to international circumstances, the precise timing for H200 shipments remains uncertain, leading to an expected reduction in the Hopper series' shipment share from 10% to 7%. In contrast, the more mature Blackwell solution is projected to surpass 70% of shipments, with the GB300/B300 series forming the core of this volume. While shipments of the GB200/B200 series will be comparatively lower, continued fulfillment of existing 2025 orders and demand from cost-sensitive customers are expected to support its supply through the second half of 2026.
It is noteworthy that beyond strengthening its position in the high-end AI training market, NVIDIA is also actively expanding into AI inference applications. Demand for its new LPU solution is forecasted to reach several hundred thousand units by 2026, with a target of doubling that figure by 2027. Additionally, to address the mid-to-low-end and edge AI application markets, NVIDIA is promoting solutions like the RTX PRO 4500/6000 series. These efforts are expected to increase the proportion of mid-to-low-end products to over 32% of its total shipments in 2026. Amidst product generation transitions and a complex international environment, NVIDIA is reinforcing its absolute advantage in the AI computing power market through diversified product configurations and the development of new applications.
Comments