Supply chain challenges risk delaying Nvidia's Rubin GPUs • The Register

Nvidia's highly anticipated Rubin GPUs, set to succeed the Blackwell architecture, are facing significant supply chain challenges that could delay their release and reduce shipment volumes, according to industry analysts at TrendForce. The firm has revised its forecast, now expecting Rubin to account for just 22 percent of Nvidia's high-end GPU shipments in 2026, down from an earlier projection of 29 percent.

Memory and Technical Hurdles

The primary bottleneck appears to be the validation process for HBM4 memory, the next-generation high-bandwidth memory that Rubin GPUs will utilize. This validation process is proving more time-consuming than initially anticipated, creating a ripple effect throughout the production timeline. Additionally, the transition to Nvidia's faster ConnectX-9 network interface cards (NICs) presents its own set of technical challenges that are contributing to the delays.

Beyond memory concerns, the Rubin architecture introduces more demanding power requirements and advanced liquid cooling needs. These factors compound the complexity of bringing the new GPUs to market, as system integrators and data center operators must adapt their infrastructure to accommodate these higher-performance components.

Hopper Shipments Also Affected

The supply chain issues extend beyond Rubin, impacting Nvidia's current-generation Hopper GPUs as well. TrendForce now forecasts that Hopper accelerators will represent approximately 7 percent of Nvidia's GPU shipment mix in 2026, down from a previous estimate of 10 percent. This reduction is particularly notable for H200 accelerators destined for the Chinese market.

Despite the Trump administration's December 2024 decision to allow exceptions to US export rules governing high-end AI accelerators to China, with formal approval following in January 2025, the process of restarting H200 production for Chinese customers has been slower than expected. Under the new arrangement, Nvidia must share 25 percent of revenue from these sales with the US government. CEO Jensen Huang revealed at GTC last month that the company is in the process of ramping up manufacturing capacity for H200s for the Chinese market, with purchase orders already in hand.

Blackwell Fills the Gap

While Rubin and Hopper shipments face downward revisions, TrendForce analysts expect Blackwell GPUs, including models like the GB300 and B300, to maintain strong market presence. The firm now anticipates Blackwell shipments to account for 71 percent of Nvidia's GPU sales in 2026, effectively filling the void left by the delayed Rubin rollout.

Groq LPU Demand Surges

In a separate development, TrendForce expressed optimism about demand for Nvidia's newly announced Groq LPUs (Language Processing Units). These specialized chips, designed to work alongside GPUs like Rubin, accelerate the token-generating decode phase of the inference pipeline. Unlike conventional GPUs, Groq LPUs don't rely on traditional DRAM memory but instead use on-chip SRAM, which limits their capacity and necessitates deploying large quantities for effective operation.

TrendForce anticipates demand in the "several hundred thousand units" range for 2026, with projections suggesting this could roughly double to around one million units in 2027 as AI inference workloads continue to grow.

Broader Memory Market Impact

The supply chain challenges affecting Nvidia's product roadmap are part of a larger trend in the memory market. TrendForce warned this week that consumer DRAM prices could rise another 45-50 percent in the second quarter of 2026, building on the 75-80 percent increase seen in the first quarter. This surge has pushed prices for products like DDR5 memory and SSDs to more than triple what they were retailing for at the same time last year.

The combination of robust demand for AI infrastructure and the highly cyclical nature of memory markets is largely responsible for these sky-high prices. As AI training and inference workloads continue to expand, the pressure on memory supply chains shows no signs of abating.

Industry Context

These developments come amid broader industry challenges, including export control violations and geopolitical tensions affecting semiconductor supply chains. Supermicro recently launched an internal probe after staff were charged with China export violations, while Intel found itself entangled in discussions about megafab developments. Meanwhile, Alibaba has produced 470,000 of its own AI chips but admits they remain inferior to Nvidia's offerings and may always lag behind.

As the AI hardware landscape continues to evolve, the interplay between technological advancement, supply chain constraints, and geopolitical considerations will likely remain a defining factor in the industry's trajectory. Nvidia's ability to navigate these challenges while maintaining its technological leadership will be crucial in determining its competitive position in the years ahead.

We've reached out to Nvidia for comment on the potential delays to its Rubin lineup and will update this story if we receive a response.