Systems

Nvidia's Rubin GPU is likely to be late thanks to memory shortage and technical challenges

China-bound Hopper accelerators are also likely to ship in smaller volumes than previously forecast, industry watchers say

Nvidia's next-gen Rubin GPUs may end up shipping later and in smaller volumes than anticipated due to supply chain challenges, TrendForce warned on Wednesday.

The industry watchers now expect Rubin to account for 22 percent of Nvidia's high-end GPU shipments in 2026, down from their previous forecast, which had pinned the mix at 29 percent.

TrendForce cited the time required to validate the newer HBM4 memory used by the chips, challenges with the migration to Nvidia's faster ConnectX-9 NICs, the system's higher overall power consumption, and the more advanced liquid cooling requirements as contributing to the delays.

Shipments of Nvidia's Hopper GPUs, including H200s bound for the Chinese market, are also expected to be lower than initially forecast due to ongoing geopolitical issues between the US and China.

In December, the Trump administration said it would allow exceptions to earlier US export rules governing sales of high-end AI accelerators to China, with formal US approval following in January. The decision meant Nvidia could sell its older, but still potent H200 accelerators to Chinese customers for the first time. In exchange, Nvidia would just have to fork over a quarter of the revenue from those sales to Uncle Sam.

Despite this, it's taken months to convince Beijing to sign off on the deal. At GTC last month, CEO Jensen Huang revealed Nvidia was in the process of firing up its manufacturing capacity to produce H200s again for the Chinese market and that it had purchase orders in hand.

TrendForce now expects Hopper accelerators to account for about 7 percent of Nvidia's GPU shipment mix this year, down from its earlier forecast of 10 percent.

While Rubin and Hopper shipments are expected to be lower than initially forecast, TrendForce says Blackwell GPUs, like the GB300 and B300, are likely to fill the void.

Analysts now anticipate Blackwell shipments to account for 71 percent of Nvidia GPUs sold this year.

Finally, TrendForce is rather bullish about demand for Nvidia's newly announced Groq LPUs, which we explored in depth here. These chips don't rely on conventional DRAM memory, and are designed to work alongside GPUs like Rubin to accelerate the token-generating decode phase of the inference pipeline. 

However, due to their limited on-chip SRAM, large quantities are required for this purpose. As such, TrendForce anticipates demand in the "several hundred thousand units" range this year, and roughly double that in 2027.

In related news, TrendForce also warned this week that consumer DRAM prices could rise another 45-50 percent in the second quarter. That's on top of the 75-80 percent pricing increase we saw in the first quarter.

Memory prices have surged in recent months with many products like DDR5 and SSDs now selling for more than triple what they were retailing for at this time last year.

As we've previously reported, demand for AI infrastructure, combined with the highly cyclical nature of memory markets, is largely to blame for the sky high prices.

We've reached out to Nvidia for comment on potential delays to its Rubin lineup; we'll let you know if we hear anything back. ®