Nvidia's Rubin GPU likely to be late as memory validation and cooling headaches bite

Supply chain and technical hurdles

It has been reported that industry watcher TrendForce now expects Nvidia's next‑gen Rubin GPUs to ship later and in smaller volumes than previously forecast. Rubin is forecast to make up about 22 percent of Nvidia's high‑end GPU shipments in 2026, down from an earlier 29 percent projection. The reasons are nitty and technical: time to validate the newer HBM4 memory, migration issues to faster ConnectX‑9 NICs, higher system power draw, and more demanding liquid‑cooling requirements. Sounds boring? It matters — a mismatch in any one of those areas can ripple through the whole AI supply chain.

China‑bound Hopper shipments trimmed

TrendForce also says Hopper accelerators — including H200s headed for China — will ship in smaller volumes, with forecasts sliding to roughly 7 percent of shipments from an earlier 10 percent. It has been reported that the US relaxed some export restrictions late last year and that formal approvals followed, with terms allegedly including a cut of revenue going to the US; convincing Beijing to go along apparently took months. Nvidia CEO Jensen Huang has said the company is ramping capacity and taking orders, but geopolitics and paperwork have clearly carved a chunk out of near‑term expectations.

Blackwell and Groq will pick up the slack

Not all is doom and gloom. TrendForce expects Blackwell GPUs — think GB300 and B300 — to fill much of the gap, projecting roughly 71 percent of Nvidia's GPU shipments this year. Meanwhile, demand for Groq’s LPUs looks brisk: TrendForce expects “several hundred thousand” units this year and roughly double that in 2027. Those chips don’t use conventional DRAM and are pitched to speed up the token‑decode phase alongside GPUs. Translation: more specialized silicon, more complexity — and a lot more chips chasing AI workloads.

Memory prices squeeze customers

There’s a cash sting to all of this. TrendForce warned consumer DRAM prices could jump another 45–50 percent in Q2, after a 75–80 percent increase in Q1; DDR5 and many SSDs are fetching more than triple year‑ago prices. The culprit is familiar: intense AI infrastructure demand layered on the memory market’s usual boom‑and‑bust cycles. We have reached out to Nvidia for comment on Rubin’s timing and the company’s production plans and will update if we hear back.

Sources: The Register