Nvidia may be the latest casualty of the ongoing memory drought amid reports of delays to its upcoming next-generation flagship graphic processing units (GPUs).

Trendforce reports shipments this year of Nvidia’s Rubin series are expected to drop from 29% to 22%, with the chip giant struggling with delays from suppliers like SK Hynix and Samsung for their HBM4 high-bandwidth memory product.

Nvidia CEO Jensen Huang told the company’s recent GTC conference that the vendor has $500 billion worth of purchase orders for its current-generation Blackwell and upcoming Rubin lines through 2026, with “at least $1 trillion” in demand through 2027.

But the ongoing memory shortage – which has seen major suppliers pushing lead times out beyond 2028 – has impacted several chip firms, and Trendforce projections suggest Nvidia could as well be affected.

Prior reports suggested SK Hynix was delaying plans to ramp up HBM4 production to the third quarter of 2026, instead of the second quarter as originally planned. The firm is believed to have instead favored a more gradual increase to HBM4 production, with a view to refocusing on HBM3E production for existing Blackwell chips.

Nvidia previously denied claims that Rubin could face a production delay off the back of those prior reports, with taping out of the design (the final design process prior to production) having occurred last August. Huang recently confirmed production on Nvidia's Vera data center central processing unit (CPU), which will be tied to the GPU as part of the Vera Rubin platform, is also underway.

Beyond memory constraint concerns, Trendforce also suggests Nvidia is having problems navigating the transition from ConnectX-8 (CX8) to ConnectX-9 (CX9).

Part of the overall Vera Rubin generation sees Nvidia unveil its next-generation SuperNIC to support 800 Gb/s GPU-to-GPU connectivity. It’ll provide advanced routing and telemetry-based congestion control capabilities for the tech giant’s next-generation flagship product line.

CX8 was introduced in 2024, but Trendforce suggests the chip giant is facing “challenges” transitioning to its next SuperNIC. That same report suggests shipment delays around the Rubin series also stem from issues around managing “significantly higher power consumption and optimizing performance under more advanced liquid cooling solutions.”

While demand for Nvidia’s hardware looks to naturally satiate the inevitable hunger to fuel AI workloads, Trendforce noted the annual growth rate for Nvidia’s GPUs to be revised downward, a measly drop from 26.8% to around 26%, but a drop nonetheless.

While it remains to be confirmed if Rubin delay rumors are accurate, Nvidia is notorious for its meticulous supply chain management.

It shored up supplies of electro-absorption modulated lasers (EML); used in its network switches; and has made substantial investments into securing optical interconnects from brands including Innolux, Lumentum, and Coherent. Add to that Huang’s infamous "chimaek summit," in which he, along with two of Asia's most powerful men, walked into a fried chicken restaurant, chugged beer, and ultimately secured Nvidia's supply of memory hardware.

During Nvidia's recent GTC event, Huang detailed its supply chain efforts, saying he was “constantly just looking across the entire life cycle of the technology and manufacturing across the supply chain” to “prepare our company for growth.”

“When you look upstream, you'll come to the conclusion that we're scaling,” the CEO explained during a press conference attended by SDxCentral. “We're starting to scale our silicon photonics technology with Spectrum-6, and we're starting to add it to our scale-up technology with NVLink in the coming years, which means the amount of silicon photonics technology capacity that we need is substantially higher than the world has today.

“So we work with the supply chain to make sure that we can help them build up the capacity in advance," Huang continued. "Sometimes we give them prepayment. Sometimes we just give them a forecast. Sometimes, if the technology is in capacity or is quite light but needs to be much, much greater, we might decide to invest in the company and provide all of the other things as well. We did that with Coherent, Lumentum, etc.”

It wouldn’t be the first time Nvidia has had to navigate delays for its flagship GPU product.

The company saw a three-month deferment for shipments of its Blackwell GPUs before release after manufacturer Taiwan Semiconductor Manufacturing Corp (TSMC) uncovered a production error related to the processor die that connects two Blackwells in a GB200.

Though Nvidia’s latest efforts pertaining to memory scarcity did see prior reports, it was looking to unbundle video RAM (VRAM) from its GPUs for the consumer and gaming market – meaning add-in card (AIC) partners would have to source their own memory.