Nvidia CEO Jensen Huang reiterates Vera Rubin production is on track and confirms company will produce 'giant amounts' of the GB200-based system.

Jensen Huang, chief executive of Nvidia, denied reports about delays of the company's next-generation AI platform, stating that production volumes of the upcoming Vera Rubin platforms are "giant." Speaking to reporters on the sidelines of an event in Japan, Huang said: "[The reports about Vera Rubin delays are] not true. Vera Rubin is already in production. Giant amounts of production incoming."

Nvidia confirmed production of its Vera Rubin platform in January and sampling in February. The emphasis on "giant amounts of production" incoming is intended to reassure investors that the company remains on track to deliver substantial quantities of its next-generation Vera CPUs, Rubin GPUs, and Vera Rubin NVL72 systems in coming quarters.

However, Huang did not address—or was not asked about—Nvidia's rumored delay of its Kyber NVL144 rack-scale solution. According to SemiAnalysis, this system, which uses copper interconnects and carries 144 AI GPUs, has been delayed by more than a year from 2027 to 2028. An alternative dual-rack design has reportedly been canceled, and a larger CPO-based NVL576 configuration may face delays or limited availability. The setback could constrain Nvidia's Rubin Ultra platform with a smaller NVLink scale-up domain than originally planned.

The Kyber NVL144 architecture was designed to connect 144 Rubin Ultra GPUs using a copper-based NVLink 7 scale-up fabric, requiring a sophisticated PCB midplane to carry high-speed electrical links between system components. SemiAnalysis reports that this midplane proved challenging to manufacture—not due to defective chips or component issues, but specifically because of the manufacturability constraints of the PCB infrastructure itself.

Nvidia's statement on the matter neither confirms nor denies the report but indicates the company will deliver products mentioned in its roadmap. "Our roadmap is intact," an Nvidia spokesperson told Tom's Hardware, without clarifying whether launch schedules remain unchanged.

Nvidia reportedly considered an alternative copper-based design called NVL72x2, which would have placed two Oberon racks back-to-back to expand the NVLink scale-up domain without optical interconnects. However, customers rejected this design due to its unusual operational requirements, though specific objections—potentially involving serviceability, cooling, cabling, or data-center layout—were not detailed.

The planned NVL576 rack-scale solution, which was to combine eight Oberon racks interconnected using co-packaged optics between NVSwitches, has also been postponed or shipped in limited quantities due to "ongoing CPO challenges," according to SemiAnalysis. This suggests Nvidia had been developing CPO-enabled NVSwitch connectivity for the Rubin generation, though it remains unclear whether this technology could replicate Kyber's topology, bandwidth, and latency characteristics or achieve the maturity required for high-volume deployments.

With Kyber NVL144 delayed and NVL72x2 cancelled, Nvidia will only offer 72-way scale-up systems until 2028, potentially giving AMD and Google more competitive scale-up options during 2027–2028. AMD's Mega Pod, based on Verano CPUs and Instinct MI500-series accelerators, is expected to support up to 256 accelerators. Google's TPU 8i can provide roughly 1,024–1,152 accelerators within one low-latency domain, while the TPU 8t can scale to 9,600 chip packages per domain.