Nvidia posted record-breaking data center revenues of $115.2 billion for FY25, a 142 percent increase on the revenue generated by the segment during the previous financial year.
Fourth quarter data center revenue also broke records, totaling $35.6bn, up 16 percent from the previous quarter and up 93 percent from a year ago.
Full-year revenue for the GPU giant totaled $130.5bn, up 114 percent from FY24, while Q4 2025 revenue totaled $39.3bn, up 12 percent from Q3 2025, and 78 percent year-on-year (YoY).
Of the $39.3bn Q4 revenue, $11bn came from sales of the company’s new Blackwell chips. Speaking on an earnings call after the results were published, Colette Kress, CFO and EVP at Nvidia said Blackwell sales had “exceeded expectations,” adding that “this is the fastest product ramp in our company's history, unprecedented in its speed and scale.”
She went on to say that Blackwell production was now in full-year across multiple configurations and the company was increasing supply to quickly expand customer adoption.
“With Blackwell, it will be common for these clusters to start with 100,000 GPUs or more,” Kress said. “Shipments have already started for multiple infrastructures of this size.”
The CFO also used the earnings call to announce that Nvidia would be providing its Spectrum-X ethernet networking platform for the “first Stargate data centers.”
Announced on the first day of President Trump’s second term, Stargate is a $500bn joint venture between OpenAI, Oracle, SoftBank, and Abu Dhabi's MGX to deploy a number of large data centers across the US.
The xAI Colossus supercomputer in Memphis, Tennessee, also uses Nvidia Spectrum-X for its Remote Direct Memory Access (RDMA) network.
Nvidia said that networking revenue declined three percent sequentially but Spectrum-X and NVLink Switch revenue increased, representing a “major new growth vector.”
“We expect networking to return to growth in Q1 [2025],” Kress said, adding that Nvidia was transitioning from small NVLink 8 with InfiniBand, to large NVLink 72 with Spectrum-X.
Looking ahead, CEO Jensen Huang said on the same earnings call that Nvidia’s next AI chip, Blackwell Ultra, would be available in the second half of next year, with new networking, new memory, and new processors but the same system architecture as Blackwell.
Appearing to reference the overheating challenges suffered by early iterations of the GB200 NVL72, Huang said the “hiccup” cost the company a couple of months, but the “team did an amazing job recovering all our supply chain partners, and so many people helped us recover at the speed of light.”
He added that while the transition from Hopper to Blackwell was “challenging” as it required moving from an NVLink 8 system to an NVLink 72-based system, Blackwell Ultra “will slot right in.”
While Huang also made reference to Nvidia’s next chip, Vera Rubin, he didn’t share any further details about its architecture, instead telling those dialed into the call that he’d be talking more about it at the company’s annual GTC event in San Jose, California next month.
“Come join us at GTC in a couple of weeks,” Huang said in his closing remarks. “We're going to be talking about Blackwell Ultra, Rubin, and other new computing, networking, reasoning AI, physical AI products, and a whole bunch more.”
