Nvidia broke all previous records for data center income in fiscal year 2025 due to its extraordinary success. The computer giant’s profits from this division alone increased by 142% to an incredible $115.2 billion. Equally impressive was the fourth quarter, which brought in $35.6 billion, a 93% year-over-year improvement and a 16% increase from the prior quarter.
The company’s entire revenue, which was up 114% from fiscal 2024 to $130.5 billion for the year, also reached unprecedented heights. Blackwell, the company’s newest AI chip, is mostly responsible for this phenomenal success, as it generated $11 billion in sales in Q4. Nvidia’s CFO, Colette Kress, referred to it as the company’s quickest product ramp ever, emphasizing its record-breaking manufacturing size and quick adoption.
The implementation of Blackwell has been nothing short of revolutionary. Kress disclosed that the fact that large AI clusters now start with at least 100,000 GPUs indicates a significant change in processing power. Numerous large-scale infrastructures have already started receiving shipments, confirming Nvidia’s dominance in AI acceleration.
The significant role that the company’s Spectrum-X Ethernet technology will play in future large-scale projects is another indication of its growing networking capability. Nvidia’s networking technology is particularly noteworthy because it will power the xAI Colossus supercomputer in Memphis and support the very ambitious $500 billion Stargate data center initiative, a joint venture between OpenAI, Oracle, SoftBank, and Abu Dhabi’s MGX. This further solidifies Nvidia’s dominance in AI infrastructure.
Nvidia forecasts a resurgence in the first quarter of 2025, despite a little sequential dip in networking revenue. The company anticipates that switching from the outdated NVLink 8 systems to the more sophisticated NVLink 72 with Spectrum-X will spur further expansion.
CEO Jensen Huang said that the company will release the enhanced AI chip Blackwell Ultra in the second half of next year. He mentioned minor hiccups in the integration of NVLink 8 to NVLink 72 as reasons for the difficulties he encountered when switching from Hopper to Blackwell. Though he assured stakeholders that Blackwell Ultra will blend in perfectly with Nvidia’s developing AI ecosystem, he commended the team for their quick recovery efforts.
Huang also hinted at the next-generation AI chip, Vera Rubin, but withheld details, promising a full reveal at Nvidia’s upcoming GTC conference in San Jose. “Join us at GTC in a few weeks,” he said. “We’ll be unveiling Blackwell Ultra, Rubin, and a host of innovations that will redefine computing and AI.”
As Nvidia continues to push the boundaries of AI and data center technology, its record-breaking performance signals a future where AI-driven infrastructure dominates the computing landscape.
