Nvidia has continued its dominance of synthetic intelligence (AI) datacentres, with its newest quarterly outcomes displaying income development of 16% – a 93% enhance from the identical interval final yr.

The corporate’s datacentre enterprise reported quarterly income of $35.6bn, and income of $115bn for the complete yr – a 142% enhance in comparison with final yr.

In his ready remarks, Nvidia CEO and founder Jensen Huang mentioned: “Demand for Blackwell is wonderful as reasoning AI provides one other scaling regulation – rising compute for coaching makes fashions smarter and rising compute for lengthy considering makes the reply smarter.

“We’ve efficiently ramped up the massive-scale manufacturing of Blackwell AI supercomputers, attaining billions of {dollars} in gross sales in its first quarter. AI is advancing at gentle velocity as agentic AI and bodily AI set the stage for the subsequent wave of AI to revolutionise the biggest industries.”

In the course of the earnings name, financial analysts questioned Nvidia over DeepSeek, which requires much less highly effective graphics processing items (GPUs), and the truth that cloud service suppliers (CSPs) similar to Microsoft are designing their very own customized chips optimised for AI workloads.

Based on a transcript of the earnings name posted on Seeking Alpha, CSPs account for about half of Nvidia’s enterprise. However there may be additionally rising demand from enterprise clients. Huang mentioned: “We see the expansion of enterprise going ahead,” which he believes represents a bigger alternative to promote Nvidia GPUs long run.

Huang used the earnings name to debate why he thinks new AI fashions will drive up demand, whilst AI fashions develop into extra computationally environment friendly. “The extra the mannequin thinks, the smarter the reply,” he mentioned. “Fashions like OpenAI, Grok-3 and DeepSeek-R1 are reasoning fashions that apply inference time scaling. Reasoning fashions can eat 100 occasions extra compute. Future reasoning fashions can eat far more compute.”

When requested in regards to the danger that CSPs have been growing application-specific built-in circuits (ASICs) as an alternative of utilizing GPUs, Huang responded by speaking in regards to the complexity of the expertise stack that sits on high of the problem, implying that this could be a problem if customized chips have been deployed as an alternative of ordinary GPUs. “The software program stack is extremely exhausting. Constructing an ASIC is not any totally different to what we do – we construct a brand new structure,” he mentioned.

Based on Huang, the expertise ecosystem that sits on high of this Nvidia structure is 10 occasions extra complicated right now than it was two years in the past. “That’s pretty apparent,” he mentioned, “as a result of the quantity of software program that the world is constructing on high of structure is rising exponentially and AI is advancing in a short time. So bringing that complete ecosystem [together] on high of a number of chips is tough.”

Discussing the Nvidia outcomes, Forrester senior analyst Alvin Nguyen mentioned: “Having one more file efficiency from Nvidia appears commonplace regardless of the enormity of the feat. The file earnings symbolize the continued demand for Nvidia AI merchandise. The emphasis on reasoning fashions driving extra, not much less, computation is an efficient verbal counter to the troubles about DeepSeek impacting their demand.”

Nevertheless, in Nguyen’s opinion, Huang’s responses to questions on customized chips that provide a substitute for Nvidia’s GPUs was “dismissive”.

“Their response to the query about customized chips from Amazon, Microsoft and Google threatening their enterprise was dismissive and ignores the necessity for these firms to have choices outdoors of Nvidia and to have semiconductors tailor-made particularly to their AI coaching and inferencing wants,” he mentioned.