Nvidia has continued to see document progress in its datacentre enterprise, pushed by the acceleration of synthetic intelligence (AI) workloads.

The corporate reported quarterly income of $35.1bn, up 17% from Q2 and up 94% from a 12 months in the past. Its datacentre enterprise, which offers graphics processing models (GPUs) for AI servers, contributed $30.8bn in the course of the quarter, representing the majority of the corporate’s income.

“The age of AI is in full steam, propelling a worldwide shift to Nvidia computing,” mentioned Jensen Huang, founder and CEO of Nvidia.

Discussing the corporate and next-generation chips, he added: “Demand for Hopper and anticipation for Blackwell – in full manufacturing – are unbelievable as basis mannequin makers scale pre-training, post-training and inference.

“AI is remodeling each trade, firm and nation,” mentioned Huang. “Enterprises are adopting agentic AI to revolutionise workflows. Industrial robotics investments are surging with breakthroughs in bodily AI, and nations have woke up to the significance of creating their nationwide AI and infrastructure.”

In response to a transcript of the earnings name posted on Seeking Alpha, chief monetary officer Colette Kress mentioned gross sales of Nvidia H200 GPUs elevated considerably to “double-digit billions”. She described this as “the quickest product ramp in our firm’s historical past”, including that cloud service suppliers accounted for about half of Nvidia’s datacentre gross sales.

In the course of the earnings name, Huang mentioned strategies to enhance the accuracy and scaling of huge language fashions. “AI basis mannequin pre-training scaling is unbroken and it’s persevering with,” he mentioned. “As you understand, that is an empirical legislation, not a elementary bodily legislation, however the proof is that it continues to scale.”

Huang mentioned that post-training scaling, which started with reinforcement studying human suggestions, is now utilizing AI suggestions and artificial generated knowledge.

One other method is check time scaling. “The longer it thinks, the higher and higher-quality reply it produces, and it considers approaches like chain of thought and multi-path planning and all types of strategies essential to mirror, and so forth and so forth,” he mentioned. “It’s a bit of bit like us doing the pondering in our head earlier than we reply a query.”

Huang mentioned rising the efficiency of GPUs reduces the price of coaching and AI inferencing. “We’re lowering the price of AI in order that it may be far more accessible,” he added.

elements that would curb the corporate’s phenomenal progress trajectory, Huang mentioned: “Most datacentres at the moment are 100 megawatts to a number of hundred megawatts, and we’re planning on gigawatt datacentres.

“It doesn’t actually matter how massive the datacentres are, the power is limited,” he mentioned, including that what issues to Nvidia datacentre prospects is how they’ll ship the best efficiency per watt, which interprets straight into the best revenues.