Nvidia CEO Jensen Huang said the future of artificial intelligence will be services that can “reason”, but such a stage requires the cost of computing to come down first.
Next-generation tools will be able to respond to queries by going through hundreds or thousands of steps and reflecting on their own conclusions, he said during a podcast hosted by ARM Holdings CEO Rene Haas.
That will give this future software the ability to reason and set it apart from current systems such as OpenAI’s ChatGPT, which Huang said he uses every day.
Nvidia will set the stage for these advances by boosting its chip performance every year by two to three times, at the same level of cost and energy consumption, Huang said. This will transform the way AI systems handle inference — the ability to spot patterns and draw conclusions.
“We’re able to drive incredible cost reduction for intelligence,” he said. “We all realise the value of this. If we can drive down the cost tremendously, we could do things at inference time like reasoning.”
The Santa Clara, California-based company has more than 90% of the market for so-called accelerator chips — processors that speed up AI work. It has also branched out to selling computers, software, AI models, networking and other services — part of a push to get more companies to embrace artificial intelligence.
Read: Nvidia to face US antitrust probe
Nvidia is facing attempts to loosen its grip on the market. Data centre operators such as Amazon.com’s AWS and Microsoft are developing in-house alternatives. And AMD, already an Nvidia rival in gaming chips, has emerged as an AI contender. AMD plans to share the latest on its AI products at an event on Thursday. — (c) 2024 Bloomberg LP
