Nvidia Unveils New Chip System For AI Video, Software Creation
Nvidia says that deploying $100 million of the new chip-based hardware will allow customers to bring in $5 billion of revenue.
Nvidia Corp., whose chips and systems are at the heart of the artificial intelligence computing boom, said it plans to offer a new product designed to handle demanding tasks such as video generation and software creation.
The product, Rubin CPX, will debut at the end of 2026, the company said. It will be offered in the form of cards that can be incorporated into existing server computer designs or used in discrete computers that can operate separately alongside other hardware in data centers.
The chipmaker argues that the design, a derivative of the new Rubin product line coming next year, is necessary to make certain types of AI work more efficient. It’s another example of Nvidia’s relentless pace in introducing new hardware and software — a speed that competitors have yet to match — aimed at keeping the industry hooked on its offerings.
Some elements of the inference portion of AI computing — the process of generating answers to requests — aren’t as efficient as they should be, Nvidia said. That’s because one chip, a type of graphics processor, or GPU, handles the understanding of the incoming input, then also creates and supplies the response. By separating the understanding portion from the response generation, which the new type of GPU chip will handle, Nvidia said its customers will get more efficient hardware.
Nvidia says that deploying $100 million of the new chip-based hardware will allow customers to bring in $5 billion of revenue. Chief Executive Officer Jensen Huang is increasingly emphasizing metrics of that kind as the technology industry seeks to quantify returns on the massive amounts of cash companies have already sunk into new equipment. Nvidia’s data center unit is on course for revenue of $184 billion this fiscal year, according to analysts’ estimates — more than any other company in the industry generates in total.
According to Huang, CPX is the first chip built specifically for AI models that do reasoning with large amounts of knowledge —millions of so-called tokens — at once.
In software generation, systems built on the new chip will evolve far beyond their current capabilities, the company said. Where systems can now generate code suggestions, with the new hardware they’ll be able to understand “large-scale software projects,” Nvidia said in a statement.
For video generation and search, the new offering will be capable of decoding, encoding and processing on a single chip.
