Nvidia Rolls Out New AI Chip Amid Soaring Demand

There's plenty of hype in the current wave of artificial intelligence evangelism, but when the CEO of the company that makes most of the chips powering the current surge in generative content creation promises upgraded models each year to meet demand, it's worth listening. Recent comments by Nvidia chipmaker CEO Jenson Huang suggest AI is moving even faster--in ways that are truly transformational for business users--than even the most breathless media accounts indicate. Huang on Sunday addressed the annual Computex conference in Taiwan, and said the push to develop new systems to keep pace with AI development forced Nvidia itself to double its pace of chip upgrades. Huang revealed the company's next generation of AI-enabling chips, dubbed "Rubin," a mere three months after unveiling its then-new "Blackwell" model. The move reflected Nvidia shifting to what Huang described as a "one-year rhythm" going forward, to meet continually rising demands from business users. "Today, we're at the cusp of a major shift in computing," Huang said in his keynote address. "With our innovations in AI and accelerated computing, we're pushing the boundaries of what's possible and driving the next wave of technological advancement." Huang spoke of "AI agents" capable of not only carrying out specific tasks, but also combining an array of functionalities to fulfill more complex objectives, and a wide variety of use cases. Those include using mobile robots that integrate and respond to work environments, and even creating a virtual, AI-based model of life on earth. But for that to happen, faster, stronger, and more energy-efficient chips must be able to handle data bottlenecks, currently a major obstacle. Automated organizing, planning, accounting, and content creation are already being transformed by AI business applications. Huang said new high-powered chips will weave different kinds of generative AI applications together, and make the tech available to businesses, organizations, and eventually individuals mining its capabilities with simple prompts. This transformation will only work if developers transition away from six-decade-old computing concepts and techniques for retrieving stored data, Huang said. Instead, next-generation systems will continually create new, constantly updated solutions that AI agents generate from the most recently available information. The unrelenting pace of development has reset Nvidia's chip upgrade plans and delivery schedules. The "Blackwell" platform unveiled in March will reach customers later this year, and a "Blackwell Ultra" successor is going to market in 2025. The even more powerful "Rubin" model announced Sunday will be available in 2026. The stepped-up pace reflects Nvidia's efforts to protect its leadership in AI chip sales--a market where it has an approximately 70 percent share. Rivals AMD and Intel have other ideas, and are rushing to roll out powerful AI platform chips as well. Huang said unlocking the potential of AI depends on enhancing performance, energy use, and autonomous graphics processing unit (GPU) and central processing unit (CPU) functions--even as they work independently toward common goals. Further development of those cross-nurturing platforms, he said Sunday, will permit businesses using them to speed result delivery by a factor of 100, at just 1.5 times the cost. Huang said companies will also find financial benefits in economies of scale. "The more you buy, the more you save," Huang said, before admitting some listeners might have trouble getting their heads around the reasoning. "That's called CEO math. It's not accurate, but it is correct." It's also what you might expect a maker of AI chips would say, especially as he looks to stoke sales and strengthen his market-leading position. BY BRUCE CRUMLEY @BRUCEC_INC