Jensen Huang doesn't care about Sam Altman's AI hype fears: he thinks OpenAI will be the first "multi-trillion dollar hyperscale company" | Fortune
Briefly

Jensen Huang doesn't care about Sam Altman's AI hype fears: he thinks OpenAI will be the first "multi-trillion dollar hyperscale company" | Fortune
"Just as Open AI CEO Sam Altman and Meta leader Mark Zuckerberg begin acknowledging that there may be truth to the warnings of an AI bubble, Jensen Huang is doubling down on his bullishness. In a recent podcast appearance with Bill Gurley and Brad Gerstner, the Nvidia CEO brushed aside the growing caution and instead zeroed in on the company he sees as the next dominant force: OpenAI. "OpenAI is very likely going to be the world's next multitrillion-dollar hyperscale company," Huang said."
""The longer you think, the better the answer you get - and thinking requires more compute," he explained. That framing matters because inference is where AI collides with day-to-day usage. Training runs happen in bursts, but inference happens constantly: every chatbot prompt, every AI video render, every background algorithmic tweak consumes processing power. If Huang is right, that relentless demand means AI won't follow the boom-and-bust cycles of earlier technologies"
Jensen Huang predicts OpenAI will likely become a multitrillion-dollar hyperscale company and frames AI's growth as a fundamental shift from general-purpose to accelerated computing. He identifies three scaling laws—pre-training, post-training, and inference—that each exponentially increase compute demand. Huang emphasizes that inference, unlike bursty training runs, operates continuously across every chatbot prompt, AI video render, and background algorithmic tweak, creating relentless processing needs. He argues that longer computation yields better answers and therefore more inference capacity will be required. This persistent, everyday demand for accelerated compute could prevent AI from following prior boom-and-bust infrastructure cycles.
Read at Fortune
Unable to calculate read time
[
|
]