The semiconductor industry is well known for its boom-and-bust cycles, but Nvidia CEO Jensen Huang does not believe a downturn is on the horizon—even as he leaned on reasoning that has historically been invoked to justify bubbles.

Chip stocks initially surged on the AI frenzy, only to sell off sharply in recent weeks as investors renewed concerns about the sustainability of massive capital expenditures. The sell-off came despite robust earnings and forward guidance from leading chipmakers, along with ongoing shortages fueled by seemingly insatiable demand.

In an interview with Axios cofounder Mike Allen, Huang was asked whether the sector is headed for a bust. He replied, "no, not for a while." When Allen followed up with "so this time is different?" Huang embraced the phrase.

"This time is different because this is not demand driven," Huang said. "This time is different because it's not seasonal. It's not demand driven means seasonal-demand driven. This is industrially driven, meaning the fundamental technology of computers is changing."

He added that the world requires an entirely new layer of infrastructure—AI—which in turn demands chips. He estimated the industry must grow five to 10 times larger over the next decade to meet that need.

Such optimism is perhaps unsurprising from the CEO of the leading AI chip supplier. Nvidia controls an estimated 80% or more of the market for AI chips, a position reinforced by its widely adopted CUDA software platform, which has become deeply embedded in AI development workflows. What stands out, however, is Huang's explicit endorsement of "this time is different"—a phrase historically deployed to argue that extraordinary gains can persist in defiance of fundamentals or logic, most notably during the dot-com bubble. That era also saw a massive infrastructure buildout, as telecom companies laid thousands of miles of fiber-optic cable, much of which sat unused for years before demand eventually caught up. The expression is now so notorious that it is widely treated as a red flag whenever it surfaces in bullish forecasts, not unlike the infamously premature declaration of "mission accomplished."

Meanwhile, hyperscalers have been committing hundreds of billions of dollars annually in capital expenditures to build out AI infrastructure as rapidly as possible. While these companies previously funded capex through their enormous cash-generating operations, that approach is no longer sufficient. Even Alphabet has recorded negative cash flow, and as a result, tech giants are increasingly turning to debt issuance.

Several of Nvidia's largest customers have also been developing their own custom AI chips—Google with its Tensor Processing Units, Amazon with its Trainium and Inferentia accelerators, Microsoft with its Maia chip, and Meta with its MTIA—though Nvidia's CUDA software ecosystem remains a significant competitive moat.

When pressed on whether he is concerned that Nvidia's customers are tapping the bond market to purchase his chips, Huang said he is not worried, pointing again to the broader shift in computing.

"So this future is a whole new way of doing computing that's fundamentally different than the past, and we need a lot more computers," he explained.

Huang also noted that AI has already proven profitable for companies like Anthropic, particularly as customers discover the utility of AI agents. AI is now at an inflection point, he said, where the technology must be built out further as it generates profits and enhances productivity.

He acknowledged that a bubble will eventually burst, but insisted it will not happen anytime soon, given that the AI buildout remains in its early stages. Furthermore, he argued that the limited supply of chips, land, power, and construction workers—which is holding back even faster growth—is actually beneficial, because it delays the point at which supply overtakes demand.

The power constraint he referenced reflects a broader challenge: data centers already account for an estimated 1 to 2% of global electricity consumption, and the energy demands of AI workloads are substantially higher than those of traditional computing.

"We basically are constrained in every single direction, in every single way," he said. "That constraint is good. That constraint is what holds the system back. So that gives us plenty of time to go build out these infrastructure."

This story was originally featured on Fortune.com.