“Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes,” Sam Altman told reporters at a dinner in San Francisco, and then added, “Is AI the most important thing to happen in a very long time? My opinion is also yes.”
In the weeks since, OpenAI raised billions, employees prepared multibillion-dollar share sales, and Altman publicly forecast that OpenAI will spend trillions on data centers. Meanwhile, dozens of much smaller AI companies have taken huge rounds on little product or revenue: Thinking Machines Lab and Safe Superintelligence have attracted eye-popping paper valuations with no public product; Cognition Labs and consumer-facing “AI tutor” and chatbot startups raised large sums on ideas.
Take also Perplexity, that signed lucrative licensing deals despite limited commercialization. Mercor, an AI recruiting platform, drew a large round while still early on revenue. On the infrastructure side, critics like Alibaba’s Joe Tsai have pointed to massive, data-center builds being discussed and funded before clear customer demand materializes. “I start to get worried when people are building data centers on spec,” Tsai told an investment summit.
New unicorn births collapsed after 2021 with about 721 new unicorns in 2021 versus roughly 110 in 2024. Yet AI made up a much larger share of the smaller cohort in 2024: roughly a third to nearly half of new unicorns, and new-unicorn valuations averaged higher than in the post-peak slump. That concentration means fewer new entrants are reaching scale, and those that do are heavily weighted toward AI. A broad market correction will hit the many firms that rely on continuous fundraising to survive.
Investors and industry leaders have said as much in public. “This looks quite a lot like 1998 or ’99,” Ray Dalio warned on the All-In podcast, adding that “a great company that gets expensive is much worse than a bad company that’s really cheap.”
First, early-stage companies that raised very large seed or Series A rounds on roadmaps instead of customers.
These are firms that raise hundreds of millions or even billions while still at the “concept” stage, long before they can point to customer revenue. The bet is that by stockpiling talent and GPUs early, they’ll be first to market when the big breakthroughs arrive. The risk is obvious: if the pace of adoption slows, they burn through capital without traction. Ex. The aforementioned Thinking Machines Lab (Mira Murati), Safe Superintelligence (Ilya Sutskever), Cognition (Scott Wu).
Second, consumer apps that depend on rapid, low-cost user acquisition but have weak paths to revenue.
The consumer internet taught that user growth doesn’t always translate into durable businesses. Today’s chatbots and “AI tutors” face that same trap: downloads and daily conversations come easily, but monetization is harder.
Character.AI (CEO Karandeep Anand) boasts tens of millions of users and a multi-billion valuation, but relies mainly on a $9.99 subscription tier. Perplexity (Aravind Srinivas) is building search and answer engines with real momentum, but advertising partners have already warned that scale isn’t yet big enough to justify the valuations it’s receiving. Replika (Eugenia Kuyda), one of the earliest AI companions, shows the non-financial risks: regulatory pushback in Italy and FTC scrutiny in the U.S. demonstrate how quickly growth can stall. Like early social networks in the 2000s, these companies are vulnerable if cheap user growth slows before a business model matures.
Fourth, narrow-vertical enterprise plays that need long sales cycles and integration work.
These startups target specific industries with AI, often healthcare, finance, or logistics, and raise heavily on the promise of disrupting entrenched systems. But enterprise sales are slow, integrations complex, and many buyers are cautious about regulatory and ethical risk. When funding conditions tighten, that lag becomes fatal.
Babylon Health (Ali Parsa) raised over $1 billion on its digital health pitch but collapsed when reimbursements and customer uptake didn’t arrive fast enough. Olive (Sean Lane) automated back-office tasks for hospitals, reached a $4 billion valuation, then shut down after customers hesitated and integrations proved unworkable at scale. These are the kinds of plays that need years of steady backing, and are often the first to fail when capital stops flowing.
The historical parallel is the dot com bubble. In 1999–2000 a flood of capital went into consumer experiments and into speculative infrastructure. When users and revenue didn’t scale quickly enough, many companies failed. The survivors and the winners: Amazon, Google, eBay, had real products or platform advantages that let them outlast the storm. Today, the winners will be those with deep distribution, vast compute, and sustainable monetization.
These players are already directing their resources. Meta has raised its 2025 capex guidance to roughly $64-72 billion to build out data centers and AI infrastructure; Microsoft has signaled record quarterly capital spending (about $30 billion in a recent quarter) to scale Azure and related services; Amazon’s data-center and cloud investments have run into the tens of billions as well.
“I do think we will see a collapse in valuations,” OpenAI Investor Vinod Khosla, who saw the dot com boom, said in a recent podcast, even as he argued the cycle will also create a few huge winners.
The bottomline is that a market collapse will have uneven consequences. The biggest firms and the platforms that control the most compute will survive. The long tail of startups that raised big rounds to chase scale, especially consumer apps, niche enterprise plays with weak go-to-market paths, and companies building bespoke data-center capacity will not.
