AI Bubble Risks: What May Endure After the Downturn
Only 13% of surveyed IT directors report significant AI benefits. As funding risks mount, proprietary data, durable systems and disciplined capital choices may separate lasting companies from add-ons.
By Nathan Brooks
4 min read

What's News
- Only 13% of more than 2,000 surveyed IT directors reported significant benefits from AI, while average project success stood at one in five.
- Around 40% of AI startups founded in 2024 have already shut down, according to some estimates.
- Gartner estimates more than $3.5 trillion in AI infrastructure spending could lose value by the end of 2030.
- AI21 Labs, previously backed by Google and Nvidia, cut more than 60% of its workforce this year.
- Situational Awareness’s assets fell from $45 billion to around $10 billion in two months in summer 2026.
By Nathan Brooks
The analysis cites a survey of more than 2,000 IT directors in which only 13% said their companies had reported significant benefits from AI. The same analysis says only one in five AI projects succeeds on average, even as more businesses adopt the technology and move pilots into operation.
In October, billionaire Ray Dalio, founder of Bridgewater Associates, described the AI market as a “classic bubble,” comparing its investment cycle with the late-1990s dot-com boom. Dalio pointed to companies financing AI infrastructure with debt while interest rates rise. Citadel LLC founder and CEO Ken Griffin also doubted that the massive AI investment would pay off.
The analysis says more money still flows into AI than the sector earns. Some deals are circular: one company invests in another, which then spends the proceeds buying the investor’s product.
Gartner estimates that, at current investment rates, more than $3.5 trillion poured into AI infrastructure could lose its value by the end of 2030. Rather than forecast the exact burst date, the analysis says investors should identify which businesses may give way first and what opportunities will open afterward.
The platform threat to third-party AI tools
Some estimates cited by the analysis put the shutdown rate among AI startups founded in 2024 at around 40%, or two in five. The analysis cites workflow-automation tools such as Relay, saying they are closing as Google or OpenAI fold those features into their own products. The need for such standalone services is fading.
The analysis says an AI company’s foundation must be its own data and expertise. Startups relying on third-party models risk losing their competitive edge when those model providers launch similar products. Its historical analogy is operating systems that began opening PDFs and compressing files themselves, making hundreds of separate utilities unnecessary. AI, the analysis argues, accelerates that process and leaves less room for add-ons built on someone else’s technology.
What an AI downturn could open
If the AI bubble bursts, the analysis argues that a downturn would not damage every business equally. Scarce talent, newly vacated markets and excess resources can create openings, but capturing them requires more than betting on a crash.
- Talent: A downturn can return experienced specialists to companies that previously could not reach them. The analysis points to AI21 Labs, one of the few independent companies developing its own language models. Backed by Google and Nvidia, AI21 Labs cut more than 60% of its workforce this year, including researchers and developers.
- New markets and niches: The Kauffman Foundation estimates that more than half of the companies on the 2009 Fortune 500 list were founded during a recession or economic downturn. Airbnb launched in 2008 and Uber in 2009, moving into markets that would otherwise have remained occupied in better times.
- Resources: When supply exceeds demand, prices tend to fall. The analysis cites fiber-optic cables since 2000, when Global Crossing built a network of roughly 100,000 miles before going bankrupt in 2002. Its network remained in use. AI data centers and power capacity could similarly find new uses.
Concentration risk in AI investments
The analysis uses former OpenAI researcher Leopold Aschenbrenner as a warning against one-sided bets. He founded the Situational Awareness hedge fund in 2024, investing in AI infrastructure and betting against software companies that might struggle to compete with AI. The two positions looked different but relied on the same assumption: the AI boom would accelerate.
Within two years, the fund’s assets grew to $45 billion. In summer 2026, investors began doubting how quickly the enormous AI investments would pay off. Its assets then fell to around $10 billion in two months, a $35 billion decline. The analysis says Aschenbrenner was not wrong about AI’s potential to change the world. His portfolio offered no hedge because every investment depended on the same scenario.
What should companies test before backing more AI projects?
The analysis says the greater risk falls on companies that cannot calculate an AI project’s expected return before starting it or decide when to stop a project that delivers no results. It urges operators to compare what pilots actually produce with what companies expect.
- Customer durability: Whether customers will keep paying after an AI model provider launches the same functions tomorrow.
- Proprietary foundation: Whether the product rests on the company’s own knowledge, data and experience or someone else’s model.
- Funding quality: Whether growth comes from the company’s own revenue or borrowed money.
- Operating resilience: Whether a solid system keeps the business moving regardless of who is involved, rather than a few key people and luck.
The analysis’s view is that talented people and luck offer only a temporary advantage. Established processes, unique data and expertise built over time create resilience that competitors cannot copy from another company’s model or foundation. Add-ons usually fall apart first when a crisis hits, while durable systems endure. If financing tightens, returning specialists, vacated niches and excess data-center capacity may become available to companies with proprietary data, established processes and pre-set project return tests.
Source: unite.ai
More from Nathan Brooks
Show full bio
News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.
2,276 articles