According to Fortune, Ali Ghodsi, the CEO of the $134 billion software analytics firm Databricks, issued a stark warning about AI startup valuations at a recent conference. He blasted the trend of investors pouring billions into companies with zero revenue, calling it “clearly a bubble” and “insane.” Ghodsi predicts this “circular aspect” of the market will get “much, much, much worse” in about 12 months before it corrects. He also revealed that over 80% of new databases on Databricks are now launched by AI agents, not humans. Despite flirting with an IPO, Ghodsi says staying private is a strategic buffer against the volatility he sees coming.
The Bubble Talk Isn’t Just Talk
Here’s the thing: when a successful CEO of a company that big calls something a bubble, you should probably listen. But what’s really fascinating is his claim that even the VCs writing the checks are exhausted. They’re apparently telling him they should just take a six-month break and come back when things are “really financially good.” That’s a stunning admission. It suggests everyone’s playing a game they know is unsustainable, but the fear of missing out is just too powerful. So they keep dancing until the music *has* to stop. Ghodsi’s not predicting an immediate pop, though. He thinks the circular financing—where everyone invests in each other’s deals to inflate valuations—will get even more ridiculous first. Basically, the fever gets higher before it breaks.
Why Databricks Is Hitting Pause
This skepticism explains a lot about Databricks’ own strategy. While competitors rushed to IPO in the 2021 frenzy and then had to slash costs in 2022, Ghodsi held back. And now he’s glad he did. Staying private means they don’t have to answer to public market whims if (or when) this bubble deflates. They can keep hiring and investing in long-term tech instead of scrambling to hit quarterly targets. It’s a luxury most companies don’t have. His point is blunt: going public during a hype cycle can be a trap. You get a sugar rush of valuation, followed by a brutal hangover when reality sets in.
The Real AI Problem Isn’t Tech
Now, for all his bubble warnings, Ghodsi isn’t down on AI’s potential. He’s just down on the silly money chasing it. He points out that the real bottleneck for big companies isn’t the models—it’s their own internal mess. Legacy data systems are “an absolute mess” after decades of patched-together software. Security and governance fears are paralyzing. And he even name-dropped “AI lawyers” as a new speed bump, scrutinizing every move. This is the gritty, unsexy reality of enterprise tech adoption. The flashy foundation models from OpenAI or Google are becoming low-margin commodities. The real value, and the real work, is in building applications that can actually navigate a corporation’s chaotic data landscape and strict rules. That’s a much harder problem than training a model, and it’s where the durable businesses will be built.
Where The Money Really Is
So where does Ghodsi see the real opportunity? It’s in the agents. That stat about 80% of databases being launched by AI agents is a huge signal. The future he’s betting on is one where AI doesn’t just answer questions, but *does* specific, valuable work. Think drug discovery or automated financial research. The application layer is where the revenue potential lies, not in providing the generic model itself. His final piece of advice cuts through corporate nonsense: stop letting executives fight to be the “AI person.” Pick one leader. Otherwise, you get a “three-headed monkey” and nothing gets done. It’s simple, practical advice in an industry drowning in hype. And maybe that’s the whole point of his rant. The insanity of the funding market is distracting everyone from the actual, difficult job of making AI useful.
