Last Updated on July 7, 2026 by Eytan Bijaoui
In January, Bret Taylor told CNBC that AI is “probably” a bubble and that he expects a correction in the coming years.
Four months later, he just raised $950 million for his AI startup Sierra at a $15.8 billion valuation.
Before you call hypocrisy, consider what he’s actually telling you. Because this might be the most honest signal any founder in AI has sent all year.
The Numbers That Don’t Lie
Sierra hit $150 million in annual recurring revenue in its eighth quarter of existence. For context, it took Salesforce — the company Taylor used to run as co-CEO — about six years to reach the same number. Sierra did it in two.
The round was led by Tiger Global and GV (Google’s venture arm), with Benchmark, Sequoia, and Greenoaks also participating. The valuation jumped from $10 billion last September to $15.8 billion now. That’s a 58% increase in eight months, during a period when most AI startups are struggling to explain why they exist.
And here’s the number that matters more than all of those: 40% of the Fortune 50 are now Sierra customers. Not piloting. Not evaluating. Paying. Companies like Prudential, Cigna, Rocket Mortgage, SoFi, Discord, and Rivian are running their customer interactions through Sierra’s AI agents.
What Taylor Actually Means by “Bubble”
When Taylor says AI is a bubble, he’s not saying AI is fake. He’s saying that most of the companies getting funded in AI right now are going to die.
There’s a massive difference between those two statements.
Taylor sits in maybe the most uniquely informed chair in the entire industry. As chairman of OpenAI, he sees exactly how foundation models are evolving, what they cost to build, and where the commoditization pressure is heading. As CEO of Sierra, he sees what enterprise buyers actually pay for — and it’s not a smarter chatbot. It’s a system that picks up the phone, resolves a billing dispute, processes a mortgage question, and does it without putting a human on hold for 47 minutes.
His bubble warning is specific. The companies building thin AI wrappers, the ones with no revenue and a pitch deck full of “AI-powered” buzzwords, the ones competing on which model they fine-tuned last week — those are the bubble. Sierra is not in that category. Sierra replaces actual labor costs with software that works. That’s not a bet on hype. That’s a bet on the oldest business model in tech: doing with software what used to require a room full of people.
The Dual-Hat Problem Nobody Wants to Talk About
Here’s the part that should make you think. Taylor is the chairman of OpenAI and the CEO of a company that’s built on top of OpenAI’s models (among others). He sees OpenAI’s roadmap. He sees their enterprise ambitions. And he’s building a company that could directly compete with OpenAI’s own enterprise agent products.
Reid Hoffman left OpenAI’s board specifically to avoid this kind of overlap. Taylor stayed. He says he’ll “recuse himself whenever there is potential for overlap.” Maybe that’s true. Maybe the governance is clean.
But here’s what’s actually interesting about this arrangement: it tells you something about where the value in AI is going. If the foundation model layer were going to capture all the value — if OpenAI’s enterprise products were going to eat Sierra’s lunch — Taylor would know. He has the best seat in the house. And instead of doubling down on OpenAI, he just raised $950 million to build on top of it.
That’s not a conflict of interest. That’s a roadmap.
What This Means If You’re Building a Startup
The lesson from Sierra isn’t “raise a billion dollars.” The lesson is about where the correction will and won’t land.
Taylor is betting that the application layer — the companies that take AI models and turn them into products that replace real costs for real businesses — will survive. And that the model layer will commoditize. DeepSeek already dropped costs by 90%. Google is giving away Gemini. The old SaaS playbook of charging per-seat is already dead. When the correction Taylor is predicting actually arrives, the companies standing will be the ones with revenue tied to measurable business outcomes, not API wrapper margins.
Sierra’s customers don’t pay because AI is cool. They pay because every customer service call that an AI agent handles instead of a human saves real money. The ROI isn’t theoretical. It shows up on the quarterly earnings call.
If you’re a founder right now, that’s your survival test. Not “is my product AI-powered?” but “does my product replace a cost that a CFO can see on a spreadsheet?” Sierra passes that test. Most AI startups don’t.
The Quiet Race You’re Not Watching
While everyone debates whether OpenAI or Anthropic or Google will win the model wars, a quieter race is happening in the application layer. Sierra just lapped the field in enterprise customer agents. Intercom, Zendesk, and Salesforce’s own Einstein are all trying to play the same game, but Sierra got there first with purpose-built AI agents instead of bolting AI onto legacy helpdesk software.
This is the pattern that will define the next two years of AI: the model makers will fight over benchmarks and capabilities. The application builders will fight over revenue. And when the correction comes, revenue wins.
Bret Taylor told you it’s a bubble. Then he showed you where the floor is.
The floor is $150 million in ARR and 40% of the Fortune 50.
Good luck building above that line.
