Key points
- Box CEO Aaron Levie says the average strongly held AI belief now has "a half life of 6 months at best."
- One belief on his list already erased roughly $590 billion of Nvidia (NVDA) value in a day, then fully reversed.
- Our AI Bubble Index reads 56 out of 100, in the Heated band.
Aaron Levie runs Box, and he spends more time thinking out loud about AI than any other public-company CEO I follow. Early Saturday morning he posted something that stopped me. "The average strongly held belief these days in AI has a half life of 6 months at best," he wrote, and then he listed a dozen of them the industry has already cycled through. "OSS is too far behind to catch up." "The labs can't be profitable at scale." "RAG is dead." "We've hit a training wall." Every one of those was treated as obvious by serious people at some point in the last two years, and every one of them has since been argued right back out of fashion.
He was building on a post from investor Jack Altman a day earlier, who put it more bluntly: "even very smart and plugged in people are continually reversing their opinions and then re-reversing them three months later."
Levie's own landing was calm. "The key here is to remain flexible in your thinking because we're going to be in a constant state of change for a while." That's good advice, and I want to be fair to him here. He's not calling a bubble, and I'm not going to put that word in his mouth.
I'll use it myself instead.
Think about what a 6-month half life actually means for stock prices. The AI trade is worth trillions of dollars, and those valuations rest on beliefs about 2030, like how much compute gets bought and which labs actually turn profitable. One of the biggest is whether agents eat the software industry or end up feeding it, and the market just spent this week re-trading exactly that belief, with software posting its biggest day of 2026 after Salesforce's earnings. We wrote about that rally on Thursday. So when the people closest to the technology admit their strongest convictions keep flipping every two quarters, you've got a market pricing a decade of certainty on top of a foundation that won't hold still for six months. I think the bubble lives in that gap.
The market already traded one of these beliefs, twice
One line on Levie's list has cost investors real money in both directions: "Cheaper models will mean less compute." That was the DeepSeek panic of January 2025. Nvidia (NVDA) fell 17% on January 27 and shed roughly $590 billion of market value, the biggest one-day loss any company had ever taken at the time, because a cheap Chinese model convinced everyone that efficient AI meant fewer chips. The belief barely survived the quarter. Hyperscalers raised capex guidance instead of cutting it, compute demand kept climbing, and Nvidia went on to new highs. Same facts, opposite conclusion, and about half a trillion dollars of round trip in between.
The dot-com era ran the same loop, and it matters what actually ended it, because it wasn't anybody changing their mind on a message board. Money got expensive when the Fed took rates from 4.75% to 6.5%, and the vendor financing that equipment makers used to prop up their own customers reversed. We built our AI Bubble Index around today's versions of those mechanisms, and it currently reads 56 out of 100, a band we label Heated. Belief whiplash isn't one of its seven components. After this week, I think there's an argument it should be.
So what do I actually do with this? Nothing dramatic. I'm not dumping AI exposure because a CEO I like posted a sharp thread, and notice that the bears on Levie's list flipped just as often as the bulls did. "Frontier models are too dangerous to release" was a strongly held belief once too. What I take from it is sizing. If the smartest people in the industry can't hold a conviction for six months, I shouldn't bet like mine will survive to 2030. Hold positions you can afford to be wrong about, and when someone sells you certainty on AI in either direction, remember that you're probably just hearing this month's belief before its half life runs out.
Sources
- Aaron Levie post on X, August 29, 2026
- Jack Altman post on X, August 28, 2026
- AIStockWire AI Bubble Index, reading as of August 28, 2026
- Nvidia market data for January 27, 2025
