”Someone's Gonna Get Burned": Sam Altman on the AI Bubble

This editorial appeared in the August 21st, 2025, issue of the Topline newsletter.

Want all of the latest go-to-market insights without the wait? Subscribe to Topline and get the full newsletter,  including bonus commentary and takeaways, delivered to your inbox every Thursday.

I didn't believe we were in an AI bubble when researchers started getting $1B comp packages. Or when pre-revenue startups raised at $2B valuations. Not even when AI companies were closing three rounds in three months.

But then Chamath announced a SPAC.

Shit!

Same week, Sam Altman tells a room full of journalists we're in a bubble. "Someone's gonna get burned," he says. Then he says OpenAI is going to invest "trillions of dollars" in AI infrastructure.

You can't make this up.

Bubbles aren't new. We've lived through dot-com, crypto, ZIRP. Everyone gets a few during their career. But most of us never really understand the mechanics of a bubble. What drives them? Why do smart people fall for them? Or what to do when we're in one?

So, that's what we're going to focus on today.

What Is A Bubble?

A bubble is when asset prices detach from fundamental valuations by a significant margin. Companies trading at 100x revenue multiples. Everyone piles in. Then something spooks the smart money and they bail. Panic spreads, everyone starts selling, and the bubble bursts. Lots of money is lost. Exuberance turns to despair.

Not every bubble needs technology. But the most spectacular bubbles come from the promise of breakthrough technology.

A new technology emerges and humans immediately and correctly recognize it'll change the world. What we're terrible at is predicting exactly how and when that change happens.

So we overshoot. Too much optimism. Too much capital. Valuations that require everything to go perfectly.

And nothing ever goes perfectly.

So, the bubble bursts. But what's interesting is that all that exuberance accidentally builds the future. Railway mania in the 1840s left Britain with infrastructure that powered decades of growth. The dot-com bubble gave us fiber optic networks we still use today. 

The AI bubble will probably do the same.

4 Signs That We Are In An AI Bubble.

1. Even Big Tech can't afford this: Morgan Stanley says we need $3T in data centers by 2029. Just the data centers. Then add power to run them and water to cool them. Big Tech always funded data centers through profits. Not anymore. The spend is so massive their profits can't cover it. So they're turning to private equity and debt. Debt is a terrible way to fund a bet that requires AI to keep improving exponentially, every company on earth to adopt it, and all of them to pay premium prices. Remember, things rarely go perfectly.
   
   
2. Valuations have completely detached from reality: AI startups command the highest valuations in history. VCs are investing 75% more YoY but raised 34% less than last year. Reserves are depleting fast. They need fresh capital from LPs, but LPs are cautious because the returns from the last few years have not been great.
   
   
3. Individual researchers get billion-dollar packages: One person. One billion dollars. We're finding creative ways to justify it. Maybe the math works if that one researcher delivers a breakthrough that changes everything. But Satya Nadella and Sundar Pichai combined made $89M in 2024. When individual contributors make 10x what the CEOs of trillion-dollar companies make, something's off.
   
   
4. Investors think Q2T3 growth is the new normal: Bessemer just invented a new benchmark: Q2T3. Quadruple, quadruple, triple, triple, triple. They say this is what good looks like now. T2D3 - the gold standard of SaaS - is suddenly average. Nobody's actually achieved Q2T3 yet because AI is so new. But during bubbles, impossible math feels inevitable.

Jitters

If you watch CNBC, tech stocks have been in the red for days. If you read tech media, everyone's screaming bubble. Two things spooked them:

1. GPT-5 landed with a thud. The market wants - actually needs - every new model to be an exponential improvement on the last. But GPT-5 was just an incremental improvement, making everyone question if this is the end of the exponential improvements we have been seeing.
   
   
2. A reality check from MIT. Their study found 95% of enterprise GenAI pilots yield zero measurable ROI. Only 5% deliver any value. AI is still early, but people expected better results.

Oh, and Sam Altman admitted we're in a bubble.

So, will the bubble burst tomorrow? Predicting when a bubble pops is as hard as predicting when a technology changes the world. Both require timing the irrational. But, to me these feel like jitters, not the end. Something bigger would need to break for this bubble to truly burst.

What To Do

During bubbles, GTM gets easy. We saw it during ZIRP. Everyone hits their number. Renewals are automatic. Expansion is everywhere. Mediocre salespeople look like heroes.

When the bubble bursts, reality hits hard. Churn spikes. NRR tanks. The same reps who were crushing it can't close a deal to save their lives. Only the truly great survive.

We don't know when this bubble pops. Could be tomorrow. Could be years from now. But GTM professionals need to prepare now.

Be ruthless about company selection. Every startup looks amazing in a bubble. They're all growing 300%. They're all "crushing it." Look deeper. Which ones have real customers paying real money for real value? Join those.

Hire for the downturn, not the bubble. The rep who can only sell when everyone's buying? They're useless when the music stops. Find people who can be effective in both markets. 

Focus on quality revenue. The deals that close themselves right now? They'll churn the second things get tough. Build real relationships. Solve real problems. That revenue sticks around.

But here's the most important part: Master AI. Not tomorrow. Now. Yes, we might be in an AI bubble. But the technology is real, and the GTM pros who master it will dominate both sides of this cycle.

The best operators in tech were forged during the dot-com boom and bust. ZIRP made the next generation stronger. This AI cycle will build the next class of greats - if we treat it as an opportunity to level up, not an excuse to sit out.