In 1992, Hock Tan became vice president of finance at Commodore, maker of the Commodore 64. Two years later, Commodore was bankrupt. Tan had a front-row seat to a tech company dying in the middle of a boom.

Today he runs Broadcom, one of the biggest AI chipmakers on Earth. On Monday, he took the stage in Singapore and raised doubts about the investors bankrolling the AI boom.

Many AI investors, he said, have “no clue what they are investing in.” Their expectations are so extreme it’s almost like a casino.

Bearish, right? The market thought so. 

Reuters said the mood on chip stocks soured after Tan’s casino line. In Tokyo, SoftBank, one of OpenAI’s biggest backers, fell 8.5% the next day. Masayoshi Son lost an estimated $5.5 billion on paper. Broadcom fell Monday and again Tuesday.

Of course, Tan had help. 

The same weekend, China’s biggest memory-chip maker, CXMT, said it had found a way to squeeze 30–50% more memory onto each chip. Investors panicked, and memory stocks sank. 

Few bothered to check the details. South Korea’s SK Hynix had laid out plans for the same design back in 2025. And CXMT is still locked out of the high-end memory that AI runs on. It can’t get the machines needed to make it.

Even so, a key index of chip stocks had fallen five days straight. Down more than 8%.

Two scary headlines. Billions gone.

Here’s the thing. The selling started before either headlines. Investors were hunting for any excuse to cash out. Tan and CXMT handed them two.

And everyone ignored the rest of the speech. 

What Else Tan Said

In the same speech, Tan said demand for computing power will stay strong until at least 2030. And his own forecast? Broadcom expected about $58 billion in AI chip sales this fiscal year. Tan sees $230 billion by fiscal 2028.

So which is it? Casino or boom?

Both. 

The gamblers are betting on which AI company wins. The house is everyone selling the hardware those bets run on. Broadcom owns a piece of it. And the house is getting paid from two places. 

The first is code. Earlier this year, the machines started using more AI than the people did. Over the next six months, bot usage grew 14-fold. Human usage didn’t even triple. A coding bot can burn about 1,000 times the computing of a simple coding chat. 

The second is physical AI. Amazon already runs a million warehouse robots, nearly one for every worker. Aurora’s trucks haul freight across Texas with nobody in the cab. And Zipline’s drones have made over 2 million deliveries, and its business grew about 15% a week for seven months straight.

All of them run on computing power, and lots of it. 

Jeffrey Cleveland, chief economist at Payden & Rygel, roughly estimates that one hour of work by an autonomous robot takes as much computing as 1,000 ChatGPT questions.

And no one is betting bigger—or doubling down faster—on physical AI than Elon Musk. 

Elon’s Not Slowing Down

Of course, Tan’s right about the gamblers. They’ll sell whatever they have on the next scary headline.

Commodore’s bankruptcy was a scary headline too. The PC boom rolled right over it. Over the next five years, PC sales more than doubled. Intel, which made the chips inside them, nearly tripled its sales.

Now look at today. Tesla’s Cybercab is already on the road. No steering wheel. No pedals. Paid rides began in Austin on September 4. 

And Elon’s just getting started. He wants robotaxis and robots by the millions, and more satellites in the sky. Every one of these machines works for a living. Every hour on the job runs up the meter.

So here are the rules. One, to Tan’s point, know what you’re invested in. Second, when a scary headline hits, ask one question: Do the machines still need it? If the answer’s yes, the headline is noise. (The answer’s yes.)