Broadcom's Hock Tan: Many AI Investors 'Have No Clue'

Broadcom has generated about $68 billion in AI semiconductor revenue since late 2023 — and its own CEO says many of the investors funding that boom have no idea what they are buying. Hock Tan, speaking Monday at the DBS Asia Future Forward Conference in Singapore, likened those investors to gamblers, according to Bloomberg.

"A lot of investors have no clue what they are investing in," Tan said. They "set very extreme expectations only to find out that it's almost like they are in a casino."

The irony is hard to miss. Broadcom designs custom AI chips for the hyperscalers pouring billions into data centers, making it one of the biggest beneficiaries of the buildout. In its fiscal third quarter, which ended Aug. 2, AI semiconductor revenue more than tripled from a year earlier to $16.7 billion. In plain terms, that single quarter's AI chip sales nearly matched a quarter of the company's cumulative total since the boom began.

What does the CEO actually doubt?

Not the demand for computing. Tan's skepticism targets the economics on the other side of the trade — the frontier AI labs that buy Broadcom's chips and must find enough paying customers to cover the cost of developing and running their models.

He pointed to OpenAI's reported $50 billion in annualized revenue at the end of September as evidence of a viable business model. But annualized revenue extrapolates a company's current sales pace across a year; it does not measure cash collected over 12 months. And the costs are enormous: OpenAI expects to spend over $25 billion on training alone this year — half its reported annualized revenue figure.

"For now it's still losing money because they're spending a lot on training," Tan said. "So frankly, it doesn't tell anything at this point except that there is a lot of potential revenue growth out of providing frontier generative AI models."

Is money the next AI bottleneck?

Tan said a lack of capital "might be the biggest constraint" on the AI industry — a bigger risk than shortages of memory, semiconductors or electricity. His warning lands as Wall Street debates whether the AI bubble will burst and some analysts predict AI capex growth will slow next year.

The market has already delivered one warning shot. Firmus, an Australian data center company backed by Nvidia, withdrew its planned IPO on Oct. 9 after weak investor demand for its marketed share prices, Bloomberg reported.

That does not necessarily signal weakening demand for AI computing. Intel CEO Lip-Bu Tan, who shared the panel at the conference, said hardware could remain in short supply for another two to three years, per Bloomberg. Chip demand, however, does not guarantee that the companies buying the chips will earn enough to justify their investments.

Are enterprises 'wasting money' on AI?

Tan's criticism extends to the companies paying to use AI. He accused them of "wasting money" on tokens to experiment with AI instead of identifying where the technology generates returns — and he drew an unflattering analogy to childhood arcades.

"When you are a kid you go to a video arcade, you play. Your parents give you $10 of tokens, you finish it in 10 minutes and you haven't learned anything," he said. "That's what's happening in enterprises today."

The numbers behind his jab exist. A McKinsey survey on AI's return on investment, published in August, found 80% of respondents said AI had improved their individual productivity. Yet only 37% reported AI had contributed to their organization's earnings before interest and taxes. Roughly one in five said AI-related operating expenses, including token costs, had constrained their use of the technology.

The gap between those figures — individual gains without corporate profits — is the quiet engine of Tan's casino warning. The industry's next constraint, by its supplier's own account, may not be silicon or power but investor patience.