Broadcom's CEO says many AI investors are gambling in a casino -- while raising his own chip-revenue target to $230 billion

The chief executive whose company is one of the biggest beneficiaries of the AI capital-spending boom spent part of this week warning that a lot of the money behind it doesn't know what it's doing. Speaking at an event in Singapore, Broadcom CEO Hock Tan said many AI investors are behaving like gamblers at a table they don't understand. "A lot of investors have no clue what they are investing in," Tan said, according to Bloomberg's report of the remarks -- adding that the sector, at the level of individual investor behavior, is "almost like a casino."

Tan aimed the remark mainly at capital flowing into frontier large-language-model companies -- OpenAI and Anthropic by name -- rather than at the infrastructure layer Broadcom itself sells into. He cited OpenAI's roughly $50 billion annualized revenue run rate in September as real evidence of demand, but noted the company is still losing money because of how much it spends on training. He went further with a second comparison: enterprise customers experimenting with AI tokens, he said, are often like a child burning through arcade credits, spending without learning much from it.

None of that skepticism shows up in Broadcom's own numbers. The company's AI semiconductor revenue more than tripled this fiscal year, reaching roughly $17 billion in its third quarter and a guided $22 billion for the fourth -- and Tan reaffirmed, in the same week as the casino remark, that he still expects $115 billion in AI revenue next fiscal year and $230 billion the year after.

That's the tension at the center of Tan's remarks: he is simultaneously the industry's most prominent skeptic of AI-investor sophistication this week, and the chief executive of the company with arguably the clearest financial interest in that same spending continuing uninterrupted. Nothing in his comments disputes that the underlying demand for Broadcom's custom AI silicon and networking gear is real -- his target for 2028 assumes it keeps growing. The casino he's describing is one level up, in the valuations and expectations attached to the labs and model-makers spending the money, not in whether Broadcom gets paid for the hardware underneath them.

Intel's Lip-Bu Tan made the more structural version of this worry the same week, without using Hock Tan's language. His company has already raised its full-year capital-expenditure forecast to more than $20 billion, up from $18 billion, to keep pace with what he's called one of the industry's most severe supply constraints in its history -- spanning leading-edge logic, memory and substrates all at once, with memory the tightest of the three and no quick relief in sight. His point isn't that the spending is foolish; it's that building the physical capacity to meet AI demand takes years, while raising the capital to fund that build can, in theory, happen much faster -- which is exactly why he's flagged financing access as a constraint worth naming on its own, separate from the chip shortage everyone already tracks.

Read that list and Tan's remarks sound like a straightforward bubble call. They aren't quite that, and the distinction matters for anyone deciding how much weight to give the warning.