Broadcom CEO Hock Tan Compares AI Investing to Gambling

Broadcom CEO Hock Tan has compared AI investing to gambling. In his words, many investors don't fully understand what they are financing and hold extreme expectations about returns. His words carry extra weight because Broadcom itself earns money in this market: the company makes custom accelerators and networking chips for large cloud providers.

This is the view of a supplier who sees orders and delivery schedules, not just stock quotes. Mandeep Singh of Bloomberg Intelligence picks up the theme: on Bloomberg Open Interest, he discussed why financing markets in particular could become a constraint on AI. The announcement gives no figures, dates or deal names, so below we examine the logic of the argument and its context, not the numbers.

What exactly Tan is saying

A gambler in a casino has no cash-flow model, only faith in a win. Tan applies that picture to part of the AI capital: money flows into projects whose mechanics the investor doesn't examine. The second half of his thesis is about expectations. If the market prices in rapid revenue growth from AI services, any pause in orders or delay in launching a data center looks like a failure, even when the business is on plan.

Who is speaking matters. Tan is neither an outside skeptic nor someone with a short position. His company depends on hyperscalers continuing to buy custom chips. When an executive like that warns of overheating, it is a signal about the quality of capital in the industry, not a rejection of the idea itself.

Why the bottleneck may be money

Until recently, the main constraints on AI buildouts were said to be chips, memory, electricity and sites. Singh's argument shifts the focus to financing. The largest platforms long paid for capital expenditures out of operating cash flow. But data centers keep growing in scale, and borrowed money increasingly shows up in the funding structures. In 2025, the press reported on Oracle corporate bonds worth $18 billion and on project financing for a Meta campus through a structure involving Blue Owl. Credit funds, bonds and special legal structures have become part of AI infrastructure.

Borrowed capital is sensitive to interest rates and to sentiment. While investors are confident, it is cheap and available. If confidence drops, every subsequent site gets more expensive, and projects with long payback periods are the first to be postponed. For suppliers like Broadcom, this means a risk that orders get revised not because of technology, but because of lending conditions.

• Source: Bloomberg Tech, October 12, 2026.
• Speaker: Hock Tan, CEO of Broadcom.
• Commentator: Mandeep Singh, Bloomberg Intelligence.
• Tan's thesis: investors don't understand what they are buying and expect too much.
• Singh's thesis: financing markets could limit the growth of AI.
• What the announcement lacks: amounts, timelines, company names and deals.

Which signals to watch

For readers who follow the industry or make purchasing decisions, it is more useful to look past loud statements and watch three things. First: the share of borrowed money in the capital plans of cloud companies and the non-operating structures around them. Second: the terms of new bond issues and private loans for data centers, meaning spreads, maturities and collateral. Third: suppliers' comments on their order backlog and delivery schedules. If caveats about customer financing appear there, that is an early sign of strain.

Companies that build products on AI infrastructure would be wise to build in a margin on prices and timelines: the cost of compute depends on the price of capital too, not just the price of chips. Singh's full analysis is worth watching on the Bloomberg broadcast, because the short summary does not disclose his arguments or data.