Broadcom CEO Hock Tan: Many AI Investors Are Like Casino Gamblers
Broadcom CEO Hock Tan stated at an event in Singapore that many AI investors behave like casino gamblers, lacking a clear understanding of the technologies they invest in.
He remarked, "Many investors don't even know what they are investing in," setting extreme expectations and finding that it feels almost like being in a casino. Tan specifically questioned the investment frenzy in cutting-edge large language models, suggesting it has outpaced understanding of the underlying economics, and pointed out that AI model developers need to find customers who can generate returns from technology deployment, rather than relying on users who are currently "wasting money" on tokens for experiments.
He likened corporate AI spending to children in an arcade: "When you were a kid and went to the arcade with $10 in tokens, you spent it in 10 minutes and learned nothing. That's what's happening in corporations today." Tan also mentioned that OpenAI's annual revenue reaching $50 billion in September is a positive signal, but considering the costs of ongoing training and capital expenditures, sustainable profitability has not yet been achieved.
Broadcom raised its fiscal year 2026 AI semiconductor revenue guidance to $58 billion, expecting it to reach $230 billion by fiscal year 2028. The company's AI revenue comes almost entirely from about six major clients. Tan believes that computing demand will remain strong at least until 2030, but capital acquisition may become the next bottleneck. He discussed AI infrastructure financing constraints with Intel's CEO at the same event.
Elon Musk replied to a related post, stating that spending $10 in an arcade as a child is "really a fun way" and that playing many video games can indeed teach something. Michael Burry also shared the related report and commented on it.
Market Mechanism: Hock Tan, as an executive of an AI chip supplier, publicly questioned investors' and companies' understanding and returns on token spending, warning of sustainability against the backdrop of expanding AI capital expenditures and ongoing losses for model companies; funds are flowing from investors and corporate experimental budgets to chip and computing power suppliers, benefiting infrastructure providers like Broadcom (which gain clear demand and revenue guidance), while those lacking return validation in frontier model investments and corporate token experiments face pressure from expectation gaps and capital constraints.
Source: Public Information
ABAB AI Insight
Hock Tan's Broadcom is a major supplier of custom chips and networking equipment for data centers, with clients including OpenAI and Anthropic. Its AI semiconductor revenue has rapidly grown in recent years, providing guidance that it could reach $230 billion by fiscal year 2028. By publicly comparing some investors to casino gamblers in Singapore and criticizing corporate token experiments as wasteful, Tan continues his long-standing cautious stance on the sustainability of the economic model for frontier models.
In terms of capital pathways, the company concentrates most of its AI revenue on a few large clients while emphasizing through public statements that capital could become a bottleneck for expansion. This aims to guide the market's focus toward real deployment returns rather than mere token consumption, thus supporting the long-term visibility of its chip orders. This aligns with its business model of providing custom ASIC services to large model companies, converting demand into predictable semiconductor revenue.
This is analogous to historical cases where chip suppliers questioned application layer valuations and actual utilization rates during the internet bubble or early cloud computing phases, or when leading semiconductor companies publicly stated the efficiency of downstream capital expenditures during semiconductor cycles. Currently, AI is transitioning from a competition in model capabilities to corporate deployment and return validation, with Broadcom positioned at the intersection of expansion and constraints.
Essentially, this represents a restructuring of the industry chain under a transfer of pricing power: chip and computing power suppliers emphasize customer returns and capital constraints, shifting the focus of the AI value chain from showcasing model capabilities to actual economic returns, partially tilting pricing power from pure model companies to deployment and infrastructure providers that can continuously generate verifiable income. The mechanism is that when token consumption cannot be converted into stable customer returns, capital and orders will prioritize segments with clear revenue paths.
ABAB News · Cognitive Laws
- Token consumption does not equal customer returns.
- Extreme expectations make investment feel more like gambling.
- Infrastructure providers first remind about economic models, then discuss expansion limits.