The CEO of the company that sells the most chips to the AI boom said publicly that investors are in a casino. Hock Tan, of Broadcom, at a conference in Singapore: most investors have no idea what they are investing in, and they set expectations so extreme it is almost like being in a casino.
The sentence worth a whole post
On AI pilots in companies, the analogy: when you are a kid, you go to the arcade, play. Your parents give you 10 dollars in tokens, you run out in ten minutes and learned nothing.
It is the most precise description I have heard of the average economics of the corporate AI pilot: tokens spent, time passed, zero learning. Not because the technology does not deliver, but because the pilot is born without a hypothesis, without a success metric and without a decision date.
The market answered the same day
That same day had the worst semiconductor session in months: Kospi -6.25%, SK Hynix -10%, Samsung -7.4%, SoftBank -8.5%. The cited trigger was China''s CXMT announcing progress on a new DRAM architecture that increases memory density, with a product plan.
Whoever depends on a single supply vector feels that volatility in the roadmap. It is the same reading as the Silicon Curtain, now with short-term impact on the portfolio.
The OpenAI reading
On the 50 billion dollars of annualized revenue, Tan''s reading: it is still losing money because it spends too much on training, and that at the moment says nothing, except that there is a lot of potential revenue growth in supplying frontier generative AI models.
What remains
The CEO of the company that sells into the boom has the sharpest view that the boom does not pay for itself. When the vendor says the buyer is gambling, the reading left is not structural optimism: it is whoever sees the front of the line.
It is the same lesson the market has given repeatedly this cycle: Medallia erased 5.1 billion in equity when the debt window closed, and the FinOps that rules is the consumption rule, not enthusiasm.
Conclusion
A mature AI pilot has a hypothesis, a metric and a decision date. What Tan''s speech installs is not pessimism, it is criteria: tokens spent with zero learning is not an experiment, it is an arcade. At Tech86, the test we apply to every pilot is simple: in which currency does it prove the result, and when does it close with a decision.