Big Ai companies may soon go public, but their path to profit remains unclear. Retail investors will surely end up carrying the risk.
The uncomfortable question about the biggest Ai companies is no longer whether the technology is impressive. It plainly is. The question is whether the businesses built around it can make money. And that’s a very different test.
In this Bloomberg interview, Ai skeptic Ed Zitron makes the case that the current Ai boom is being confused with something more solid than it is. His argument is that investors are looking at a semiconductor rally, vast data centre spending, and breathless corporate announcements, then treating all of that as evidence of a successful underlying business.
But expensive infrastructure is not the same as profit. It may just be expensive infrastructure.
The central problem is simple enough. If companies cannot properly measure the cost of an Ai task, and cannot clearly measure the return from that task, then the commercial case is fantasy. It may look dramatic from a distance, but up close there is very little to hold.
Zitron gives the example of enterprise customers suddenly facing more realistic token-based charges. Once that happens, the question changes. It is no longer “what clever thing can this tool do?” It becomes “how much are we spending, and what are we getting back?”
That is the question every ordinary business eventually asks. It is also the question every investor should ask before buying into a public listing.
The danger is that some of the largest Ai companies may reach public markets before that question has been answered. If OpenAI, Anthropic or similar companies go public while still burning huge amounts of cash, the IPO may function less as a growth opportunity and more as an exit route for earlier investors. Venture capital gets liquidity. Retail investors get the story.
And the story is always seductive. There is a new platform. There will be winners. Costs will fall. Adoption will rise. Monetisation will follow. We have heard versions of this before.
The comparison often made is Amazon. For years, Amazon did not appear to be a conventional profit machine. Then AWS became a giant business, and the patient investors were rewarded. But Ed Zitron challenges that comparison. Amazon had a developing commercial engine and useful infrastructure. Ai model companies just have vast ‘compute costs’, uncertain margins, and products whose current pricing may be subsidised to encourage adoption.
That subsidy point is crucial. Many people experience Ai through cheap monthly subscriptions. That gives the impression that the service is affordable at scale. But if the true cost of serving those users is much higher than the price they pay, the business is not being proven. It is being disguised. Sooner or later, the bill has to land somewhere.
It may land with business customers through higher prices and usage caps. The interview mentions Uber setting limits on some Ai tools after exceeding its Ai budget, and suggests similar stories may become more common. If companies start discovering that the productivity gains do not justify the cost, enterprise growth may slow just when public investors are being sold the opposite narrative.
It may also land with shareholders. Once these companies are public, they are not just speculative private bets. They will be pulled into indices, funds and retirement accounts. That means people who never chose to make a direct bet on loss-making Ai companies may still end up exposed to them through passive investment.
That is the part which should worry us the most.
Retail investors are often the last to receive the full picture. By the time a hot private company reaches the public market, founders, early staff and venture investors may already have had years of upside. The public buyer is offered access at the point where the valuation is largest and the scrutiny should be fiercest.
If the business is sound, that scrutiny should be welcome. If the path to profitability is real, it should survive contact with an S-1 filing, analyst questions and the dull old discipline of cash flow.
But if the numbers reveal that revenue growth depends on underpriced computing, subsidised usage, financial engineering and optimism about future efficiency, then the public market is not being offered a mature business. It is being asked to fund a last-chance rescue.
There may, of course, be winners from the Ai boom. Nvidia has already been a spectacular one. Construction firms, data centre operators and suppliers may do well. Some software companies will find valuable, narrow uses for Ai inside existing profitable businesses. But that does not automatically make the big model companies good investments.
The difference between a technology being useful and a company being profitable is the gap where bubbles form. Ai may change how we work, search, write, code and organise information. It can also be a poor business at the scale and cost structure currently being built.
For retail investors, the warning is not that Ai is worthless. It is that the public markets may soon be asked to value Ai companies as if their profitability is inevitable – and it is not.
Before the ordinary investor buys into the grand Ai flotation opportunity, one question should be asked persistently: where does the profit come from? “Trust us” is not an investment case. And I don’t trust them.
- More reading: Will mega IPOs really hurt index investors?
Image: Deborah Lupton / Pop Chips / Licenced by CC-BY 4.0