Why Billionaires Don't Invest in AI and Trust Humans More

Not long ago, news of SpaceX's IPO spread around the world — a company operating at the intersection of space and artificial intelligence. In a single day, it raised $75 billion, and its market capitalization reached $2.1 trillion. Next in line are IPOs of "pure" AI companies — OpenAI and Anthropic.

Media outlets enthusiastically reported that investors were rushing to pour hundreds of billions into "the most advanced technology on Earth." Even then, it seemed to me that there was more hype than rational thinking here.

I know that billionaires are, to put it mildly, cautious about such IPOs. And life soon proved the validity of this approach: by June 26, SpaceX shares had approached their offering price. The hype margin had virtually evaporated.

What did the SpaceX IPO show?

SpaceX posted a net loss of $4.94 billion in 2025 (with the AI segment losing $6.35 billion) and negative free cash flow of $19.78 billion. The company is unprofitable because it spends even more on AI data centers than on hardware for its space program. And this despite the fact that within four years, equipment depreciation could reach 90%, given how increasingly powerful and faster chips continue to emerge.

The billions raised through the IPO are also going toward M&A deals: the company plans to acquire Cursor, an AI-agent developer, for $60 billion. Corporate strategy may vary, but by spending such sums, this unprofitable organization clearly plans to raise significantly more — hundreds of billions — in the foreseeable future to purchase new chips, rather than be left with outdated data center equipment.

But where will the money come from? Retail investors provided at least one-third of every dollar raised in the IPO. Ordinary Americans invested record amounts in the shares. Where did the other two-thirds come from? Sovereign wealth funds from Southeast Asia and the Persian Gulf (with the White House's promise of a rapid resolution to the US-Iran conflict being a key condition), but the bulk consists of Americans' pension savings.

In Q1 2026, the share of US household and nonprofit sector wealth held in equities was estimated at 45.7%, and for US households alone, it reached a record high of around 50%. So it's no surprise that ordinary retail investors are exactly the ones the American AI business is targeting.

Why are AI IPOs happening in the US right now?

We are witnessing a parade of American AI companies going public, even though many of them are posting growing financial losses. OpenAI, for example, saw its losses increase nearly eightfold in 2025, to $38.5 billion — yet its IPO filing is already prepared.

All major US AI players are heading for IPOs. And this is no accident. The reason is that large US neural networks have reached the peak of their capabilities under their current development model. Most training on human-written texts is already in the past. Now AI is increasingly training itself — but this is a path to degradation.

The problem now is precisely this: there are no new high-quality datasets for training public neural networks. Corporate AI is a different story, but overall, the limitations of neural networks are obvious — both in terms of tasks and their potential for growth.

Each new piece of unique information created by humans will become increasingly expensive, while information generated by neural networks will become cheaper — despite the rising aggregate costs for AI developers.

From a financial perspective, the current development model of well-known AI companies does not promise a sustainable breakeven point. Yet this does not prevent them from promoting the idea of some "bright AI future" to the masses.

And financing follows these ideas. But dreams are bought by those who prefer compelling narratives over cold calculation. If you buy on such foundations rather than investing based on ROI and other financial metrics, you can easily go broke — and people with substantial wealth want to avoid that.

You need a human

Consider this: there is currently not a single AI-powered automated portfolio management platform that is popular among billionaires. With very few exceptions, they all entrust their money only to a human. Notice that for ultra-high-net-worth investors, chauffeur-driven cars remain more popular than autonomous toys. And this holds true for many aspects of life among the wealthy: they are surrounded mostly by people, not AI agents.

I was pleased to note that the Bank of Russia's methodological recommendations on information security for AI development and use in the financial sector, issued June 16, address many points I've previously raised — including who gains access to information fed into public neural networks, and the quality of the data on which they are trained.

I would highlight the following from the recommendations: "When using AI for automated operations in critically important processes (e.g., payment processes, accounting systems reflecting core business activities), where information security risks are assessed as high, the organization is recommended to implement validation of automated AI-generated results by a human with the ability to modify such results."

It is a fact that humans must remain involved in critical processes.

It's noteworthy that SpaceX managed to negotiate with Wall Street investment bankers on IPO costs roughly six times below the market average. And there is no doubt that such a deal was negotiated not by AI agents, but by humans!

What if you read every book in the world?

When discussing AI, people often repeat the claim that neural networks process information tens or hundreds of times faster than humans, thereby "processing" far more data.

However, speed is not always the most important thing in the world of finance. Even if you read every book in the world and memorized their entire contents, it would not guarantee the creation of something fundamentally new.

In fact, large fortunes are not managed — and never will be — by AI, no matter how far it develops. In their current form, neural networks represent a major source of threats: a tool that can make mistakes but bears no responsibility for them.

I would note that digital art never truly took off in any serious or lasting way, while paintings created by a human hand, sculptures — these only appreciate in value.

AI's limitations and insatiable appetite

Yes, an AI-powered robot will be able to do many things. But it remains a highly limited technological entity — a "man in the image of a machine." It still has very weak memory, and it often struggles to select the most appropriate response to a query.

And despite all this, now and in the foreseeable future, there are enormous bills for chips and electricity, while "saving" on data (AI training on AI-generated texts), and no ability to raise subscription prices to cover real costs.

So far, it has been possible to attract investor funds. Since the beginning of the year, the amount of capital poured into global AI is approaching $5 trillion, by my estimate. This does not include private credit, which flowed without issues until recently to cover chip costs and data center construction for US AI giants — until it became clear that the return flow was weak, and investors are now demanding returns.

There is no dispute that neural networks are good at routine tasks, especially when resources remain to keep employees monitoring them. Due to data leakage risks, the corporate sector will increasingly shift toward closed, organization-specific AI platforms on blockchain. But even then, they will only be viable if the returns from such platforms sustainably exceed their costs.

One cannot simply say: "we're experimenting here, burning through money — maybe something will work out." AI has already passed the phase of enthusiasm and naive wonder from creators and users of new technology. Now it's time to understand how every ruble invested in it pays off.

Can losses from AI adoption be measured?

Perhaps the mass retail investor in the West doesn't care that they are investing in shares of a knowingly unprofitable company that doesn't even explain how it will become sustainably profitable. Perhaps they aren't fully aware that their private pension savings are flowing into such securities.

But the following is clear: users of neural networks won't pay more. Some companies will develop their own neural networks; others will abandon AI entirely as an unjustifiably expensive innovation.

Already, this surge in costs among Western AI developers is fueling global inflation, which will seek to infiltrate every country. This explains both the European Central Bank's monetary tightening and the Federal Reserve's decision on June 17 to hold the base rate steady for the fourth consecutive time.

But monetary tightening in major economic sectors against a backdrop of sluggish global economic growth risks triggering a worldwide recession — a scenario where no one will want to pay "more and more" for AI.

Human existence is geared toward survival — that is a biological law. Business is geared toward generating income and achieving sustainably high and growing equity valuations. And I am convinced that humans are the key link in this.

No one can truly assess the damage when following the logic: "AI can replace an employee." But the damage is significant. AI, for example, cannot replace the connections and contacts a person has within and outside an organization.

Essentially, it's like trying to replace a working heart with an artificial one because it's "cheaper." How much would a business owner lose from ideas that only arise in the mind of a professional? AI is a shovel — but where, how, and for what purpose to dig — that must be decided by a human.

Interestingly, Nvidia — which is directly invested in AI development since it sells chips for the technology — recently stated that they don't understand why layoffs are being linked to AI: "How did it happen that AI became truly productive and useful only six months ago, but people have been getting laid off because of it for two years already? There's no logic to this."

Of course, one could argue with this statement, since elements of AI as a technology for routine tasks have existed for many years, and I am aware of successful examples of its implementation and workforce optimization.

And of course, within any organization, when implementing neural networks, one must carefully assess the potential losses from staff reductions. And have a plan to bring them back if necessary. The key for individuals to remain valuable is to invest in self-development.

I am convinced: the larger the assets under management, the more evident the value of a human being becomes in workflows — a reliable guardian of entrusted data, a professional with extensive experience, unwavering ethics, and clear values.