A few days ago I wrote about taking my Roth IRA to zero on the Palm IPO in 2000. The bigger point of that piece was that the dot-com era did not hurt the people you would expect. It hurt regular people. Wealthy investors got in early, made huge returns, and got out at the IPO. The general public bought at the top, held through the crash, and absorbed the losses.

This piece is about how the same thing could happen in AI. And about who actually gets hurt this time.

The Three Names You Know

Right now there are three pure-play AI companies everyone is talking about.

Anthropic, the company that makes Claude, is reportedly raising money at a valuation of $900 billion. OpenAI, the company that makes ChatGPT, was valued at $852 billion earlier this year. xAI, the company that makes Grok, was valued at over $200 billion before Elon Musk's rocket company SpaceX absorbed it in a $250 billion all-stock deal in Q1 of this year. The combined SpaceX entity is now preparing to go public this summer at a target valuation of $1.75 trillion. That would be the largest stock market debut in history.

I am leaving Google's Gemini out of this list on purpose. Gemini is built by Google DeepMind and sits inside Alphabet, which is already publicly traded. So the dynamic I am about to describe, where private market valuations get handed off to public investors at the IPO, does not apply to it.

These three pure-play AI companies, Anthropic, OpenAI, and what is left of xAI inside SpaceX, are not where the most extreme risk lives. They are real businesses with real revenue. They are not going to disappear. Whatever the right price for them turns out to be, they will be around long enough for the market to figure it out.

But "real business" does not automatically mean "supports a trillion-dollar valuation." Here is the part that does not get enough attention in the headlines about Anthropic's revenue going from $9 billion to over $30 billion in a few months. These companies are spending hundreds of billions of dollars on data center infrastructure. The chips they are buying for tens of billions of dollars today will be largely obsolete in three to five years. They will need to be replaced. And the replacements will not be cheaper, because the cutting edge of AI chip technology keeps moving and the leading labs cannot afford to fall behind on hardware.

This is a real accounting issue. When a company buys a chip for $40,000 and puts it on its balance sheet, accounting rules let the company spread the cost out over the useful life of the chip, typically five or six years. But the actual useful life of an AI chip in a frontier model training environment is probably closer to three years. The chips are wearing out faster than the books say they are. The companies are showing more value than they really have, and they will have to keep buying more chips at increasing prices just to stay at the frontier.

So even if the revenue keeps growing, the cost of staying in the game grows alongside it. Anthropic's revenue may double again next year. So might its infrastructure spend. Even strong revenue growth may not be enough to support a trillion-dollar public market valuation when the capital required to maintain that growth never slows down.

This is the part of the story I would watch most carefully if I were investing in any of the three. The revenue numbers look extraordinary. The capital requirements behind them look more extraordinary still.

The real risk for everyone else, though, is below them.

The Companies You Have Never Heard Of

There is a financial term called a "unicorn." It refers to a private company valued at over one billion dollars. The term used to be rare, which is why it was named after a mythical creature. Now there are over a thousand of them.

Specifically in AI, there are approximately 500 unicorns globally, with a combined valuation of roughly $2.7 trillion. That number comes from CB Insights as of late 2025, and it has likely grown modestly since.

Take out the three giants I just named. The three of them are worth nearly $2 trillion combined. That leaves about $750 billion spread across the other 498 companies. Most of them are companies you have never heard of. They were valued at anywhere from one to ten billion dollars in the last year or two because money was flowing freely and any company with the word "AI" attached could raise at extraordinary prices.

I think at best, half of those 498 companies will eventually be worth what they raised at. Probably less than a quarter.

The rest will quietly disappoint, get marked down in private, or fail to ever sell shares to the public. Some will go out of business entirely. Some will get acquired for a fraction of what they raised at. A few will pivot into something completely different. This is not a prediction that AI is fake. It is a prediction that the price tags do not match the underlying businesses.

These will be the Pets.coms of this era. The companies that raised on hype, ran out of road, and serve as the cautionary tale of the cycle.

Who Gets Hurt

This is where my background in finance matters, so let me explain it the way I would explain it at a dinner table.

Right now, the people who own these AI companies are mostly venture capital firms, sovereign wealth funds, private equity, and very wealthy individuals. None of those are regular people. None of those represent your 401(k). Your retirement money is not in these companies yet.

The way the system works, it gets into your retirement money when these companies go public and start selling their stock on the regular stock market. That is when normal investors buy in. That is when pension funds and mutual funds buy in. That is when the wealthy investors who got in early start selling their shares to the public and walking away with their winnings.

In the dot-com era, this is exactly what happened. Wealthy investors made fortunes selling to ordinary people at the top of the market. Then the market crashed and the ordinary people held the losses.

The question for AI is whether this same handoff is about to happen at trillion-dollar prices.

What Is Likely To Happen

This is the part most people are missing.

The SpaceX IPO this summer is the first test. SpaceX is not a pure AI company. Most of its $1.75 trillion target valuation is the rocket business and Starlink. But the market is treating it as a partial test of AI valuation appetite because xAI is now part of the entity, and because every IPO at trillion-plus valuations sets a tone for the next one in line.

If SpaceX goes public and the stock does well, that signals to the rest of the market that there is appetite for these kinds of valuations. Anthropic and OpenAI come next. The hundreds of mid-tier AI unicorns rush to follow. And the wealth transfer I described above plays out in full.

There is a less likely scenario where SpaceX disappoints, the other big offerings pause, the mid-tier unicorns cannot find their exit, and the wealthy investors who own them end up stuck holding losses themselves. That outcome would actually be better for most people reading this. The losses would stay with the investors who can afford them, instead of getting passed to the public through 401(k)s and mutual funds.

I do not think that is what happens. I think this is going to look a lot like the dot-com era. The first big offerings probably work well enough to open the floodgates. The mid-tier unicorns follow. Wealthy investors cash out. Ordinary people buy in at the top. And then, somewhere down the road, the cycle corrects.

That is the historical pattern. That is the financial pattern. That is the pattern I would bet on.

What Is Different This Time

But here is where this cycle parts ways from the dot-com era in the most important way possible.

The dot-com era was about putting commerce online. Buying books. Buying pet food. Buying cars. Useful, but mostly incremental. Most of those companies disappeared or got absorbed, and the world kept working roughly the way it had before. Amazon was the rare exception that became something larger than the era that produced it.

AI is not like that. The trend that powered the last fifty years of computing, Moore's Law, the doubling of chip capacity every eighteen months or so, ended around 2020. Chips ran into physical limits that engineering could not solve. Then AI started doubling on a much faster cycle, and on a fundamentally different mechanism. Each generation of AI is now helping to build the next generation. The improvement is self-reinforcing in a way that hardware improvement never was. I wrote about all of this in detail in Moore's Law on Steroids in February if you want the longer argument.

The short version is that AI is going to reshape how people learn, work, build, communicate, and connect with each other. It is the most important technology shift of our lifetimes, and it is probably the most important one in any of our parents' or grandparents' lifetimes either.

The transformation is real. The eventual value created over twenty or thirty years will dwarf what came out of the late 1990s.

That makes this cycle dangerous in a specific way. Because the technology is so obviously important, it becomes easy to confuse the importance of the technology with the soundness of any specific company's valuation. People look at AI, see that it will change the world, and conclude that anything labeled AI must be worth what it is priced at. That conclusion is wrong. The technology being transformative does not mean the prices are right.

What I Am Not Saying

I am not saying AI is a fad. Quite the opposite. I think it is going to create more value than any technology wave before it.

I am not saying don't invest in AI. There are real opportunities at fair prices, especially in smaller companies that have not been bid up by the hype.

What I am saying is be careful at the top of the market. Be careful about buying at trillion-dollar valuations because you love the product. That is what I did with Palm in 2000. The product was great. The thesis felt obvious. The price was wrong.

Be even more careful about the next wave of mid-tier AI companies that go public over the next year or two. The names will sound impressive. The pitches will feel familiar. The marketing will be polished. Most of them will not survive at the prices they raised at.

My Skin in This

I should tell you where I sit in all this.

I am building a company called Kenektic. We build on top of foundation AI models, not as one of them. I am not trying to be the next Anthropic or the next OpenAI. I do not expect Kenektic to be worth $100 billion or even $10 billion. If everything goes right, a billion-dollar valuation would be world-changing for what we are trying to do. A hundred million is my realistic floor.

I tell you this so you know I am not warning about valuation extremes while secretly hoping for one of those valuations myself. I am building a real business with realistic targets. I am writing this because the people who are going to get hurt by what is coming are not the ones reading the headlines about hundred-billion-dollar funding rounds. The people who are going to get hurt are the ones who will read their 401(k) statements three years from now and wonder what happened.

The next twelve months will tell us which version of this story we end up in.

If you have a view, drop a comment below or send me a DM.