Earlier this week, I wrote an article arguing that there was no obvious AI bubble. I argued that AI companies are making massive investments in data centers due to surging demand for their services, and that demand is likely to continue growing in the next couple of years.
This prompted several thoughtful comments asking variants of the same basic question: if there’s so much demand for this technology, why are AI companies losing so much money? As I thought about how to respond, I became convinced that it would be helpful for me to explain the intellectual framework I use to think about questions like this.
I’m not going to claim any kind of originality here — the ideas I’ll explain below are commonplace in startup finance. But I suspect that many readers haven’t spent much time thinking about them.
So in this piece I’m going to do three things. First I’ll present a stylized example to illustrate some key ideas about how to finance a new company. Next I’ll use real-world examples to illustrate how to distinguish healthy startups from doomed companies. Finally, behind the paywall, I’ll apply this framework to OpenAI and Anthropic.
My claim isn’t that these companies are guaranteed to succeed — all startups face risk, and these companies could certainly fail. It’s also possible that they could survive but never generate a healthy return for their investors.
But I am going to insist that OpenAI and Anthropic are following a standard tech industry playbook. The fact that they are losing more money every year does not necessarily mean they are on a road to bankruptcy — or even that anything especially unusual is going on. After all, Amazon lost money for the first nine years after it was founded. Today it’s one of the most valuable companies in the world.
Imagine you start a coffee shop. The space costs $6,000 per month. Coffee beans cost $2 per cup, and you sell each cup for $4.
The first month, you sell 250 cups, earning $1,000 in revenue. But you spend $500 on coffee beans and $6,000 on rent, so you lose a total of $5,500.
The second month, you sell 500 cups of coffee. That’s $2,000 in revenue minus $1,000 for beans. You still aren’t close to covering your store’s $6,000 in monthly overhead, though; you lose another $5,000.
Despite these early losses, you feel like you’re on the right track. Customers like the coffee. They keep coming back, and some of them bring friends. The third month you sell 750 cups and lose $4,500. The fourth month you sell 1,000 cups and lose $4,000.
Projecting forward, you estimate that you’ll break even around the one-year mark, when you expect to sell 3,000 cups. That will generate $12,000 in revenue, just enough to pay $6,000 for beans and $6,000 in rent. By the end of year two, you expect to sell 6,000 cups of coffee in a month, generating $24,000 in revenue. After subtracting $12,000 for beans and $6,000 for rent, you’ll be left with a healthy $6,000 profit.
Starting a business almost always requires spending a bunch of money up front before you earn your first dollar of revenue. Even after you launch, it usually takes a while to build up a customer base. So it’s very common for a business to lose money for at least the first few months — and sometimes the first few years — before it grows large enough to cover its overhead and start generating profits.
Now imagine that the first store does so well that you decide to open two new stores a year after the original one. So in month 13, store #1 earns a $500 profit. But your other two stores are each losing $5,500 — just as the first store did a year earlier. In total, the company is losing $10,500 — the biggest loss in its short history.
Customers love the two new stores and they grow as fast as the first one. You become so optimistic that you decide to open four more stores at the start of year three. That month, store #1 generates $6,500 in profit and store #2 and store #3 each generate $500 in profit. But stores 4 through 7 are brand new, and so they each lose $5,500. In total, your company has lost $14,500 — another record loss.
A financial analyst writes an article arguing that your company is doomed: the larger your company gets, the more money it loses.
But you’re confident the analyst is wrong. Sure, your newest stores are losing money, but that’s temporary. You expect the new stores to become profitable over time, just like the earlier ones did.
This could go on for a while. Maybe you open eight stores in year four and 16 in year five. If you are particularly ambitious — and have sufficiently patient and deep-pocketed investors — you might be able to open new stores for a decade before you turn your first profit. But eventually, you’ll stop (or at least slow down) the pace of openings, and at that point you will wind up with a big, profitable company.
This is a common pattern in the business world. Once investors are confident that a company has a clear path to profitability, they are often willing to fund another round of expansion — designing another chip, releasing another software version, expanding into another city — without waiting for the previous round of investments to pay off. This is why it’s common to see startups do a series of larger and larger fundraising rounds — $1 million, $5 million, $20 million — before they generate a single dollar in profit.
This is especially common in the technology sector because these are often winner-take-all markets. Frequently there are economies of scale, network effects, or other factors that make the most popular search engine, social network, or online retailer much more profitable than the also-rans. You’d much rather be Google than Lycos or Ask Jeeves. So once you (and your investors) are confident you have a viable business model, it often makes sense to spend heavily to stay ahead of your competitors.
Amazon famously did this for a decade. In the late 1990s and early 2000s, it lost more and more money as it expanded from books to CDs to DVDs to consumer electronics and then to many other products. The company didn’t earn its first full-year profit until 2003, nine years after it was founded.
In the early years, a lot of people questioned whether Amazon would ever turn a profit. But the doubters were ultimately proven wrong. Today Amazon is one of the five most valuable companies in the world. It earned $77 billion in profits in 2025.
It doesn’t always work out that way, of course. In 2017, the startup MoviePass announced a service where customers could pay $9.95 to watch one movie per day in movie theaters. A month of movie tickets costs a lot more than $9.95, and in a 2018 interview, MoviePass CEO Mitch Lowe admitted that the company was losing $21 million per month on the service. But he argued that he was just following in the footsteps of Jeff Bezos.
“Remember Amazon, for what, 20 plus years, lost billions and billions of dollars,” he said. “And today is now the most valuable company out there.”
But MoviePass and Amazon were different in a crucial way. Amazon generally sold products above cost; if a CD cost $9.95 on Amazon, the retailer might have paid $7 or $8 for it. Amazon was only losing money because it was rapidly expanding into new markets where — due to startup costs — it wasn’t profitable yet.
In contrast, a typical customer on a $9.95 MoviePass plan got more than $9.95 worth of movie tickets. MoviePass was buying those tickets from theaters at the full retail price and just eating the losses.
The technical term for this is gross margin:
-
My hypothetical coffee shops had gross margins of 50% because the cost of the beans ($2) was 50% lower than the cost of the coffee ($4).
-
In 2001, Amazon had a gross margin of 21% — if you bought a CD for $10, Amazon’s costs were likely around $7.90.
-
In the first half of 2018 MoviePass charged customers $121 million for MoviePass subscriptions, but had a cost of revenue (i.e. the money they paid for movie tickets) of $313 million. That works out to a negative 159% gross margin.
If a company has positive gross margins — that is, if it’s making some money on every sale — then scaling it up should help it get to profitability. A company with negative gross margins, on the other hand, likely needs a fundamental rethink.