Is it really worth that?

On Tuesday a company called Cognition — it makes two AI tools that write software, Devin and Windsurf — took $2 billion from investors. The deal says the whole company is worth $48 billion. In May it was worth $26 billion.

Four months. Nearly double.

If you've been seeing headlines like this every week and quietly thinking how can that possibly be right, that's the correct reaction. Here are the questions people actually ask about these numbers, and the answers — which turn out to be more interesting than "it's all crazy."

Why would anyone pay $48 billion for a company I've never heard of?

They didn't. Nobody handed over $48 billion.

Investors bought a small slice of the company and paid a price for that slice. The $48 billion is what the whole thing would come to at that rate. It's an opinion with money behind it, not a receipt.

Worth saying plainly, because the headlines never do: none of these companies are on the stock market. You can't buy a share of Cognition, or of OpenAI or Anthropic. There's no share price. These figures come from private deals between a company and a handful of investors, which is why they arrive as one enormous number with no explanation attached.

The way to test that number is to ask what the company earns. Cognition is taking in about $900 million a year — with a caveat the headlines always drop. That's a run rate: a recent month's income, multiplied by twelve. Cognition has never actually collected $900 million in a year, nobody audits the figure, and it covers two products, since Cognition bought a rival tool called Windsurf in 2025. Every revenue number below works the same way.

Take it at face value anyway. $48 billion divided by $900 million is 53, so the company was valued at 53 times what it earns in a year — money coming in the door, note, not profit left at the end.

Imagine a coffee shop that takes $100,000 a year, and someone pays $5.3 million for it — 53 times. Nobody does that for this year's coffee. They do it because they don't think it stays one shop.

Has it got more expensive, or just bigger?

This is the question the headline can't answer, and it's the interesting one.

Back in May, when Cognition was worth $26 billion, it was earning about $492 million a year. Run the same sum — $26 billion divided by $492 million — and you get 53 again.

It's the same price. Exactly the same. The company's worth doubled because what it earns doubled — investors are paying today what they paid in the spring, no more. It grew into the bigger number rather than being talked into it.

"Worth twice as much" and "twice as expensive" sound identical in a headline. Only the first one happened here.

Why is the smallest company the biggest stretch?

Because this ratio has nothing to do with size — which you can see the moment you line them up:

• Cognition — worth $48B, earns ~$900M a year → 53 times
• Cursor, a rival coding tool bought by SpaceX — $60B, earns $2B+ → under 30 times
• Databricks — $134B, earns $5.4B → about 25 times
• OpenAI, which makes ChatGPT — $852B, earns over $40B → about 21 times
• Anthropic, which makes Claude — $965B, earns $47B → about 20 times

The smallest company on the list has the highest ratio.

Not the highest price — $48 billion is a fraction of $965 billion, and Cognition would cost far less to buy outright. But set each company against what it actually earns, and Cognition is the biggest stretch of the group, while the two names with the most ridiculous-looking headlines are the most grounded.

For comparison, an established software company you can buy shares in usually sits at about 7 times what it earns in a year, so every company on that list is far above the normal range. The useful point is that the size of a headline number tells you nothing about how big a gamble sits behind it. Anthropic's $965 billion is resting on a great deal more earnings than Cognition's $48 billion is.

So are they really worth that much?

Honestly: it depends on one assumption, and it's worth knowing which.

Cognition is expected to reach $4–5 billion a year by the end of 2026 — a projection reported by The Information, not a promise the company has made. If it gets there, $48 billion works out at about 12 times its earnings — cheaper than Anthropic is today.

So nobody is really paying 53 times. They're paying roughly 12 times what they expect it to earn soon. These prices aren't bets on how big a company is. They're bets on how fast it's moving.

That makes the risk narrower and more boring than "AI is a bubble." It's one sentence: this rate of growth continues. Cognition earned half as much four months ago, and now has to repeat that trick several times over, while a rival sits inside SpaceX and every large AI lab ships a competing tool of its own.

Isn't this just the dot-com bubble again?

On these numbers, no — and it's not close.

In 1999 people were paying around 100 times sales — and those companies were on the stock market, so ordinary people bought in — frequently for businesses with almost no sales at all, and most never made a profit. Today's prices are high, but they sit on real money that is genuinely growing fast.

What is worth worrying about is different: how few players there are, and how often the same ones sit on both sides of a deal.

Nvidia, which sells the expensive chips all these companies run on, is one of the investors in this round. The chip seller is putting money into a customer who will spend it on chips. Cognition's closest rival is owned by SpaceX, which has already absorbed Elon Musk's AI company. Buyer, seller and backer keep turning out to be the same few balance sheets.

In the 1990s, telecom suppliers lent their customers the money to buy their equipment. The sales were real — right up until the customers ran out of money. That's the echo to listen for, rather than the big numbers.

What should I watch, if I only watch one thing?

There is at least a figure with a deadline on it: $4–5 billion a year, by the end of 2026. Treat it as what it is — a reporter's estimate, not a commitment the company has made — but it's specific and dated, which is more than most of these headlines offer.

If Cognition hits it, $48 billion was cheap and the headline was the wrong thing to react to. If it comes in at $2 billion, the problem was never the size of the number — it was that one assumption underneath, and the same assumption is holding up every other name on that list.

So the next time one of these goes past: ask what it earns. If you've ever bought a car, a flat, or a small business, you already know the difference between paying for what something earns now and paying for what you hope it becomes. Which do these look like to you?