The cheap AI era is ending, and DeepSeek just proved it

DeepSeek, the Chinese AI lab that became famous for being cheap, has put its prices up, and its customers don't seem to mind. According to The Information, DeepSeek raised the price of its flagship V4 models by between 2.3 and 4.5 times, with peak-hour prices quadrupling. Its annualised revenue has since passed $1 billion.

That's a big moment for a company whose whole reputation was built on doing AI for less. It also says something about where AI prices are heading for everyone else.

From price war to price rise

DeepSeek burst into the headlines in January 2025, when it released a powerful model that it said had cost a fraction of what US labs spend to train. The news wiped nearly $600 billion off Nvidia's market value in a single day, as investors wondered whether the AI boom really needed quite so many expensive chips.

Since then, DeepSeek's cheap API (the paid service developers use to build its models into their own apps) has been a favourite of startups trying to keep costs down. Now it's charging more. According to reports, chief executive Liang Wenfeng has told investors that demand wasn't affected by last month's price rise. Even after the increase, DeepSeek's prices are still among the lowest of the major AI providers.

Read the numbers carefully

A $1 billion "annualised run rate" isn't $1 billion in the bank. It means DeepSeek's recent monthly revenue, multiplied by twelve, comes to about $1 billion. Earlier reporting put its revenue at around $71 million for the first seven months of this year, so the jump is enormous, and higher prices are clearly part of it.

DeepSeek is now finalising a funding round of around 50 billion yuan (about $7.5 billion) that would value it at roughly $75 billion, and it's preparing to list on the Shanghai Stock Exchange. Almost all of these figures come from unnamed sources and investor materials, so they're best treated as a strong signal rather than audited fact.

Why UK businesses should pay attention

Plenty of UK startups and developers have built products on the assumption that AI will keep getting cheaper. DeepSeek's move is a reminder that the price you pay today is a business decision, not a law of physics. Once a provider has enough customers who depend on it, it can charge more.

Then there's the data question. When DeepSeek first took off, the UK's then AI minister, Feryal Clark, told Bloomberg that using it was "a personal choice", but urged people to be "alert to the potential risks and they know how their data will be used". That advice still stands. For UK firms, the sensible approach is the same whichever provider you use: know where your data goes, and make sure you can switch if the price or the terms change.

The case for DeepSeek

Charging a sustainable price isn't a scandal. Running AI models costs a lot of money, and a company that makes a profit is more likely to stick around than one burning cash to win a price war. If DeepSeek's customers are happy to pay more, that suggests it's offering something they value.

Mind the hype. And read the price list.

Sources: PYMNTS (citing The Information), Tech Startups, Insurance Journal (Bloomberg).