Does more software mean more businesses?
Not so far, which isn't what I expected. In my last piece I guessed that subscription companies were slowing down because AI has made software so cheap to build that more products are going after the same buyers. So I went looking for all these new products.
They're certainly there. US applications for a new business in the information sector, the closest government figures get to software, are up 60% since 2024. In the EU they're up 37%.
But they don't seem to be turning into businesses. New employers in that US sector haven't moved, and in the UK last year 23,645 information businesses closed while 22,165 opened. So I think I was wrong about what all this means.
The closer it gets to a real business, the flatter it gets
Each bar shows how much something has grown since 2024. Things you can do alone in a weekend are up two to five times, registering a company is up by about half, and anything that involves paying someone has barely moved.
It started in spring 2025, in almost every country
This isn't just an American thing. ChatGPT came out in late 2022, and new companies in this sector barely moved for more than two years. Then they took off in the US, the UK and 26 of the 33 European countries Eurostat covers. The phrase vibe coding dates from February 2025, so I'd guess that's why, though I can't prove it.
Each chart is one country, and the line turns blue at spring 2025. The thin grey line is new companies of every kind.
Applications have tripled since 2005, and employers haven't
The US figures go back further, and they show this is an old pattern. Applications, in blue, have more than tripled since 2005, while the two lines at the bottom, the ones likely to hire and the ones that went on to, have barely moved.
There aren't enough new competitors to explain the slowdown
So does any of this explain the slowdown? The number I'd look at is new revenue, roughly what a company adds from new customers each month. At the median company it fell by about a fifth in a year.
If that was only because the same buyers were being split between more companies, you'd need about 25% more companies. The UK is the one place I can count them, and the number on its register grew 5.5% in the year to March, even counting companies that only exist on paper.
It does look like crowding at the cheap end, though. This is how much new revenue fell, split by what a customer pays each month.
Fall in new revenue over a year, by what a customer pays a month
I'd guess that's because the flood is mostly cheap self-serve products, all chasing the same attention. New App Store apps about doubled in the first half of this year while downloads grew 2%, and Arthur Cnops noticed the same thing on Hacker News.
Find your peer group
Pick your peer group to see how much new revenue companies like yours add now, compared with a year ago.
Could it be something boring?
The US did drop company ownership filings just as this started, but a US rule doesn't do this in Sweden. These are the three explanations I can't dismiss so easily.
Some of them must be. One London postcode holds 12% of this year's new UK software companies, and only about half of the UK companies formed in early 2024 still exist.
Delaware doesn't fit, though. It's the obvious place for shells in the US, and its applications are flat since 2024.
That fits Germany and France, where software job postings have kept falling. It doesn't fit the US, where postings have risen 26% since mid-2025 while applications sped up.
It could be. Census works out the sector from a business's name and description, and UK companies pick their own code, so an agency that's started calling itself an AI company would count.
The US sector also includes things like film and telecoms, though all of its rise in 2025 came from the part that holds software publishers.
So more software hasn't meant more businesses, at least not yet, and I think I can see why. Writing the software was never the hard part. The hard part is finding people who'll pay for it, and AI hasn't made that any easier, so we've got a huge rise in people trying and very little rise in people getting anywhere. RevenueCat says the typical new subscription app makes about $72 a month after a year.
That means I overstated things last time. If you sell something for $20 a month, I think you probably are up against the flood. If you sell something bigger, I'd look at where budgets have gone, because Epoch AI puts OpenAI and Anthropic at about $105B of annualised revenue by August, up from about $30B in January.
I could easily be wrong. These companies are young, so new employers could still jump next year, which is what Census's model expects. A handful of very strong new companies could also be doing the damage without moving these counts, because Stripe says its 2025 companies grew about 50% faster than 2024's.
And none of this explains why it all landed in October, because nothing here does anything that month. I've saved Stripe's count of companies in each peer group, so next time I pull the data I can at least check whether my first guess was right after all.