Maor Shlomo had an idea during his post-military service vacation. The 31-year-old Israeli entrepreneur wanted to build an app that would let anyone (not just programmers) create software by just talking to it. You describe what you want. The AI builds it. No code required.
He launched Base44 in late 2024. Within three weeks, 10,000 people were using it. Within six months, he had 250,000 users and was already profitable. Wix, the website-building company, saw what was happening and paid him $80 million to acquire the entire company. This was a six-month-old startup with eight employees.
A year later, Base44 is hitting $150 million in annual revenue. On June 29, it did something nobody else in the "vibe coding" business has done yet: it released its own artificial intelligence model.
This is the moment things get serious.
Why this matters more than it sounds
Think of the economics this way. Every time someone uses Base44 to build an app, the platform needs to run code through an AI model. ChatGPT, Claude, or something else. Wix pays OpenAI or Anthropic by the token—the building blocks of text. At $150 million in revenue, those token costs are brutal. If your inference bill is even 30 percent of revenue, that is $45 million a year paying someone else to think for you.
Now imagine you own the brain. You train your own model on all the apps your users have built. You run it on your own computers. Your costs plummet. Your margins explode. That is what Base44 just did.
The competitive problem this creates
There are other vibe-coding platforms. Lovable, a Swedish startup, reached unicorn status (valued over $1 billion) and claims $500 million in ARR. Replit, Bolt, Cursor—they are all trying to let people build software without coding. None of them have built their own AI model. None of them can afford to right now.
Why? Because you need three things: massive amounts of real data on how people use your platform, enough money to burn while you train and fine-tune a model, and a business large enough that the investment pays off before your competitors catch up. Base44 has all three. Lovable probably does too. Everyone else is in trouble.
The data flywheel
Here is what makes Base44 defensible. The company trained its model on "tens of millions of real user interactions." Every app someone builds, every prompt they write, every mistake the AI makes and learns from—that becomes training data. The model gets better. More people use it. More data flows in. The model gets even better.
Your competitor, Replit, can rent Claude from Anthropic. It works fine today. But six months from now, Base44's model is trained on hundreds of millions of interactions specific to app building. Claude is trained on the entire internet. Who wins for actually building apps? Base44.
Why Wix made a good bet
When Wix paid $80 million for a six-month-old startup, it looked crazy. Now it looks like stealing. Maor Shlomo is the kind of entrepreneur who solves a problem so completely that it becomes obvious everyone else is doing it the hard way. He built a business that hit profitability in months and $100 million in revenue within a year.
Wix also knew something else: if you want to own the future of no-code development, you cannot rent the AI brain. You have to own it.
What happens now
The vibe-coding market just got a lot more expensive to compete in. Lovable might build its own model. Replit might try. But they are a year behind now, playing catch-up. Shlomo, meanwhile, is building defensibility the way tech companies used to—by owning something your competitors cannot easily replicate.
That is the difference between a successful startup and one that lasts.
