When Wix paid $80 million for Base44 last year, the startup was barely six months old and had a team of eight. Now the company is doing something that most AI startups only talk about: training its own large language model. According to TechCrunch, Base44 has started rolling out Base1, a custom model built on tens of millions of real user interactions from its platform.
The move is a direct response to a question that has been hanging over the AI startup world for a while now: if your product runs on someone else’s model, how defensible is your business really? For vibe coding platforms like Base44, which let users build apps using plain language instead of code, that question has become impossible to ignore.
Base44 founder Maor Shlomo put it plainly. “Training and owning the model as part of our entire stack allows us a lot more optimizations on latency, cost, and efficiency,” he said. The goal is for Base1 to eventually outperform the frontier models Base44 currently relies on for app creation tasks.
The timing matters. The vibe coding space has gotten crowded fast. Swedish startup Lovable hit unicorn status in its Series A last summer and recently reported $500 million in annual recurring revenue. Base44 passed $100 million in ARR a few months ago, which is strong growth but still a significant gap. Both companies are chasing the same users, but Base44 is now betting that owning its infrastructure will be the deciding factor.
Jonathan Userovici, a general partner at VC firm Headline, whose portfolio includes Mistral AI, sees three things that make an AI startup defensible over time: distribution, data, and tech stack. Base44’s play hits all three. The company says Base1 was trained on data generated from its own user base, which keeps growing. That data flywheel is hard to replicate from scratch.
Shlomo also expects competitors to follow. “At least the players that have gotten enough scale and velocity to have enough data” will eventually train their own models, he said. Lovable, which still relies on external models, is the obvious comparison. But Shlomo is careful not to frame this purely as a startup rivalry.
The bigger threat may actually come from the foundation model labs themselves. Claude Code has become a serious vibe coding product in its own right. Cursor and xAI, the company behind Grok, are now both part of SpaceX. These companies have access to massive data and feedback loops that they can feed directly back into model development. That puts them closer to Base44’s core product than ever before.
Shlomo thinks specialization is still an advantage. “Models are progressing, but they’ll stay very general in what they can do,” he said. Userovici is more cautious. He pointed to Harvey, the legal tech startup that ultimately abandoned plans to train its own model, as a reminder that building foundation-level AI is expensive and hard to justify at every scale.
What has changed the calculation for a company like Base44 is the cost of inference. Running large frontier models at scale is not cheap, and enterprise customers are starting to push back. Userovici described a shift already underway: “They don’t necessarily see a return on investment when using the latest models for all use cases, so an entire infrastructure is being set up to do orchestration and optimization to select the right models for them so that costs don’t skyrocket while maintaining the same or similar performance across the majority of use cases.”
Enterprise customers are still a minority on vibe coding platforms, but they represent a growing share of revenue. And cost pressure is no longer just an enterprise concern. Everyday users are also starting to think about what AI tools actually cost them. Base44’s model development was driven by several factors, but a lower cost per interaction is likely one of the outcomes Shlomo is counting on.
“We want to get a model that is going to be more aligned to what we think is the right thing, is going to be more optimized to what we see users like in terms of the results we’re getting, and is going to be faster and cheaper for customers eventually than using the frontier models like Opus,” he said.
For Base44’s parent company, the margin angle matters too. Wix recently announced it would cut 20% of its workforce. Base44, by contrast, has been growing its headcount since the acquisition. The company’s own statement on Base1 noted that owning the model gives it “direct control over compute and inference spend, expected to result in a structurally stronger margin profile over time.” That phrase does a lot of work, but the direction is clear: better margins, eventually.
Shlomo describes the engineering effort behind Base1 as significant. The payoff, in his view, is that Base44 becomes the only vertically integrated vibe coding platform, meaning it owns its distribution, its data, and its model stack all at once. Whether that combination is enough to hold off both well-funded rivals and frontier labs expanding into app building is the real question. For now, Base44 has at least made a bet worth watching.
