In recent months several companies that previously relied on third-party large language models (LLMs) have started deploying or developing their own proprietary models. The shift is driven by rising inference costs and the strategic risk that model providers are entering their customers’ markets.
Specific moves
- Base44, the app-building platform that Wix acquired for $80 million last year, has put its first proprietary model, Base One, into production. Base One is a fine-tuned model based on an open-source foundation and trained on tens of millions of real user interactions.
- MyClaw.ai, a notable agent-hosting platform, said it will soon ship its in-house model, MyClaw Pro.
- Cursor was the first to move; Base44 and MyClaw are following that lead.
Why this is happening now
The conventional playbook had been to rent the ‘‘engine’’. That approach has become less tenable because of two developments:
- Inference costs are eating into margins: using third-party models is an ongoing expense that becomes material at scale.
- Model providers are increasingly competing in the same markets as their customers, creating strategic conflict.
For these reasons, owning the model is shifting from a luxury to a baseline requirement: companies with sufficient scale and data want to control the layer they can no longer afford to rent.
The rationale and implications
Maor Shlomo, founder of Base44, expects that narrow, task-specialized models will outperform broad frontier models when it comes to building apps. The practical conclusion is that players with the necessary scale and data will likely follow suit to reduce long-term costs and gain better product control.
As more firms in a category move to build the core they used to buy, the market signal is clear: renting the core was never a durable moat, merely a lease that is now expiring.
One specific note: the sector’s current category leader, Lovable, with a reported valuation or position at $500 million, appears—based on the same reasoning—vulnerable to similar pressure to develop its own model.
What to watch next
- More companies may choose in-house models if they have sufficient scale and data.
- Beyond cost savings, owning the model increases strategic control and reduces dependence on external providers.
- However, building and maintaining proprietary models also requires substantial resources, so the shift is not an automatic fit for every player.
In short, rising costs and competitive risks are prompting many firms that once rented LLMs to invest in their own models to lower costs and retain strategic control over their products.
