The Shift Toward Vertical Integration

Base44, a "vibe coding" platform acquired by Wix, has launched its own custom LLM, "Base1." This move marks a strategic pivot away from relying solely on frontier models, aiming to achieve a fully vertically integrated stack. By owning the model, the company gains direct control over its infrastructure, which founder Maor Shlomo argues is essential for optimizing latency, efficiency, and long-term cost structures.

Defensibility Through Data and Infrastructure

In the competitive landscape of AI-powered development tools, defensibility is increasingly tied to three pillars: distribution, tech stack, and proprietary data. Base44’s model was trained on tens of millions of real user interactions, creating a feedback loop that the company believes will provide a competitive edge over general-purpose models. While competitors like Lovable currently rely on external LLMs, the industry trend suggests that companies with sufficient scale and velocity will eventually move toward proprietary or specialized models to maintain margins and product differentiation.

Managing Inference Costs and Performance

Beyond defensibility, the move is a response to the growing pressure of inference costs. Enterprise customers are increasingly scrutinizing the ROI of AI features, often finding that general-purpose frontier models are overkill for specific tasks. By training a specialized model, Base44 aims to deliver a product that is faster and more cost-effective for its specific use cases. While some industry experts caution that training custom models is a massive engineering undertaking—citing examples like Harvey that abandoned similar plans—Base44 is betting that vertical integration is the only way to secure a sustainable, high-margin business model in the long term.