Base44 Shifts Toward Vertical Integration With Launch of Proprietary AI Model
Table of Contents
The Pivot to Proprietary Intelligence
Base44, the Tel Aviv-based “vibe coding” platform acquired by Wix for $80 million just a year after its inception, is attempting to solve one of the most pressing dilemmas facing the current generation of AI startups: the problem of defensibility. The company has begun rolling out Base1, its first proprietary large language model (LLM) designed specifically to help users build applications using natural language.
For the past year, like many in the “applied AI” space, Base44 relied on frontier models—the massive, general-purpose LLMs provided by companies like Anthropic or OpenAI. While these models offer immense power, they create a precarious dependency. If a foundational provider updates their API or launches a native feature that mimics a startup’s core value proposition, the startup’s moat evaporates instantly. By training Base1 on tens of millions of real user interactions, Base44 is betting that a vertically integrated stack—owning the data, the distribution, and the infrastructure—is the only way to survive long-term.
The Battle Against ‘Frontier’ Dominance
The move is a direct challenge to the prevailing industry trend where startups act as thin wrappers around external APIs. Maor Shlomo, founder of Base44, argues that specialization is the key to outperforming generalists. According to Shlomo, owning the model allows for aggressive optimizations in latency, cost, and efficiency that are simply impossible when renting compute from a third party.
This strategy puts Base44 in a distinct competitive position compared to rivals like the Swedish startup Lovable. While Lovable has seen explosive growth—reaching unicorn status in its Series A and reporting $500 million in annual recurring revenue (ARR) earlier this month—it continues to rely on external LLMs. Base44, which reported crossing $150 million in ARR in May, is positioning itself as the more sustainable, integrated alternative.
The Economics of Inference
Beyond the strategic moat, there is a stark financial reality driving this shift: the cost of inference. As AI features move from novelty to enterprise-grade tools, the cost of querying high-end models like Claude Opus can become prohibitively expensive, eroding profit margins.
Jonathan Userovici, a general partner at VC firm Headline, notes that enterprise customers are increasingly demanding a better return on investment (ROI). He suggests that the industry is moving toward a model of orchestration, where the “right” model is selected for the specific task to prevent costs from skyrocketing. For Base44, Base1 is that optimized choice. The company claims that direct control over compute and inference spend will result in a structurally stronger margin profile over time.
A Risky Engineering Bet
Despite the promising ARR growth, the path to proprietary AI is fraught with risk. The engineering effort required to maintain a custom LLM is massive, and the success of such a venture depends entirely on the quality of the feedback loop. Base44 is banking on its vast dataset of user interactions to provide the “edge” that general models lack.
However, the competition isn’t just other vibe-coding startups. The frontier labs themselves are encroaching on this territory. With tools like Claude Code and the integration of Cursor and Grok under the SpaceX/xAI umbrella, the providers of the foundation models are now building the very application layers that Base44 occupies. Shlomo remains optimistic, however, predicting that frontier models will always remain too general to match the precision of a specialized tool.
This strategic pivot arrives at a complex time for the broader organization. While Base44 continues to scale its headcount and revenue, its parent company, Wix, recently announced a 20% workforce reduction, highlighting the tension between the lean, high-growth AI sector and the stabilizing needs of established tech giants.
