Wix’s Base44 Unlocks AI’s Next Frontier

Hustler Words – In a significant strategic pivot, Base44, the innovative "vibe coding" platform acquired by Wix for a staggering $80 million just a year ago, has unveiled its proprietary artificial intelligence model. This move, coming from a company that was barely six months old with a lean team of eight at the time of its acquisition, marks a bold step towards empowering users to craft applications using natural language, directly challenging prevailing industry norms.

The decision by the Bay Area-based Base44 arrives amidst escalating debates within AI circles regarding the universal applicability of generalized "frontier" models. A core question at the heart of this discussion is the long-term viability and defensibility of businesses constructed atop third-party AI infrastructure. Base44’s latest initiative directly addresses both concerns, signaling a potential shift in how specialized AI applications are developed and sustained.

While its custom large language model (LLM), dubbed Base1, is still in its initial rollout phase, Base44 harbors ambitions for it to eventually surpass the performance of more generalized frontier models. Maor Shlomo, the company’s founder, emphasized the strategic advantages, stating that "training and owning the model as part of [our] entire stack allows us a lot more optimizations on latency, cost, and efficiency." This integrated approach aims to deliver superior performance tailored specifically to Base44’s unique "vibe coding" environment.

This proactive stance could be interpreted as a preemptive strike against competitors, such as the Swedish unicorn Lovable, which achieved its impressive valuation last summer and currently relies on external LLMs. However, Shlomo anticipates that this trend of internal model development will become more widespread, particularly among "players that have gotten enough scale and velocity to have enough data."

Jonathan Userovici, a general partner at VC firm Headline, whose portfolio includes notable AI entities like Mistral AI, underscores the critical components of defensibility for AI startups: data, distribution, and the underlying tech stack. Base44’s strategy aligns perfectly with this framework. The company asserts that Base1 was developed and trained on a massive dataset comprising "tens of millions of real user interactions on the platform," providing a rich, proprietary foundation.

This invaluable dataset is poised for continuous growth alongside the company. Yet, the competitive landscape is not static. The most formidable challengers might not be other "vibe-coding" startups, but rather the frontier AI labs themselves, increasingly encroaching on Base44’s specialized territory. Companies like Cursor, Grok’s parent xAI (now part of SpaceX), and Claude Code are rapidly evolving into significant players in the app creation space, leveraging their foundational AI capabilities.

While these developments grant foundational AI providers access to crucial data and feedback loops for refining their app creation models, Shlomo remains confident in Base44’s specialized approach. He posits that despite advancements, general models "will stay very general in what they can do," leaving ample room for niche-focused solutions to excel.

Userovici, while acknowledging the power of frontier models (citing the legal tech startup Harvey’s decision to abandon its own model training efforts), frames Base44’s move within a broader economic context: the escalating costs of AI inference. He suggests that this cost pressure is driving a demand from enterprise customers for more efficient solutions. "They don’t necessarily see a [return on investment] when using the latest models for all use cases," Userovici explains, highlighting the emergence of infrastructure designed for "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."

Though enterprise clients currently represent a minority of "vibe coding" platform users, their share of platform revenue is expanding, and cost concerns are becoming universal across user segments. Base44’s decision to develop its own LLM was multi-faceted, but cost reduction undoubtedly stands as a significant benefit. Shlomo articulated the company’s ambition: "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."

For Base44 itself, the financial benefits, particularly improved margins, are a clear long-term goal. The company stated in a press release that "ownership of the model gives Base44 direct control over compute and inference spend, expected to result in a structurally stronger margin profile over time." This improved financial outlook would be welcome news for its parent company, Wix, which recently announced a 20% workforce reduction. In contrast, Base44 has experienced consistent headcount growth since its acquisition and recently announced it had surpassed $100 million in annual recurring revenue (ARR).

While this is still less than Lovable’s reported $500 million ARR earlier this month, Shlomo is betting that the "huge engineering effort" invested in Base1 will solidify Base44’s position as the "only vertically integrated vibe-coding application." In Userovici’s terms, this means Base44 would be a player that simultaneously owns its distribution, data, and core infrastructure, establishing a formidable and sustainable competitive advantage in the rapidly evolving AI landscape.