Base44, the natural-language app-creation platform acquired by Wix for $80 million in June 2025 when it was roughly six months old with eight employees, has launched its own proprietary large language model: Base1. The model was trained on tens of millions of real user interactions on the Base44 platform — interaction data that no external AI lab possesses or can replicate.
Founder Maor Shlomo framed the move in explicitly economic and architectural terms. Rather than continuing to route every inference call through a third-party frontier API, Base44 now owns the full stack from user prompt to deployed app. The strategic rationale: domain-specialized training data, lower latency, reduced compute cost, and optimization leverage that generic frontier models cannot offer for this specific task surface.
The launch lands at a revealing competitive moment. Rival Lovable is reported at approximately $500 million ARR — roughly five times Base44’s current scale. Base44’s answer is not to match Lovable dollar-for-dollar on marketing or distribution, but to structurally differentiate at the model layer. That is a different kind of race entirely, and a bet that vertical integration compounds faster than top-line growth alone.
Base44 — Vertical Integration Timeline
Late 2024 — Launch
Base44 founded by Maor Shlomo; vibe-coding platform ships with ~8-person team, built on frontier APIs.
June 2025 — Wix Acquisition
Wix acquires Base44 for $80M at roughly 6 months old. Wix gains distribution; Base44 gains runway to build proprietary infrastructure.
2025–2026 — Data Accumulation
Platform scales to ~$100M ARR. Tens of millions of real user app-creation interactions accumulate as proprietary training corpus.
June 2026 — Base1 Launch
Base44 launches Base1: a proprietary LLM trained on its own platform data. Full vertical integration across data, model, and distribution.
The key insight: Base44 did not train Base1 because it wanted to be an AI lab. It trained Base1 because the moment you own a vertical with enough usage data, renting inference from a frontier API becomes both a margin tax and a ceiling on product quality. Base1 is a cost-reduction play disguised as a capability announcement.
The Structural Read
For two years, the venture consensus on AI application companies was “distribution beats model.” Build fast, acquire users, and let OpenAI, Anthropic, or Google absorb the R&D cost of the underlying intelligence. The strategy worked — Base44 hit ~$100M ARR, Lovable reached ~$500M ARR, and a dozen others scaled on commodity inference. The wrapper-to-scale playbook was real.
But the playbook has a structural ceiling. When your core product is AI-generated output and the model powering it is available to every competitor at the same API price, differentiation collapses to UX and brand — fragile moats in a market where UX can be cloned in weeks and brand has a short half-life. Base44 recognized this. Base1 is the structural response.
The deeper dynamic is what the Map of AI framework identifies as layer compression: companies that span multiple layers of the AI stack simultaneously — data, model, application, distribution — create compounding advantages that single-layer players cannot match. Base44 now touches all four. Wix’s acquisition, which looked like a talent-and-product grab in 2025, now reads as the financing event that enabled a full-stack build. The $80M purchase price may prove to be the cheapest foundation for a durable AI applications business of this generation.
Maor Shlomo, Base44 Founder — via TechCrunch
“Training and owning the model as part of our entire stack allows us a lot more optimizations on latency, cost, and efficiency.”
Map of AI — Layer Compression
The Wrapper Escape Sequence
Phase 1: Build on frontier APIs to acquire users fast. Phase 2: Accumulate proprietary usage data at scale. Phase 3: Train a vertical model on that data. Phase 4: Use cost and latency advantages to widen the product gap over competitors still in Phase 1. Base44 has now entered Phase 4. Most of its rivals are still in Phase 1 or 2.
Three Implications
IMPLICATION 1 — THE INFERENCE COST WAR IS NOW VERTICAL
Generic frontier model prices have fallen dramatically, but per-call costs still aggregate into meaningful margin drag at $100M+ ARR scale. A domain-trained model like Base1 can run more efficiently on narrower tasks — app-code generation has a constrained output distribution compared to general reasoning. Every percentage point of inference cost savings flows directly to gross margin, and gross margin is what funds the next round of model improvement. This is a self-reinforcing loop that API-dependent rivals cannot enter.
IMPLICATION 2 — LOVABLE’S SCALE LEAD IS MORE FRAGILE THAN IT LOOKS
At ~$500M ARR, Lovable has a 5x revenue lead over Base44. But if both platforms currently deliver similar output quality via similar underlying APIs, Lovable’s lead is a distribution and brand advantage — not a product advantage. If Base1 produces meaningfully better app-creation outputs at lower cost, that gap can close faster than the revenue numbers suggest. The question is whether Lovable moves to vertical model ownership before Base44 proves the thesis. In AI, the second mover on infrastructure can leapfrog quickly.
IMPLICATION 3 — WIX JUST BECAME AN AI INFRASTRUCTURE COMPANY
The $80M Wix paid for Base44 in June 2025 now looks less like a product acquisition and more like the foundational capital expenditure for Wix’s AI infrastructure layer. If Base1 extends to power Wix’s broader product suite — site creation, e-commerce flows, content generation — the ROI calculation changes entirely. Wix is no longer just a website builder with an AI feature; it is a company with a proprietary model trained on one of the world’s largest real-world app-creation datasets. That is a materially different competitive position in the website and SaaS creation market.
Business Engineer Framework
Map of AI — Where Does Base44 Sit in the Stack?
The Map of AI tracks 200+ companies across 9 layers of the AI stack — from chips and foundation models through infrastructure, tooling, and application layers. Base44’s Base1 launch is a textbook example of upward layer compression: an application-layer company acquiring model-layer capabilities to lock in structural advantages. Understanding which layer a company occupies — and which layers it is moving into — is the single most useful analytical lens for evaluating AI competitive dynamics in 2026.
Base44 and Base1 are the clearest proof yet that the applied AI playbook has entered a new phase: companies that used frontier APIs to acquire users at speed are now turning that usage data into proprietary model advantages that frontier API providers cannot sell back to them. The “wrapper” era is not ending — it is graduating. The winners will be the ones who recognized early that scale without data ownership is a rented moat, and moved to own the stack before the window closed. Base44 just moved.
Sources: TechCrunch — Base44 Base1 launch reporting, June 2026; Wix acquisition details and ARR figures via TechCrunch. Competitive ARR figures (Lovable ~$500M, Base44 ~$100M) as reported by TechCrunch. Analysis and framework application by FourWeekMBA / Business Engineer.
91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
