← Sticker Wall AGENT · DETAIL

In-house Model Cost Reduction: Base44 Surpasses $200M ARR Post-Acquisition, Targeting 60% Gross Margin

Workflow: The daily workflow is as follows: users describe their needs in natural language (e.g., "a bakery management system with

AGENT

Key Fields

FIELD STAMPS

IndustryFintech

RegionGlobal（以色列）

ScaleSME

ChannelOnline

🔧 Workflow

The daily workflow is as follows: users describe their needs in natural language (e.g., "a bakery management system with inventory management and reminders"). The AI agent automatically generates the database, front-end interface, login authentication, and back-end logic, and completes the deployment. The Workflows module uses triggers such as schedules, data changes, webhooks, and payments to connect the application to real-world business, allowing AI to interact with real-world orders and payment streams. The founder acts as a human referee, reviewing user feedback and inference cost bills, and gradually migrating high-frequency paths to the in-house Base 1 model. The input is the requirement description, and the output is a runnable full-stack application with month-over-month declining unit costs.

🛠 Setup Requirements

Requires full-stack prototyping capabilities: knowing how to use Claude and Gemini APIs to build a generation pipeline from "natural language to deployable application," along with an understanding of model routing, distillation, and cost attribution. Founder Maor Shlomo himself has a background at data company Explorium. For tools, prepare a vLLM inference cluster and a cost observability platform like Langfuse, coupled with application hosting and database services. Going from a prompt prototype to a launchable MVP takes about 6 to 8 weeks.

🧰 Toolchain

🔧 Claude API
🔧 Gemini API
🔧 vLLM Inference Deployment
🔧 Langfuse Cost Observability

💰 Revenue

In May 2025, prior to the acquisition, monthly profit reached $189,000 with over 1,300 paying customers, resulting in an estimated ARR of about $3.5 million. Following the all-cash acquisition by Wix, the 2026 ARR surpassed $200 million, translating to a monthly revenue of approximately $17 million (based on Wix's consolidated financial reporting). Founder Maor Shlomo has cumulatively received over $150 million, including $38 million in performance earnouts in Q1 2026 alone.

💸 Cost

Prior to the acquisition, approximately 89% of costs consisted of LLM token usage fees, primarily paid to Claude and Gemini, which was the core motivation for developing the in-house Base 1 model. After launching Base 1 in 2026, the target is to raise the gross margin to around 60%, significantly driving down inference costs. The absolute monthly cost figures have not been publicly disclosed by the official channels.

⏱ Time Investment

During the first 6 months of the startup, the founder worked almost around the clock, waking up every two to three hours to check servers. For a replicable solo version, it is recommended to invest 4 to 6 hours daily, focusing on two dashboards: inference cost billing and user feedback.

🚀 Getting Started

Step 1: Validate the minimal closed-loop of "one-sentence generation of deployable applications," output detailed token consumption breakdowns for every complete generation, and calculate the gross margin per application. Step 2: Use cost observability tools to attribute costs by prompt path, routing the most expensive routes to cheaper smaller models or distilled models. Step 3: Build in public on LinkedIn and X to drive viral growth, using human feedback as the iteration referee.

🔑 Keys to Success

✅ Treat inference cost as a first-principles problem: calculate unit application gross margin before discussing growth.
✅ Build in public combined with human customer-service-style feedback for zero-marketing customer acquisition, using humans as iteration referees.
✅ Proprietary models and distillation form a scalable moat, carving out about 60% gross margin space from nearly 90% token costs.
✅ Use Workflows to connect generative applications into real business flows such as payments and webhooks, ensuring AI generates real-world orders rather than remaining on demo pages.
✅ Leverage giant distribution channels as a multiplier: post-acquisition by Wix, ARR scaled from about $3.5 million to over $200 million within half a year.

⚠️ 风险

⚠️ High investment and long cycles for in-house models; if Claude and Gemini continue to drop prices, the profit window for in-house inference may narrow, flattening the moat via price wars.
⚠️ Post-acquisition traffic relies heavily on the Wix ecosystem; independent branding and pricing power may be diluted.
⚠️ Under a free tier plus credits billing model, low-quality applications generated by users may pile up inference costs. The gross margin target is sensitive to retention and unit economics, and once growth slows down, earnout pressures will correspondingly amplify.

📌 Real Cases

📌 Base44 (Israel): Founder Maor Shlomo launched an AI website builder as a personal side project in February 2025. Within 3 weeks of launch, it gained 10,000 users, reaching about 250,000 users and over 1,300 paying customers in 6 months. In June 2025, it was acquired by Wix for $80 million in cash, plus a $25 million retention bonus. After releasing its proprietary model Base 1 in 2026, ARR surpassed $200 million, and the founder's cumulative cash-in exceeded $150 million.
📌 Bakery Scenario (Official Documentation Example): A user describes in one sentence, "a bakery management system with inventory management and reminders." Base44 automatically stitches together the database, authentication, and backend functions within the workflow, and uses scheduled triggers to remind the store owner when inventory falls below a threshold, validating the instant availability of personal business tools.
📌 Nonprofit Volunteer Scenario (Founder's Account): Maor Shlomo discovered during volunteer work that building a simple system for a charitable organization was quoted externally at a million and took months, which led him to create a platform allowing everyone to "create software with chat," becoming the product's entry point.

Sources