Datagen

The problem: labeled data is expensive

Datagen was founded in 2018 by Ofir Zuk (Chakon) and Gil Elbaz, both graduates of the Technion. It operated in Tel Aviv and was built on a simple fact: a computer vision system learns from labeled images, and sometimes it needs tens or even hundreds of millions of them. The tech giants hold data generated by their users, but other companies have to take the photos themselves or buy data, and both ways are expensive and slow. In a 2019 interview, Chakon described an example: a startup that wants to recognize passenger behavior inside a car has to bring a hundred different people into a car and stage scenarios in front of them, just to collect the training material.

The solution: a 3D simulator

Instead of taking photos, the company built an engine that produces synthetic data. It created a 3D simulation of reality and generated images from it that come already labeled, because the system knows exactly what is in every pixel. According to Chakon, the engine takes requirements in on one side and puts data out on the other, and can vary lighting, people and environments. Beyond regular images, it produced depth maps, matching between objects in different images and simulated infrared imaging. The customers described came from virtual and augmented reality, security and smart stores.

The peak: a $50 million round

In March 2022 the company completed a $50 million Series B round led by Scale Venture Partners, with participation from TLV Partners, Viola Ventures and Spider Capital, bringing its total funding to $70 million. The company reported at the time an eightfold increase in revenue within a year, and that three of the world's five largest tech giants use its product. These are figures the company itself provided, as is common in funding announcements. In May 2023 Datagen still employed 110 people.

The blow: generative AI

According to an August 2023 report, the rise of ChatGPT, Bard and other generative AI tools made the company's solution less relevant. As a result the company laid off many employees and negotiated a possible acquisition with Meta, which did not come through. Within three months almost all the remaining employees were laid off, leaving a team of ten people looking for a new business direction. This was a surprising point in the story: generative AI, which is itself a way of producing synthetic content, competed directly with the company's method of producing training images.

The shutdown and the question that remains open

In July 2024 TheMarker reported that the company's CEO had officially announced its closure, after attempts to change the business failed. According to the report, about $20 million was left in the bank, and people in the market struggled to understand the founders' decision a year earlier. The Datagen case does not mean that synthetic data failed as an idea; the field continues to develop. It shows how quickly a technological breakthrough can erode a young company's advantage, even when it has money in the bank, large customers and growing revenue.