Arago, the Photonic Processor for AI

While computing power is essential in the race for generative AI, the startup Arago aims to transform data centers. Its vision: photonic chips. These chips use lasers instead of the transistors found in the graphics cards currently used to train and run the latest models. They rely on photons—particles of light—rather than electrons.

According to Arago, this approach would reduce energy consumption by a factor of ten—or even more—compared to graphics cards, a major advantage in lowering AI costs. In addition, these chips will remain compatible with existing software and infrastructure. Founded in the summer of 2024, the startup has just raised $26 million to continue R&D and launch its first photonic processor in the coming months.

Dottxt, the AI that talks to computers

While generative artificial intelligence can mimic human syntax, it has not yet learned computer syntax. Dottxt aims to address this limitation. Its solution: structured generation. The French startup explains that this technique acts as a kind of spell-checker, forcing large language models to produce results that conform to computer languages.

Launched a year ago, dottxt offers an open-source library that has already been downloaded more than three million times on GitHub, the platform used by developers. OpenAI and Cohere are reportedly among its first users. Building on its early success, the startup raised 12 million euros from investors during two funding rounds held a few months apart. Its executives hope to drive adoption of their technology before beginning to market a paid solution for businesses.