NVIDIA has gained Gretel general data company for nine numbers, according to two people with direct knowledge about this agreement.
Gretel's most recent buying price is $ 320 million, the sources said that although the exact terms of the purchase were not known. Gretel and the group of about 80 employees will be ranked in NVIDIA, where its technology will be deployed as part of the AI Service Development Service based on clouds, cloud -based for developers.
The acquisition takes place when NVIDIA has deployed general data creation tools, so that developers can train their own AI models and tweak them for specific applications. Theoretically, synthetic data can create a source of training data AI near the term and help solve the problem of data scarcity that has appeared in the AI industry since chatgpt by 2022, although experts say that using synthetic data in AI generation.
Nvidia spokesman refused to comment.
Gretel was founded in 2019 by Alex Watson, John Myers and Ali Golshan, who also worked as a CEO. Startup provides a synthetic data platform and API sets for developers who want to build generation AI models, but do not have access to enough training data or have a concern about privacy around the use of people's data. Gretel did not build and licensed their own AI Frontier models, but refined existing source models to add different security and safety features, then the packages together to sell them. The company has mobilized more than $ 67 million in advanced money to sponsor venture capital before buying, according to Pitchbook.
Gretel spokesman also refused to comment.
Unlike human data created or in the real world, synthetic data is created and designed to imitate data in the real world. Proposals say this makes the data creation needed to build AI models can expand, less labor and more accessible to smaller or less resources developers. Privacy protection is another important sales point of synthetic data, making it an attractive option for health care providers, banks and government agencies.
NVIDIA has provided general data tools to developers for many years. In 2022, it launched Omniverse Replicator, providing developers to create custom 3D data, physically accurate to train nerve networks. Last June, NVIDIA started launching a family of AI models to create general training data for developers to use in building or tweaking LLM. Called Nemotron-4 340B, these mini models can be used by developers to drumming general data for their own LLM on the basis of health, financial, production, retail and all other industries.
In his main presentation at NVIDIA's annual developer on Tuesday, the co -founder of Nvidia and CEO Jensen Huang talked about the challenges that the industry faced in quickly expanding AI in an effective way.
There are three issues that we focus on, he said. One person, how do you solve data problems? How and where do you create the data needed to train? Two, what is model architecture? And then three, what is the law? Huang continues to describe how the company is currently using to create synthetic data in its robotic platforms.