Nvidia relies large on synthetic data

Nvidia has acquired a Gretel synthetic data company for nine figures, according to two people with direct knowledge of the deal.

The cost of the acquisition exceeds Gretel’s most rating of $ 320 million, sources say, although the exact terms of the purchase remain unknown. Gretel and his team of approximately 80 employees will be folded in NVIDIA, where its technology will be located as part of a growing set of cloud, generative services based in cloud, generative AI services.

The acquisition comes as NVIDIA introduces synthetic data generation tools so that developers can train their own AI models and refine them for specific applications. In theory, synthetic data can create an almost infinitive offering of AI training data and help solve the problem of data shortages that emerges above the AI ​​industry, as Chatgpt went to the basis of 2022, although experts say that the use of synthetic data in generative AI comes with its own.

A spokesman for Nvidia declined to comment.

Gretel was founded in 2019 by Alex Watson, John Myers and Ali Golshan, who is also CEO. The launch offers a synthetic data platform and a set of APIs for developers who want to build generative models of AI but do not have access to sufficient training data or have concerns about confidentiality of using real people data. GRETEL does not build and license its own Frontier AI models, but refinement of existing open source models for adding differential confidentiality and safety features, then packs these together to sell them. The company has raised over $ 67 million at risk funding before the acquisition, according to Pitchbook.

A spokesman for Grete also declined to comment.

Unlike human or real -world data, synthetic data are generated computer and intended to mimic real -world data. Proponents say that this makes the data generating needed to build AI models larger, less laborious and more accessible to smaller or less resource AI developers. Privacy is another key point for selling synthetic data, which makes it an attractive option for healthcare providers, banks and government agencies.

NVIDIA has already offered synthetic data for developers for years. In 2022, he launched Omniverse Replicator, which enables developers to generate personalized, physically accurate, synthetic 3D data on neural networks. Last June, NVIDIA began to unfold a family of open AI models that generate data on synthetic training for developers to use in the construction or fine -tuning of LLM. Called the Nemotron-4 340B, these mini-models can be used by drum developers synthetic data on their own LLM in Health Care, Finance, Production, Retail and any other industry.

During his main presentation at the NVIDIA Annual Conference on NVIDIA this Tuesday, the co -founder of NVIDIA and CEO Jenson Huang talks about the challenges that the industry is in the rapidly scathing AI in a profitable manner.

“There are three problems we focus on,” he said. “One, how do you solve the problem of data?