NVIDIA purchased a synthetic Gretel data company for nine numbers, according to two people with direct knowledge about the contract.
Sources say that the acquisition price exceeds the latest Gretel valuation of $ 320 million, although the exact purchase conditions remain unknown. Gretel and his team with about 80 employees will be submitted to NVIDIA, where his technology will be implemented as part of the growing package of the Giant Giant Giant in the cloud, generative AI services for programmers.
The acquisition appears when NVIDIA introduces synthetic data generation tools, thanks to which developers can train their own AI models and tune them to specific applications. Theoretically, synthetic data can create almost infinite supply of AI training data and support solve the problem of data shortage that appears over the AI industry, since ChatgPT has become the mainstream in 2022-even though experts say that the operate of synthetic data in the generalization of AI has its own risk.
Nvidia spokesman refused to comment.
Gretel was founded in 2019 by Alex Watson, John Myers and Ali Golshan, who also acts as a general director. The startup offers a synthetic data platform and API interfaces package for programmers who want to build generative AI models, but do not have access to a sufficient number of training data or have concerns about privacy related to the operate of real people. Gretel does not build and licens its own AI Frontier models, but refuses the existing Open Source models to add differently privacy and security functions, and then packs them together to sell them. According to Pitchbook, the company collected over $ 67 million in financing Venture Capital before taking over.
Gretel spokesman also refused to comment.
Unlike data generated by people or real data, synthetic data is generated by computer and designed to imitate real data. Supporters say that data generation required to build AI models is more scalable, less labor -intensive and more accessible to smaller or less separated AI programmers. Privacy is another key point of sale of synthetic data, which makes it an attractive option for healthcare providers, banks and government agencies.
NVIDIA has been offering synthetic data tools for programmers for years. In 2022, he launched the Omniverse replicator, which gives programmers the opportunity to generate non -standard, physically right, synthetic 3D data to train neural networks. In June last year, Nvidia began to introduce a family of open AI models that generate synthetic training data for programmers to operate in building or refineing LLM. These mini-models, called Nemotron-4 340B, can be used by programmers to download synthetic data for their own LLM in “healthcare, finance, production, retail sales and any other industry.”
During his main presentation at the annual NVIDIA developer conference on Tuesday, co -founder of Nvidia and general director of Jensen Huang spoke about the challenges that industry faces in the rapidly scaled artificial intelligence in a profitable way.
“We focus three problems,” he said. “First of all, how do you solve the problem with the data? How and where do you create the data necessary to train AI? Two, what is the architecture of the model? And then three, what are the provisions on scaling?” Huang described how the company now uses the generation of synthetic data on its robotics platforms.