MosaicML Launches AI Services at Lower Prices than OpenAI and Anthropic

JAKARTA - Artificial intelligence (AI) startup MosaicML founded by former Intel executives and academic researchers released two new products on Wednesday, May 3 with the aim of beating industry giants like OpenAI in terms of price.

San Francisco-based MosaicML, which has raised $64 million to date, is launching in 2021 with a suite of software tools designed to make artificial intelligence work less costly, which often involves training AI algorithms on trove big data using expensive computer chips. The company makes money selling the tool to companies that want to develop their own AI systems.

But on Wednesday, MosaicML announced several new services that will compete more directly with companies like OpenAI or Anthropic, which build AI systems and then charge access to them.

MosaicML said it will offer inference services to software developers looking to add features to their applications such as the ability to read and respond to text or generate images from a prompt.

MosaicML offers this service for a claimed fee that is 15 times lower than competing services.

"We understand how to make these things highly optimized. And so do they," Naveen Rao, CEO of MosaicML, told Reuters. "There's a lot of margin building on this, and at 15 times less, we're still making money on it."

Behind this new service, MosaicML has developed its own underlying model, which is the core technology category behind services such as Microsoft Corp's chatbot-enabled products and Alphabet Inc's Google.

Like many of its competitors, MosaicML will lease its technology to customers, but unlike most of them, the company will also provide customers with their code to run on their own hardware, so that MosaicML never sees the data. "Because so much of the value of an AI system comes from the data used for training, many enterprise customers want to do it," said Rao.

"People buy from us because we can do things on their personal data. They have assurance that the models built on personal data are owned by them," said Rao.