Generative A.I. has emerged as one of the hottest areas for venture investments, seeing $12.37 billion in VC funding so far in 2023, according to PitchBook. In January, Microsoft invested $10 billion in OpenAI, which owns the popular A.I. tool ChatGPT. Microsoft is one of the big players in the space, along with some familiar rivals: Alphabet, Amazon, and Meta. Nvidia, which makes chips for computers that run A.I. models, is also positioning itself to be at the center of the boom—its shares have soared 189% since January.

With the consolidation of such power comes the danger that companies seeking to build their own A.I. models, with their own data, may be forced to rely on Big Tech. Training an A.I. model is computationally intensive, uses lots of electricity, and is expensive. Startups may have little recourse but to use Google, which has embedded its A.I. technology into 25 products, or Microsoft, which is developing its own A.I. chip, code-named “Athena,” The Information reported in April. Startups could lean on the traditional tech giants to build out their A.I.s, but it could be very expensive, and it would make them captive to someone else’s timelines, two venture capital executives said. Microsoft and Google could not be immediately reached for comment.

Ali Ghodsi, cofounder and CEO of Databricks, said he has met with several leaders of startups during the past six to seven months who want their own A.I. models. “The question recently is, Can they build their own models? Do they have the resources, or [are] the resources just reserved for companies with massive funding?” Ghodsi told Fortune.

Enter MosaicML, which is providing software that lets startups use their own data to train and deploy large language models as well as other generative A.I. That software also lets developers retain control over the models they build. Companies can save millions of dollars, as well as weeks to months, using MosaicML instead of a product from Google or Microsoft, the VCs said.

MosaicML began partnering with Databricks about a year ago, Ghodsi said. Founded in 2021, the startup has raised $64 million in funding from venture firms like DCVC, Lux Capital, and Playground Global. Customers include AI2 (Allen Institute for A.I.), Generally Intelligent, Hippocratic AI, Replit, and Scatter Lab.

Databricks began considering whether to buy MosaicML a few months ago, Ghodsi said. “It became clear that if we join forces, we can democratize A.I. even more.”

The two companies share a mutual respect and an aligned mission to free A.I., said one of the VCs familiar with the transaction. “This is not science fiction. MosaicML has demonstrated that they have the software solution to democratize A.I.”

On June 26, Databricks and MosaicML announced their stock deal, valued at $1.3 billion, which includes retention packages. All 62 of MosaicML’s employees are expected to stay. The transaction is slated to close in July, a spokeswoman said. Neither Databricks nor MosaicML used an outside investment bank on the deal.

The acquisition, however, is not an exit for MosaicML’s venture investors, the VC told Fortune. No investor is cashing out with the sale, and MosaicML’s backers are unanimously behind the future of the two companies together, the VC said. Venture investors will likely have to wait for Databricks to go public or get sold to realize a return. But if Databricks does exit, MosaicML’s earliest investors are on a path to achieve “a significant double-digit cash-on-cash return on their total investment,” the VC added.

Databricks, which describes itself on its website as “the data and A.I. company,” has raised $3.6 billion over 10 rounds, according to Crunchbase. The company was valued at $38 billion in 2021, and an IPO has long been expected, but a stock market downturn beginning last year has stifled those opportunities.

“The IPO market is shut right now,” another person familiar with the situation told Fortune, adding that Databricks would “definitely” consider an IPO once market conditions improve.