Oracle said to weigh acquisition of Cohere in enterprise AI push

Oracle is exploring a purchase of Cohere, according to a report published Wednesday that cited people familiar with the talks, in a move that would fold one of the leading enterprise AI model providers into Oracle Cloud Infrastructure. The Information reported that discussions in recent weeks have centered on a price above 8 billion dollars, and that terms could still change. Oracle and Cohere did not immediately comment.

The potential deal matters because Oracle is trying to narrow the AI infrastructure gap with Microsoft Azure and Google Cloud while locking in enterprise demand for foundation models that run on OCI. Synergy Research said in 2024 that AWS held about 31 percent of global cloud infrastructure spend, Microsoft 25 percent, Google 11 percent, and Oracle about 3 percent.

Cohere, led by chief executive Aidan Gomez, has focused on controllable large language models for business use, including the Command and Embed families, and retrieval augmented generation tooling. The company raised 270 million dollars in 2023 at a roughly 2.2 billion dollar valuation, with backers including Nvidia, Oracle, Salesforce Ventures, and Inovia Capital, and it was reported in 2024 to be seeking additional funds at a higher valuation.

An acquisition would deepen Oracle’s existing partnership with Cohere, whose models are already available on OCI and through Oracle’s data stack. It could also create friction with other distributors, since Amazon’s Bedrock service lists Cohere models alongside Anthropic, Meta, and Mistral.

Any transaction involving Cohere, which is headquartered in Toronto and has operations in the United States, would be reviewed under Canada’s Investment Canada Act. Depending on deal structure, U.S. agencies including the Federal Trade Commission could examine competitive effects in enterprise AI and cloud infrastructure.

Oracle has been investing heavily to meet AI compute demand. In its fiscal fourth quarter 2024, Oracle reported that cloud infrastructure revenue rose 42 percent year over year to about 2.0 billion dollars, and executives said the company had signed multi year capacity commitments linked to Nvidia GPU clusters. Chair and chief technology officer Larry Ellison has positioned OCI’s RDMA networking and data locality as differentiators for AI training and inference.

If Oracle brings model IP in house, it would intensify vertical integration across the AI stack, from GPU capacity to model development to enterprise applications. That would signal further consolidation between model firms and hyperscalers, a trend shaping how businesses buy AI capabilities and how regulators define market power in cloud services.