Oracle teams up with Cohere on generative AI cloud service

Every major cloud provider has spent the past few months watching Microsoft and OpenAI run away with enterprise AI. Now, Oracle is mounting its counterattack.

During Oracle's Q4 earnings call, founder and CTO Larry Ellison announced an upcoming enterprise generative AI service in partnership with AI unicorn Cohere. Think of it as Oracle's answer to the Azure OpenAI service, designed specifically to help businesses train and customize large language models on their private data without leaking secrets to public systems.

Teaming up with Cohere

The move follows Oracle's lead investment in Cohere's $270 million Series C round alongside Nvidia and Salesforce, which valued the Toronto-based startup at roughly $2.2 billion. As part of the deeper tie-up, Cohere relies heavily on Oracle Cloud Infrastructure (OCI) to train its models.

Ellison emphasized that data privacy is the whole pitch. Corporate legal teams are notoriously wary of public models scraping proprietary knowledge, so Oracle is building a dedicated sandbox where companies can train specialized internal models safely.

Oracle is already eating its own cooking. Using proprietary data from Cerner, the healthcare IT giant it bought last year, Oracle has tuned two specialized models: one for clinicians and another for first responders.

The secret weapon: spare GPUs

While the model news grabbed headlines, Oracle's balance sheet showed real momentum in AI infrastructure. OCI revenue surged 76% year-over-year to $1.4 billion. Ellison claimed that generative AI cloud customers recently signed contracts to purchase over $2 billion in capacity on Oracle's Gen2 Cloud, with Adept AI and Nvidia also tapping OCI clusters.

Wall Street cheered the print, driving the stock up past 3% after hours. Wolfe Research analyst Alex Zukin upgraded Oracle from hold to buy, pointing to a critical structural advantage: spare GPU capacity. When every AI lab on earth is starved for compute, simply having clusters ready to turn on is half the battle.

Constellation Research analyst Holger Mueller noted that Oracle's capital expenditure exceeded 50% of its cash flow over the last two quarters to fund this infrastructure buildout. Still, geographic friction remains: almost all of this surge came from the Americas, while revenue in Asia Pacific actually dipped.

Why it matters

For enterprise tech leaders, this is the first credible enterprise-grade alternative to Azure OpenAI. If your critical business records and ERP already sit inside Oracle databases, keeping your proprietary data within the same secure perimeter to fine-tune Cohere's models eliminates enormous compliance and egress headaches. It is worth evaluating immediately if your legal team flatly banned commercial LLMs.

For the broader industry, Oracle is cementing its position as the preferred neutral utility provider for AI startups. By backing Cohere rather than training its own frontier models from scratch, Oracle avoids an expensive arms race while securing long-term compute commitments.

We think Ellison's playbook here is remarkably practical. Instead of trying to out-research OpenAI or Google, Oracle leveraged its enterprise trust and physical data centers to monetize the gold rush. The real vulnerability down the road is what happens when the global GPU shortage eases. If AWS and Google catch up on compute availability, Oracle will have to prove that Cohere's enterprise models are sticky enough to keep workloads from migrating back to the incumbent cloud giants.