By ByteBot
Databricks closed a $5 billion funding round yesterday at a $190 billion valuation — and buried beneath the headline number is a more interesting story: CEO Ali Ghodsi thinks AGI has already arrived, the model race is done, and the companies that win the next five years are the ones that own the data context underneath AI agents.
The Raise That Almost Wasn’t
Databricks originally planned to raise $1 billion. Then The Information published a story about the round while Ghodsi’s team was in the middle of a conference. His phone blew up. Investor interest ballooned to $15 billion from selected investors alone, according to TechCrunch. They settled on five.
The company turned away more than $10 billion in would-be capital. That tells you something about where the market thinks enterprise AI infrastructure is heading. The round values Databricks at roughly 27x its $7 billion annualized revenue run-rate, which is growing at over 80% year-over-year and generating positive free cash flow. Coatue led, with Blackstone, MGX, Andreessen Horowitz, and about two dozen others participating.
Ghodsi’s AGI Claim Is Actually a Business Thesis
At Databricks’ Data + AI Summit in June — 31,000 attendees from 174 countries — Ghodsi polled the crowd: “Has AGI arrived?” About 90% said no. His response: you’re wrong.
However, he’s using a specific definition: systems that perform intellectual tasks at human-level or better most of the time. He argues that today’s frontier models already clear that bar as the industry defined it before 2022. But the provocation isn’t the real point. The conclusion is: “The model race is over. The context race has begun.”
If model intelligence is roughly commoditized, competitive advantage shifts to whoever connects AI to an organization’s actual data — records, rules, workflows, institutional memory. That’s Databricks’ entire product pitch, and it explains why they’re building Lakebase, Genie, and Unity AI Gateway simultaneously rather than racing to ship a better model.
Lakebase: Postgres Built for AI Agents
The most technically interesting product in the announcement is Lakebase — a serverless Postgres database engineered specifically for AI agent workloads. It’s built on Neon, the serverless Postgres startup Databricks acquired in May 2025 for over $1 billion.
The core innovation is database branching. An agent can spin up an isolated copy of a production database — at petabyte scale — in about one second, run experiments, fail safely, and discard the branch when done. Branches use copy-on-write storage, consuming space only for the bytes that change. Neon can spin up a full Postgres instance in under 500 milliseconds. Moreover, more than 80% of databases provisioned on Neon are created by AI agents rather than humans — a striking signal about how quickly this usage pattern is changing.
Lakebase is already at $100 million in annualized revenue — roughly 12 to 14 months after launch — and is generally available on Azure Databricks. The Lakehouse Sync feature automatically replicates agent scratchpads and tool outputs into Unity Catalog as Delta tables, eliminating the ETL pipelines that typically sit between operational and analytical systems.
Unity AI Gateway: The CFO Problem, Addressed
Ghodsi is direct about what’s driving Unity AI Gateway adoption: “Token maxing has freaked out CFOs.” As inference costs compound across hundreds of agents and thousands of users, enterprises need a control plane that can route tasks to cheaper models, enforce hard spend caps, and attribute costs by team and use case.
Unity AI Gateway went generally available on August 4, ten days before this funding announcement. Its Smart Routing feature claims 30%+ cost reduction per task by matching workload complexity to model capability rather than defaulting to the most expensive option. Furthermore, over a quadrillion tokens have already passed through the gateway, with Rivian, Asana, and Edmunds among the enterprise customers already live on the platform.
What Developers Should Watch
The $190 billion number is eye-catching, but the product roadmap tells you more about what Databricks actually believes. They’re building up the stack from analytics (Lakehouse) to transactional databases (Lakebase) to agent orchestration and governance (Unity AI Gateway and Genie). Every layer assumes your AI deployment will eventually involve agents running at scale, writing to databases, calling external models, and consuming enough tokens to concern your finance team.
Whether Ghodsi is right about AGI is almost beside the point. The bet is that the data layer matters more than the model layer going forward — and given that leading models are increasingly interchangeable on standard benchmarks, he has a reasonable case. The $5 billion raise — and the $10 billion in demand they turned away — suggests investors agree.
