The new round and the valuation
Databricks has closed a $5 billion round that raises its valuation to $190 billion, according to AI News Italia. The deal, announced on August 13 by the San Francisco company, was led by Coatue alongside Blackstone, MGX, T. Rowe Price and Sixth Street Growth. Six months ago the valuation was $134 billion: an increase of 42%.
The final figure exceeds the one circulating in July: on July 16 the company had indicated $188 billion, after signing a term sheet with Coatue. CEO Ali Ghodsi explained the difference from the higher amount raised and from the additional shares issued. Annualized revenue has surpassed $7 billion, with year-over-year growth of more than 80% in the second quarter.
Where the funds are going: three products
The capital is earmarked for three products. Lakebase is a serverless Postgres database designed for software written by agents and has already surpassed $100 million in annualized revenue. Genie is an agent that operates like a colleague: it reads emails, meeting recordings and internal data while maintaining security and privacy. Unity AI Gateway routes workloads among models and keeps spending under control.
The latter addresses a concrete problem described by the source: many companies deploy agents to write code and run them always on the most powerful models, even for trivial tasks, so inference costs end up exceeding the value of the work performed. A gateway that selects the right model for each request reduces the bill. The same issue emerges in analyses of the hidden costs of enterprise AI, which grow when pilot projects move into production.
The bet on the orchestration layer
Databricks is not betting on a single provider. OpenAI's GPT models and Codex, Anthropic's Claude—which arrived under a five-year agreement signed in 2025—and Google's Gemini run on the platform: the customer chooses. Investors are betting that the layer orchestrating agents will be worth as much as, or more than, the individual model makers.
Ghodsi described the market as “crazy” to CNBC, noting that everyone is using agents and the world is focused on them. He estimated that over the next nine to twelve months more software could be written than in all human history, and that every application will need a database, with agents wanting one that can start in an instant and shut down when idle.
IPO postponed and the cost issue
There is no date for the listing. Ghodsi repeated that the company wants to become public, but that for now there are too many distractions: as long as private capital remains abundant, the IPO can wait. The next test will come from the financials: if inference costs continue to rise, spending control may matter as much as model quality, and whoever governs traffic among models will become the provider on which companies depend.
