Databricks hits $4.8B revenue run rate with 55% growth as Ali Ghodsi argues LLMs are already commoditized

Summary

Databricks has crossed $4.8 billion in annualized revenue, growing at over 55% year-over-year, with the milestone disclosed by CEO Ali Ghodsi in a late 2025 appearance. The growth is broad-based, with no single customer, cloud, or product driving it disproportionately. Two product lines have independently crossed the $1 billion run-rate threshold: its data warehousing product, launched four years ago, and its AI revenue segment, which now represents more than 25% of total revenue.

LLMs Are Already a Commodity

Ghodsi's central strategic argument is that large language models have already commoditized. He draws a direct analogy to gasoline: the underlying model matters less week to week than which gas station you use. What remains scarce, and therefore valuable, is the ability to run AI against proprietary enterprise data, whether electronic medical records, financial models, or internal operational data. That scarcity is what Databricks monetizes through its Lakehouse data foundation and Unity Catalog governance layer.

This framing positions the $2 billion of Databricks revenue classified as data management not as legacy infrastructure but as the prerequisite for enterprise AI deployment. Nearly all of Databricks' AI customers were already Lakehouse customers first, not by coincidence but because unstructured or siloed enterprise data renders commodity LLMs effectively useless.

Enterprise AI Use Cases Moving Beyond Pilots

Several live deployments illustrate where enterprise AI spend is landing. Merck built a transformer model called TEDDY (Transformer Enabled Drug Discovery) to predict gene regulatory networks, enabling targeted drug discovery. 7-Eleven automated significant portions of its marketing stack, with agents now handling audience segmentation and content generation. Royal Bank of Canada uses an agent that ingests earnings calls, competitor filings, and market sentiment to produce equity research reports in 15 minutes, versus the industry standard of two or more hours.

Ghodsi categorizes this as "GSD AI" (get stuff done) rather than superintelligence research, and argues AGI is already here in a practical sense. He places himself explicitly in the camp that believes current models are already generally intelligent enough to deliver meaningful enterprise automation.

Open Source as Structural Commoditizer

On the open source model landscape, Ghodsi sees Chinese labs, particularly Alibaba's Qwen, as accelerating commoditization of the foundation model layer globally. He notes, with candor, that American startups are already distilling Chinese open source models into products with American branding, a dynamic he says is widespread but rarely discussed publicly. Databricks itself is multi-LLM, offering OpenAI, Anthropic, and Google Gemini natively to all customers alongside open source alternatives.

His prediction is that state-funded open source model development, analogous to national supercomputer programs of prior decades, will emerge in Europe and the United States as a strategic response to Chinese open source dominance. The net effect of all this pressure on the foundation model layer is lower COGS for companies like Databricks operating above it.

IPO Optionality, Not Commitment

On going public, Ghodsi does not rule out a Databricks IPO by end of 2026, but frames it as a conditional rather than a plan. His reference point is the 2021 vintage of high-growth SaaS companies that went public before rate tightening, then cut R&D and headcount to produce EBITDA in 2022, damaging both morale and innovation capacity. He attributes part of Databricks' current growth rate advantage over those peers to having stayed private through that cycle.

Open Source Monetization Framework

For founders building on open source, Ghodsi offers a two-home-run framework. The first is achieving genuine open source dominance, with the GitHub stars and adoption to prove it. The second, which must follow quickly, is building proprietary innovation on top that cannot be replicated by hyperscalers picking up the open source base. Failing to hit the second home run, he argues, leaves founders exposed to the Mag 7's distribution and pricing power. A hyperscaler launching a clone of your product is not a death knell but confirmation you hit the first home run and now need the second.

Sales Motion at Scale

Databricks operates a bottom-up enterprise sales motion grounded in a permanent free tier accessible via Gmail at databricks.com/try, followed by a $200 free trial credit. In large enterprise conversations, the free-tier adoption already present inside a target organization, employees using personal Gmail accounts, functions as a proof point and trust accelerator that supports top-down deals that can reach $250 million in contract value. The traditional enterprise steak-dinner sales approach, Ghodsi suggests, has largely given way to demonstrated value through product-led growth.