Databricks Just Hit $188 Billion — And It's Quietly Building the Infrastructure Every AI-Powered No-Code Platform Will Run On

Databricks signed a term sheet last week valuing the company at $188 billion. Led by Coatue, the round is expected to raise roughly $3 billion and close later this summer. To put that number in perspective: it's higher than Goldman Sachs. Higher than IBM. Closing in on Salesforce. And Databricks has never gone public.

The immediate reaction is to file this under "AI hype machine keeps spinning." And sure, the valuation has nearly tripled in 18 months — from $62 billion in December 2024, to $134 billion in February, to $188 billion now. Memes about running out of alphabet letters for funding rounds write themselves.

But I think the valuation tells a more interesting story. It's not a bet on AI models. It's a bet on AI infrastructure. And if you're building or evaluating no-code platforms right now, that bet changes how you should think about the entire category.

## What is Databricks actually building?

The new capital is earmarked for three products: **Lakebase**, **Unity AI Gateway**, and **Genie**. Alongside **Omnigent** (released in June under Apache 2.0), these form a stack that reads like a blueprint for where enterprise AI is heading.

**Lakebase** is a serverless Postgres database purpose-built for AI agents. Not a data warehouse with an AI sticker on it — a transactional database designed for autonomous software that reads, writes, and acts on live data without human intervention. As agentic workflows multiply, the gap between analytical infrastructure and operational systems becomes a genuine bottleneck. Lakebase is the answer to "where does the agent's state actually live?"

**Unity AI Gateway** is a governance layer. Its job: control which models employees and systems can access, enforce security policies, and track spend across every model an organisation runs. Most enterprises are already running dozens of AI models across hundreds of workflows with no coherent way to govern any of it. Unity AI Gateway is a single control plane for the entire AI stack.

**Omnigent** is the newest piece — an open-source meta-harness for orchestrating multiple AI agents. Its three design pillars are composition, control, and collaboration. Think of it as the conductor that tells Claude Code to handle this task, Codex to handle that one, and a custom agent to clean up afterwards.

**Genie** sits on top as the business-facing interface: an AI coworker that turns governed data into trusted answers without a data engineer in the loop.

Read together, this is not a suite of AI products. It's an operating system for agentic software. Lakebase is the filesystem. Unity AI Gateway is the permissions model. Omnigent is the process scheduler. Genie is the shell.

## The number that should make model companies nervous

While announcing the round, Databricks CEO Ali Ghodsi published internal benchmarking results that deserve more attention than they got.
The company tested AI coding tools on real tasks across its 3,000-engineer codebase — not synthetic benchmarks, but actual production work. The finding: open-weight models, particularly Z.ai's **GLM 5.2**, now handle the highest-difficulty coding tasks at lower total cost than proprietary models from Anthropic and OpenAI.

That alone is interesting. But the surprise was this: the choice of coding agent (the tool that wraps the model and manages context) mattered just as much as the model choice. Open-source tool **Pi** emerged as one of the best performers for managing context efficiently. The blog post's conclusion was unusually frank: "Model choice is only one piece of the puzzle."

Ghodsi called this shift "moving from tokenmaxxing to valuemaxxing." Enterprises are done burning expensive tokens on the smartest model for every task. They want the best outcome per dollar. That means routing work to the right model at the right cost, governed through infrastructure they control.

The model layer is becoming a commodity. The infrastructure layer — the one Databricks owns — gets more valuable with every new model that launches.

## What this means for no-code

Here's where it gets relevant if you're not running a Fortune 500 data team.

No-code platforms sit at exactly the same intersection Databricks is betting $188 billion on. They are governed runtimes. They handle permissions, audit trails, data access control, user management — all the unglamorous infrastructure that determines whether AI features actually work in production or just look good in a demo.

The valuation logic applies directly. A no-code platform that bolts on an AI chat widget is worth whatever the AI widget costs to license. A no-code platform that gives you governed access control, structured data, audit trails, and a permission model that actually controls what AI agents can see and do — that platform is building the same kind of infrastructure moat Databricks just raised $3 billion to deepen.

**Stacker** is the obvious example here. Its entire architecture is built around governed access: portals with granular permissions, audit logs, role-based data visibility. When you connect AI to that — whether it's formula fields, automations, or external agents hitting an API — the platform already knows who can see what. The governance layer isn't bolted on after the fact. It's the foundation.

Compare that to no-code tools where the AI feature is basically a text box that calls an API. There's no permission boundary between what the AI can access and what the user can access. There's no audit trail of what the AI did. The demo is impressive. The security review is a disaster.

## A framework for evaluating no-code platforms through the infrastructure lens

I've been thinking about this in terms of three questions you can ask about any platform you're evaluating.

First: where does the AI agent's state live? If the answer is "in a vector database we don't give you access to," that's a red flag. The platform that owns the transactional data layer — the Postgres database, not the embeddings store — owns the architecture.

Second: who controls what the AI can access? If the AI inherits the user's permissions automatically, great. If there's a separate AI permission model, also great. If nobody has thought about it and the AI can see everything the platform can see, that's not a product decision. That's an incident waiting to happen.

Third: can you route AI tasks to different models? The Databricks research showed that model choice is only one variable — and not always the most important one. Platforms that lock you into a single model provider are making the same mistake enterprises made when they bought one cloud vendor's entire stack. It works until it doesn't, and then it's expensive to fix.

None of this is to say every no-code platform needs to be Databricks. But the $188 billion signal is unmistakable: investors are pricing the governance layer at a premium over the model layer. The market believes, with a serious amount of money, that controlling AI access, cost, and reliability is a bigger business than building the AI itself.

If you're choosing where to build, pick a platform that already understands that.