AI needs your own knowledge. The industry now agrees.

Last month Databricks and Microsoft extended their partnership into the 2030s. The core of the announcement was not a new model or more compute. It was a principle: AI delivers value only once it is grounded in an organization's own knowledge, properly governed, with costs under control.

That principle is not new. The endorsement is. When the two largest players in the market renew a decade-long partnership on this exact point, the debate is effectively over. Grounding AI in business context is no longer a position to argue for. It is the center of the field.

Microsoft's commercial chief framed the next phase around a single idea: the organizations that win will be the ones that convert their own knowledge into intelligence.

So the principle is settled. Two questions are not, and those two decide whether it works for a given organization.

Where does it start?

A platform-first approach carries an assumption. Grounding an AI co-worker on a data platform works well for organizations that already have one, with a data team to keep it running.

Most organizations are not there yet. They have no lakehouse and no data team. They have a handful of systems, a great deal of spreadsheet, and a sense that more sits inside the organization than comes out of it. For them, a platform-first answer begins too late.

Another starting point exists. The brain of an organization can begin on the context that is already present, without building a full platform first. A single spreadsheet is often enough to reveal what is already there. Once it proves its worth, core data is connected, and a platform such as Databricks becomes the engine underneath. The platform is where an organization grows to, not where it has to begin.

Whose is it?

The second question weighs heavier. Once the meaning exists, who owns it?

In a platform-grounded model, that knowledge lives inside a vendor's ecosystem: grounded in the vendor's platform, governed by the vendor's layer, running on the vendor's cloud. For many organizations that is entirely acceptable. For others, and for certain kinds of data, it is not.

The alternative is meaning that remains the organization's own. Portable, not locked in. And where data cannot leave the building, the brain runs on the organization's own hardware, through private AI. Not rented in another environment, but owned.

One model makes an organization a tenant of its own knowledge. The other makes it the owner.

Two models side by side: as a tenant, access to your own meaning runs through the vendor; as the owner, you hold the key yourself.

What remains

On the principle, the field has converged. AI needs an organization's own knowledge, governed and trustworthy. That is not a miracle. It is craftsmanship.

What remains open is not whether to ground AI in business context. That has been answered, by the largest players there are. What remains is where the work starts, and who owns the result once it stands.

The most defensible answer is the straightforward one: start where the organization already is, and keep ownership of what gets built.

Source: Databricks and Microsoft, announcement 23 July 2026.