Row Zero's 9.5x Grid Headroom Isn't Enterprise Pricing • The Weighted Average

Databricks announced its acquisition of Row Zero on September 24, promising a governed spreadsheet experience inside Genie. Row Zero’s current non-enterprise pricing advertises a row-count reference roughly 9.5x Excel’s worksheet limit, but neither that comparison nor the $20 Business subscription buys the full enterprise governance proposition.

The spreadsheet is familiar; the entitlement is not

This is an analysis of Thursday’s acquisition and the purchasing boundaries visible today, not a claim that the deal was announced yesterday. Databricks says Row Zero will integrate with Genie across major clouds and continue supporting data sources beyond Databricks. The operative word is will. An acquisition announcement establishes product direction; it does not prove that every promised integration is already enabled in a customer’s tenant.

The capacity arithmetic comes from two first-party sources. Row Zero’s pricing matrix lists 5GB per workbook with “10M+ rows” for its non-enterprise plans. Microsoft specifies 1,048,576 rows per Excel worksheet. Using ten million as Row Zero’s advertised reference, 10,000,000 ÷ 1,048,576 = 9.54, or about 9.5x. That is a comparison of a marketed row-count reference with a worksheet limit, not a benchmark of calculations per second.

The 5GB boundary matters more than the attractive ratio. Rows can contain different amounts of data, so ten million is not a guaranteed capacity for every workbook. Row Zero’s comparison page talks about far larger enterprise scale, while its pricing matrix separates that from the standard plans. A buyer should test the actual dataset and transformations rather than infer capacity from a row count detached from row width.

Nor is the worksheet grid the entirety of Excel. Microsoft’s guidance for datasets too large for the grid describes working through Power Query and analyzing data without loading all of it into worksheet cells. The honest migration question is whether a team needs more interactive, editable grid space and different governance—not whether Excel becomes useless at its last visible row.

The commercial split is concrete. Row Zero lists Pro at $12 per user per month and Business at $20 per user per month. Business adds automated refresh and warehouse write-back. Enterprise pricing is custom, with features including SSO and SCIM, restricted export, dedicated data planes, and private deployment options listed in that tier. Buyers attracted by the acquisition’s governance language should not automatically place the Business price beside an enterprise requirements checklist.

That distinction extends the warning in our earlier analysis of Amazon’s free seller AI and the paid-tool renewal baseline: a compelling entry price is useful only when the product boundary matches the work being replaced. Here, the difficult part may be access administration and controlled write-back, not opening a larger file.

Trace the credential before trusting the cell

The governance claim deserves a closer reading than “data stays in the cloud.” Row Zero’s security documentation says connections inherit the permissions of the credentials they use, including row-level security. A user needs access to the Row Zero connection, and the associated database identity must have the necessary permissions. That means the connection’s identity is part of the control boundary. A spreadsheet cannot repair an overly broad database credential merely by presenting a friendlier interface.

For a pilot, choose a workflow where the current export creates a specific problem: stale data, a manual refresh, or a result that must be written back under review. Row Zero’s connected-spreadsheet documentation describes scheduled refresh, query-based sources, and write-back. Those are mechanisms worth testing. Their existence does not establish that a particular finance model will preserve its formulas, permissions, and review history without migration work.

Start by reproducing the existing result, not by asking an agent to redesign the process. Confirm that two users with different warehouse permissions see the expected records. Check what happens when access changes after a workbook has already loaded data. Verify who can share, export, and write back. These are proposed acceptance tests, not claims that the product currently fails them. They make the promised governance observable to the team responsible for the underlying data.

Costs extend beyond the seat. The security documentation says Private Link carries additional fees; enterprise pricing is not publicly fixed. Warehouse activity, connection administration, validation, and AI usage also need to be understood in the actual arrangement. The pricing page places bring-your-own-key AI in Enterprise, while standard plans describe models with weekly limits. Neither should be read as unlimited inference at the Business seat price.

The acquisition’s strongest case is not merely scale. Databricks says spreadsheet syntax can make agent actions interpretable and auditable to the business team using the workbook. If that holds for the team’s real models, it could give reviewers a more useful artifact than a chat answer whose transformations are difficult to inspect. But “auditable” must include the source identity, refresh state, and write-back decision, not only a visible formula.

Today’s Copilot lead examines the same gap between a familiar interface and an additional operating bill. Builders should resist treating the spreadsheet or chat window as the system boundary. What runs behind it—and which identity pays for and authorizes that work—determines the operating risk.

Switch a workflow when connected refresh, reviewable transformations, and permission tests outperform the existing process at a known total cost. Delay a broad replacement if it depends on unreleased Genie integration or enterprise controls that have not been quoted. Evidence that would change the verdict is a successful permission-preserving migration with measured refresh and review costs. 9.5x is a reason to test the grid; it is not permission to skip the contract.