The data stack moves back into the spreadsheet

The spreadsheet became strategic again

On 25 Sep, press coverage reported Databricks had acquired Row Zero. On 27 Sep, the move was described as bringing live, governed spreadsheets into Genie. That is the cleanest statement of the fortnight: the spreadsheet is not being displaced. It is being pulled inside the governed AI platform.

Sigma was already working the same seam. On 26 Sep it published eight blog posts across AI agents in demand forecasting, cloud BI and spreadsheet governance. On 27 Sep it added two case studies, one on Dotmatics shipping new workflows in days and another on Delta Defense replacing nine years of spreadsheets with a Sigma scorecard. No new customer logos were added in that update, so the point was use-case depth rather than logo expansion.

Snowflake put governance closer to distribution. On 26 Sep it launched a marketplace governance feature, while Liquibase Secure became available on the Snowflake Marketplace for automating and governing changes across data platforms. Across those events, the common object was not storage or compute. It was the uncontrolled operational layer where spreadsheets, workflows and change management meet.

Databricks is buying reach and hiring for AI

Databricks paired acquisition with geographic expansion. On 19 Sep and again on 21 Sep, coverage said the company was investing more than US$350 million in Singapore to support enterprise AI adoption, with plans reported on 19 Sep to double its regional workforce. Its hiring signal matched that posture: on 22 Sep it had 15 new roles, mostly engineering and product, on 25 Sep it had 24 new postings across engineering, sales, product and marketing, and on 27 Sep it listed engineering roles including AI Forward Deployed Engineer and Lead Solutions Architect, Generative AI.

The developer surface moved too. On 19 Sep, Databricks released TypeScript SDK v1.3.0, promoting Arrow Flight ingestion to general availability, and created repositories including sdk-js, databricks-sdk-py, sdk-go, dicer and databricks-jdbc. On 24 Sep it released CLI v1.18.0, changing command paths for AI Runtime commands. On 27 Sep, unity-gateway 0.2.0 added estimated per-model costs and Databricks created appkit, terraform-provider-databricks and zerobus-sdk.

Snowflake’s fortnight had a different shape. On 21 Sep, reports said its shares rose 23 percent after a Q2 earnings beat and raised guidance, with AI efforts cited in the coverage. It added Mandai Wildlife Group to its public customer wall on 19 Sep and Canva on 25 Sep, with case studies attached to both. The same period also brought operational and security disclosures: CVE-2026-92903, published on 17 Sep, carried a CVSS score of 8.2 for Snowflake CLI versions prior to 3.27.0, and on 26 Sep Snowflake resolved a critical outage, incident INC20000237, after about 80 minutes.

dbt and Fivetran turned agent-ready into packaging

The Fivetran and dbt Labs story was not just announcement volume. On 19 Sep, 20 Sep and 21 Sep, the companies were covered for dbt v2, dbt State and new capabilities aimed at making enterprise data agent-ready, announced around dbt Summit 2026. On 22 Sep, dbt Labs changed its homepage from an event-oriented message to the headline The open standard for modern, AI-ready data transformation, with subtext about SQL-based development and AI-ready data pipelines.

The commercial packaging changed before the homepage did. On 17 Sep, dbt Labs updated tier entitlements, including Developer and Starter references covering job scheduling and monitoring, CI checks on GitHub and GitLab, five developer seats, dbt Catalog and dbt Semantic Layer. Then on 26 Sep it added 21 product pages and two blog pages while removing 15 product pages and one blog page. The sequence matters: packaging first, homepage next, product library after.

Fivetran’s work was quieter but more mechanical. On 19 Sep it added 11 troubleshooting documentation pages and removed 11 Snowflake destination troubleshooting pages. On 22 Sep it added Odoo and CI/CD deployment documentation while removing older Odoo external API pages. On 25 Sep it added 10 more troubleshooting and user management documentation pages while removing 10 older troubleshooting pages. Its GitHub activity pointed to the same connective tissue, with repositories including fivetran-mcp, dbt_openai, dbt_claude, dbt_reddit_ads, dbt_amazon_ads and dbt_fivetran_utils created or observed during the period.

The small edits carried the market signal

The fortnight was busy, but the most revealing movements were in the edits that usually vanish between press cycles. dbt Labs moved its homepage from Community Keynote on 19 Sep to Don’t miss what’s next from dbt, then to an AI-ready transformation message on 22 Sep. Snowflake removed two pricing pages on 25 Sep while adding product pages on warehouse modernisation and data struggles. On 27 Sep it added a Spanish-language modernisation presentation page and removed an Advertising Week 2025 page.

Customer-wall changes were equally telling because they happened in sequence. Snowflake added Mandai Wildlife on 18 Sep and 19 Sep, removed booking.com from the displayed wall on 18 Sep, removed Nissan on 23 Sep, then added Canva on 25 Sep. None of that says a customer was won or lost beyond the stated public display changes, but it shows how quickly reference architecture is being curated around AI, security and industry narratives.

The ad machines were also fully loaded. On 17 Sep, Databricks had 40 new Google ad creatives observed, with 40 running in total. On 26 Sep, Snowflake had 38 new creatives, Fivetran 35 and Databricks 36, each with 40 running in total. By 28 Sep, Snowflake still showed 31 new creatives and Fivetran 36, while Databricks had 13. dbt Labs was steadier, with 11 new creatives on 17 Sep, five on 26 Sep and six on 28 Sep, while still showing 40 running in total. The public narrative was AI readiness. The operating reality was rapid message testing across every layer of the stack.