The data stack moves closer to the business user

The loudest move was Databricks buying Row Zero. From 29 Sep through 4 Oct, multiple reports described the acquisition of the spreadsheet startup, with Row Zero's live, governed spreadsheets set to feed into Databricks' Genie product. On 2 Oct and 3 Oct, Databricks was also reported to be participating in a $5 billion funding round, with SBVA among the investors and a reported valuation of $190 billion.

Snowflake's fortnight was equally active, but with a different shape. On 28 Sep, reports focused on a proposed $3.5 billion convertible debt offering and stock dilution concerns. By 30 Sep and again on 2 Oct and 4 Oct, Snowflake had announced an upsized $3.75 billion zero-coupon convertible senior notes placement, including capped calls to limit dilution. In the same window, Snowflake announced GPT-6 Sol and Luna on Cortex AI, expanded its LSEG collaboration for financial data and AI workflows, and completed the Observe acquisition for a reported $595.8 million.

The competitive line is no longer just warehouse versus lakehouse. The observed moves put the fight around who owns governed AI workflows at the point of work: spreadsheets, decision functions, financial data workflows, native apps and assistant-facing product surfaces.

Snowflake shipped a dense run of developer infrastructure changes. On 28 Sep it updated terraform-provider-snowflake to version 2.21.0 with a migration guide, then on 1 Oct released libsnowflakeclient 2.11.0 with OCSP certificate revocation checking off by default and external-browser callback changes. On 5 Oct it released snowflake-odbc 4.0.0, including breaking changes related to catalog functions. Across the same period, it added repositories including cortex-training, snowflake-kafka-connector, snowflake-telemetry-python, snowpark-java-scala, PowerPlatformConnectors and snowflake-connector-net.

Databricks was just as visible in the implementation surface. On 29 Sep, CLI version 1.18.0 moved AI Runtime commands to `databricks air`. On 30 Sep, unity-gateway v0.2.0 added estimated per-model cost in usage reports and removed some CLI flags. On 3 Oct, databricks-jdbc 3.4.3 added `EnableThriftNativeMetadata` and improved Real-Time SQL warehouse functionality. On 4 Oct, CLI 1.19.0 added support for `CLAUDE_CONFIG_DIR` in `aitools` commands and new flags for `databricks postgres create-branch`.

Fivetran's public work was less platform-theatre and more plumbing. On 29 Sep it launched repositories including dbt_workday_financial_management, dbt_tiktok_ads, dbt_ad_reporting and duckdb-iceberg. On 30 Sep it added dbt_reddit_ads, great_expectations, dbt_starter_project and go-fivetran. On 2 Oct it launched dbt_package_automations, dbt_greenhouse, dbt_ai_reporting, terraform-provider-fivetran and community_connectors, then on 3 Oct released great_expectations 1.23.2 with SQLAlchemy compatibility updates.

The paid channel was saturated. On 29 Sep, Databricks had 40 new Google ad creatives and 40 running in total, Snowflake had 39 new and 40 running, and Fivetran had 36 new and 40 running. On 1 Oct, Snowflake showed 40 new creatives with 40 running, while Fivetran had 33 new and Databricks had 30 new. By 5 Oct, Snowflake still had 31 new creatives, Databricks had 36 new, and Fivetran had 16 new, with each still showing 40 running in total. dbt Labs was quieter in new creative volume, with 6 new creatives on 28 Sep and 4 on 29 Sep, while also showing 40 running.

The more interesting signal sat in the website and documentation layer. Sigma Computing changed its hero on 1 Oct from `Go build it.IT approves.` to `Vibe-code enterprisebusiness applications.`, with subtext moving to `Sigma is the layer to build and scaleyour analytics, apps, and agents on trusted data.` On 2 Oct it changed back to `Go build it. IT approves.`, with subtext updated to `Analytics, apps, and agents on one trusted platform`. That is a two-day public oscillation around whether the product is led by vibe coding, IT approval, trusted data or a unified app surface.

Fivetran's docs told a different story. On 29 Sep it added Adobe Analytics troubleshooting while removing a Snowflake duplicates page. On 30 Sep it added Amazon Appstore and Cursor documentation while removing five Snowflake troubleshooting pages. On 2 Oct it added a Pennylane connector page and removed connector SDK best practices, then on 3 Oct added Facebook Ads and bulk row filter troubleshooting while removing Snowflake column renaming and default virtual warehouse pages. The launches matter, but the removals are where positioning and support burden often show first.

Snowflake had two operationally relevant events. On 28 Sep, it experienced a critical outage, incident INC20000239, lasting about 61 minutes before resolution. On 29 Sep, CVE-2026-81930 was published with a CVSS score of 6.3, covering the Snowflake provider in Apache Airflow failing to validate `account` and `region` fields before using them in request URLs. Both events were reported as resolved or published facts, and they landed in the same fortnight as the company's financing and AI workflow announcements.

dbt Labs had two minor incidents. On 1 Oct, dbt deps failures affected private packages for about 56 minutes before resolution. On 2 Oct, increased errors for Wizard CLI requests using Anthropic models lasted about 89 minutes before resolution. On 4 Oct, dbt Labs' YouTube output centred on context engineering with dbt, a dbt Summit 2026 recap with product launches and community engagement, and the 2026 dbt Community Awards.

Hiring and community activity remained concentrated around AI, engineering and customer adoption. Databricks had rolling seven-day job postings across engineering, machine learning, product and sales on 29 Sep, 30 Sep, 2 Oct and 3 Oct, including machine learning-focused roles. Snowflake listed engineering and machine learning roles on 1 Oct and 2 Oct across locations including Sydney, Menlo Park, Tokyo and London. Sigma listed four roles on 3 Oct across marketing, sales and customer success, and on 29 Sep published AI-agent content for sales, RevOps, marketing analytics and headless Sigma agent experiences with Claude.