AI pushes the data stack into the operator's seat

The contest moved from storage to action

This was not a quiet fortnight. On 7 Oct, Databricks raised $5 billion at a $190 billion valuation and disclosed the acquisition of Row Zero, a governed spreadsheet company. On 10 Oct, Snowflake announced the acquisition of TensorStax to advance agentic AI for data engineering. The two largest platforms both moved toward the same place: the point where data is changed, modelled, queried and acted on by business users and agents.

Snowflake had already set the frame. On 4 Oct, it priced an upsized $3.75 billion private placement of 0.00% convertible senior notes, announced GPT-6 Sol and Luna integrations on Snowflake Cortex AI, and expanded work with LSEG to connect financial data with AI workflows. On 5 Oct, its EXPEDITION 2026 messaging centred on enterprise-wide agentic transformation. On 9 Oct, snowflake-cli 3.29.0 added features for launching coding agents through Snowflake AI Gateway. On 10 Oct, TensorStax made the agentic data engineering push explicit.

Databricks answered from the application and governance side. On 7 Oct, it published lakehouse benchmark material and acquired Row Zero. On 8 Oct, it announced Workday Data Connect federation into Unity Catalog. On 10 Oct, the Row Zero deal was again described as enhancing Databricks' AI spreadsheet product. That combination says the lakehouse pitch is no longer only about where data sits. It is increasingly about who gets to manipulate it, under what controls, and inside which familiar interface.

Sigma made the clearest messaging cut

Sigma Computing changed its hero headline on 11 Oct from 'Go build it.IT approves.' to 'AI runtime for business.' Its subtext moved from 'Analytics, apps, and agents on one trusted platform' to 'Sigma is the trusted platform for analytics, apps, and agents.' That is a tighter claim. It drops the internal approval drama and claims the runtime layer for business AI.

The surrounding edits made the shift look deliberate rather than cosmetic. On 8 Oct, Sigma launched a product page outlining trial terms and promoted a webinar on building Sigma apps using ChatGPT and Grok. On 9 Oct, it launched content on Claude Dashboards and added a Treasury4 customer page. On 10 Oct, it added pages on building Sigma apps with ChatGPT, setting up Claude skills for Sigma, and optimising complex scheduling into Sigma apps, while removing an earlier blog page that paired ChatGPT and Grok.

The timing matters. A homepage rewrite on 11 Oct came after three days of AI app, assistant and customer proof content. The public surface moved in sequence: onboarding terms, AI app education, customer proof, then a new front-door claim. That is the sort of packaging change that can look minor in isolation, but it shows a vendor turning scattered AI features into a category position.

Distribution was as aggressive as product

Snowflake and Databricks both kept paid acquisition at full visible capacity, with 40 Google ad creatives running under each publisher account during the period. Snowflake logged 31 new creatives on 5 Oct, 31 on 6 Oct, 31 on 7 Oct, then further sets on 9, 10 and 12 Oct. Databricks logged 36 new creatives on 5 Oct, 23 on 6 Oct, 38 on 7 Oct, then further sets on 9, 10 and 11 Oct. Fivetran also showed 40 creatives running, with smaller new creative batches across 5 to 10 Oct.

The ad cadence lined up with product and capital news. Snowflake's heavy creative rotation started on 5 Oct, the day its EXPEDITION 2026 and AI positioning appeared in press coverage, and continued through the 10 Oct TensorStax announcement and 12 Oct stock and AI coverage. Databricks' visible rotation peaked on 7 Oct, the same day as the $5 billion financing, Row Zero acquisition disclosure, lakehouse benchmark article and Marketplace news.

This is a market where launch timing is now compressed. Capital news, acquisition news, benchmark content, customer proof and ad refreshes are landing within days of each other. Operators watching only press releases would see the headline. Operators watching the surrounding surface would see when the message was armed, where it was amplified, and which claims were being tested in market.

The plumbing did not disappear

For all the AI packaging, core platform work remained noisy. On 5 Oct, Snowflake released ODBC driver 4.0.0 with breaking changes related to catalog functions. On 6 Oct, snowflake-ml-python 2.4.0 added Python 3.14 support and schema-free dictionary support for FeatureSpec object inputs or outputs. On 7 Oct, Databricks CLI 1.20.0 removed the Terraform deployment engine, causing errors for configurations that relied on it. On 10 Oct, Databricks' VS Code extension 2.20.0 added Bundle Variables refresh behaviour and raised the minimum VS Code version to 1.104.

Fivetran's week was similarly practical. On 6 Oct, it created repositories including hybrid_deployment, dbt_claude and connector_sdk. On 7 Oct, it added connector_sdk_tools, go-fivetran, dbt_greenhouse and dbt_workday_financial_management. On 9 Oct, dbt_fivetran_log 2.5.3 added DuckDB as a supported destination, and repositories including dbt_shopify, community_connectors and dbt_claude appeared. On 10 Oct, it added dbt_qualtrics and dbt_amazon_ads.

Documentation churn showed the same integration grind. Fivetran added troubleshooting pages across Salesforce Marketing Cloud, Skilljar, Sprout Social, Adobe Analytics, Google Analytics 4, NetSuite SuiteAnalytics and SAP ECC to Oracle between 6 and 10 Oct, while removing several Snowflake destination, Redshift and SingleStore related pages. dbt Labs added two blog pages, a customer case study and nine product pages on 10 Oct, after creating a dbt World Tour Tokyo page on 6 Oct. The AI story is front of house. The back office is still connectors, drivers, CLI compatibility, docs and workflow edge cases.

Reliability stayed in the buying conversation

Two service signals stood out on 9 Oct. Snowflake reported a major incident, INC20000263, which was resolved after approximately 73 minutes and classified as a major impact. The same day, dbt Labs detected a minor incident in which run logs and artifacts were missing for some runs in US Cell 3 on GCP, with the status being monitored.

These events landed in the same window as the AI acceleration. Snowflake had just released AI Gateway coding-agent features in snowflake-cli 3.29.0 on 9 Oct and announced TensorStax on 10 Oct. dbt Labs was expanding content and product pages on 10 Oct. The market is asking vendors to make data systems more autonomous, but the operational baseline is still availability, logs, artifacts and predictable tooling.

Hiring reinforced where attention is going. In the period, Snowflake posted engineering, machine learning, compliance, sales and programme roles across locations including Menlo Park, Bellevue, Chicago, Bangalore, Amsterdam, Sydney, Pune and Dublin. Databricks posted engineering, machine learning, product, sales, marketing and solutions roles across locations including San Francisco, Seattle, New York City and Paris. Sigma listed engineering roles focused on machine learning. Fivetran listed engineering, partner, sales, tax and onboarding roles across the United States, India, Canada, Austin, Bengaluru and Dublin. The people plan matched the product plan: more AI engineering, more field motion, and more operational support around the data stack.