# Pulse ยท September 4, 2026 > A judge spares Google's ad tech business from a breakup, Google's WeatherNext 3 lands in Search, and BigQuery adds stateful streaming joins. Canonical: https://brandonlazovic.dev/pulse/2026-09-04/ Author: Brandon Lazovic Published: 2026-09-04 ## Judge rejects the Justice Department's push to break up Google's ad tech business Status: Confirmed | Topics: ads-paid Judge Leonie Brinkema ruled on September 2, 2026 that Google will not have to divest its AdX ad exchange or DFP ad server, rejecting the Department of Justice's structural remedy in the ad tech monopoly case. Instead, she ordered behavioral fixes, requiring Google to end 'first look' and 'last look' bidding advantages, deprecate Unified Pricing Rules, let publishers set different price floors per bidder, and expose real-time bid data to rival ad servers. In plain terms: An ad exchange is the marketplace software that auctions a website's ad space to advertisers in real time, and DFP (DoubleClick for Publishers) is the separate software publishers use to serve and manage those ads on their own sites. 'First look' and 'last look' were rules that let Google's own exchange see and react to competing bids before a sale closed, an edge regulators found anticompetitive. Why it matters: E-commerce advertisers keep buying into the same Google-run auction, but publishers and rival ad servers gain real-time bid visibility they've never had, which could reshape where programmatic product ads actually clear. Our take: A breakup would have taken years of appeals to resolve; behavioral remedies start changing auction mechanics on a much shorter clock. The real test is whether ending first look and last look actually moves ad dollars to rival exchanges, or whether Google's scale keeps the auction sticky regardless of the rules. - [AdExchanger: Google Won't Have To Break Up Its Ad Tech Business, Judge Brinkema Rules](https://www.adexchanger.com/antitrust/google-wont-have-to-break-up-its-ad-tech-business-judge-brinkema-rules/) ## Google's WeatherNext 3 AI model starts powering Search, Maps, and Gemini Status: Confirmed | Topics: organic-search-core, ai-data-stack Google DeepMind launched WeatherNext 3 on September 3, 2026, an AI weather model built on live satellite data instead of physics-based simulation, and immediately rolled it into Search, the Gemini app, Maps, the Maps Platform Weather API, and Earth Engine. Google says the model delivers hourly, localized forecasts five times sharper than its predecessor, with up to 60% better precipitation accuracy against satellite data and 50% better accuracy for next-day-and-beyond precipitation calls. In plain terms: Physics-based simulation is the older way weather models worked, running equations that model how air and moisture move. This model instead learns patterns directly from real-time satellite images. Why it matters: Another data-driven AI model just got a direct line into Search results, adding to the list of first-party AI answers that never route through a webpage at all. Our take: Weather joins flights, sports scores, and stock quotes on the list of query types Google answers entirely inside its own interface. None of that traffic was ever earned by ranking, so its loss to third-party weather sites is a shrinking-pie problem that started well before generative AI existed. - [Google: Introducing WeatherNext 3, our most advanced and accurate global weather AI model](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/) ## BigQuery continuous queries add joins, aggregations, and windowing Status: Confirmed | Topics: ai-data-stack Google added stateful processing to BigQuery continuous queries in preview on September 3, 2026, letting streaming SQL pipelines run joins, aggregations, and windowing functions directly on live data. Teams can now calculate rolling metrics, such as a 30-minute average, without landing data in a table first, feeding richer real-time signals to downstream applications and AI agents. In plain terms: A streaming SQL pipeline is code that processes data continuously as it arrives, instead of waiting for a batch job to run later on data already saved to a table. Why it matters: Teams building real-time dashboards or agent-facing data feeds get streaming joins without standing up a separate stream-processing layer first. Our take: This closes a real gap for anyone piping live search or crawl data into an agent: the workaround used to be landing data in a table first, then joining after the fact. It's a preview feature, so the thing to watch is whether it reaches general availability before anything production-critical gets built on it. - [Google Cloud: What's new with Google Data Cloud](https://cloud.google.com/blog/products/data-analytics/whats-new-with-google-data-cloud/) ## xAI's Grok 4.6 launches inside Snowflake Cortex AI Status: Confirmed | Topics: ai-data-stack, llm-models-agents Snowflake announced public preview availability of xAI's Grok 4.6 inside Cortex AI on September 3, 2026, positioning the model for long-running agentic work such as multi-step research, codebase analysis, and turning an idea into a working application. Paired with Snowflake's CoCo tooling, it runs inside the platform's governed security perimeter rather than calling an external API, so enterprise data never has to leave Snowflake to reach a frontier model. In plain terms: Public preview means the feature is live for any customer to try, not restricted to a small invite-only test group. A governed security perimeter means the data stays inside Snowflake's own access controls instead of being sent to an outside company's servers. Why it matters: Enterprise teams running SEO or product data pipelines in Snowflake get frontier-model reasoning without an added data-export decision. Our take: This is the second frontier model Snowflake has landed inside Cortex in three days, after Claude Fable 5.1 on September 1. The pitch to enterprise buyers is model choice without a data-export decision, and that only holds up if Cortex's own inference costs stay competitive with calling the model directly. - [Snowflake: Announcing Grok 4.6 on Snowflake Cortex AI](https://www.snowflake.com/content/snowflake-site/global/en/blog/announcing-grok-4-6-snowflake-cortex-ai) ## Shopify lets store staff see a customer's abandoned online cart at checkout Status: Confirmed | Topics: platform-ecommerce Shopify shipped a POS update on September 3, 2026 letting authorized retail staff view an identified customer's items from an abandoned online checkout, directly inside the point-of-sale system, to help close the sale in person. It requires Shopify POS version 11.14 or later, an explicit 'View abandoned checkouts' permission set per role, and a customer identified by email with an eligible abandoned checkout; staff see cart contents only, not browsing history. In plain terms: POS, or point of sale, is the checkout system a store uses to ring up in-person purchases. An abandoned checkout means a shopper added items online and started paying but left before finishing. Why it matters: Closes a real cross-channel gap for retailers, since an online cart abandonment stops being a dead end the moment that same shopper walks into a store. Our take: This only works where the retailer already asks online shoppers to identify themselves by email before checkout, so it rewards stores that built that habit rather than creating a new one. Expect competitors to frame their own cross-channel cart-recovery features around whatever numbers Shopify publishes for this one. - [Shopify Changelog: Permitted store staff can now see an identified customer's online cart in POS](https://changelog.shopify.com/posts/permitted-store-staff-can-now-see-an-identified-customer-s-online-cart-in-pos)