Broadcasters' infrastructure playbooks

Broadcasters are sketching out new playbooks for AI-heavy video systems that can survive huge traffic spikes without breaking. (cloud.google.com) (broadpeak.io) Google Cloud’s current stack pairs BigQuery, its data warehouse, with Vertex AI for model training and serving, while Dataflow handles streaming data and Cloud Run runs containerized services on demand. Google’s documentation says the combination is aimed at recommendation, prediction and other machine learning workloads that need data preparation, model deployment and monitoring in one workflow. (docs.cloud.google.com) (cloud.google.com 1) (cloud.google.com 2) For video teams, that means the AI work is no longer just encoding and delivery. Google’s recent video-insights codelab shows Cloud Run, BigQuery and Gemini processing hundreds or thousands of video Uniform Resource Locators in parallel to generate summaries, chapter titles and question-and-answer pairs. (codelabs.developers.google.com) (cloud.google.com) A content delivery network is the layer of edge servers that stores copies of video closer to viewers, like stocking inventory in local warehouses instead of one distant depot. Broadpeak said on April 8, 2026 that its video-focused content delivery network as a service is being sold with “guaranteed bandwidth” options for premium live events, where ordinary best-effort delivery can fail under sudden demand. (broadpeak.io) Broadpeak tied that pitch to recent streaming peaks that reached national-scale network loads. The company cited Netflix’s Jake Paul-Mike Tyson fight at 65 million viewers and an estimated 65 terabits per second in November 2024, and JioHotstar’s Champions Trophy final at 60 million viewers and 60 terabits per second in March 2025. (broadpeak.io) The company has also been building out the physical side of that promise. Broadpeak said on September 4, 2025 that it added HyperPoP caches in England, Switzerland, Greece and Mexico, with each site offering more than 1 terabit per second of capacity for live sports, series launches and other traffic surges. (broadpeak.tv) The software pattern underneath these systems is simple even if the stack is not: store copies in layers, and put requests in line before they hit the expensive AI step. Google’s guidance for streaming prediction uses Pub/Sub to capture incoming data, Dataflow to clean and transform it, and Vertex AI endpoints to return predictions, which spreads work across stages instead of letting one burst overload the whole pipeline. (cloud.google.com 1) (cloud.google.com 2) Google is also pushing more of the artificial intelligence work directly into the analytics layer. In a February 29, 2024 post, the company said BigQuery can call Gemini models through Vertex AI so teams can blend structured data, unstructured data and generative models inside one serverless pipeline. (cloud.google.com) That leaves broadcasters with two linked infrastructure jobs: keep the stream moving to millions of screens, and run personalization or analysis fast enough that the extra intelligence does not become the outage. The current vendor pitches from Google Cloud and Broadpeak are aimed at making those two jobs look like one managed system. (docs.cloud.google.com) (broadpeak.io)