AWS launches Nova Forge and AgentCore to scale custom video AI

The launch of Nova Forge specifically addresses the limitations of standard Retrieval-Augmented Generation (RAG) by allowing streaming platforms to bake domain-specific knowledge—like unique content catalogs or specialized encoding rules—directly into frontier-class models. This shifts AI from a bolt-on interface to a core structural component of the video delivery stack. Within the competitive ecosystem, this deep vertical integration allows AWS users to differentiate their generative search and recommendation engines against rivals using generic third-party APIs. Watch for the initial performance benchmarks of 'Novellas'—the custom model variants produced by Nova Forge—as they hit production in early 2026.

AWS re:Invent 2025 signaled a strategic pivot toward AI-first, vertically integrated infrastructure meant to scale highly specialized workloads like live video and multimodal search. Per IDC reporting in December 2025, AWS is positioning itself as a foundational partner for enterprises navigating 'data gravity' by offering custom silicon like the Trainium3 chip alongside its software advancements. This hardware-software synergy is designed to make large-scale model training more cost-effective for media organizations that previously relied on general-purpose compute instances.

In practical application, the streaming industry is already testing these agentic capabilities to handle extreme concurrency. Per an AWS case study from December 2025, Sky utilized Redis Cloud on AWS to manage 140,000 writes per second during the Peacock-exclusive NFL Wild Card game. Sky is now integrating its GenAI platform, Genapi, with these latest AWS vector search and semantic caching tools to improve real-time content discovery. Furthermore, per AboutAmazon (December 2025), AWS introduced Nova 2 Lite models optimized for browser-based automation, achieving 90% reliability in early customer workflows for tasks like automated QA testing and metadata extraction from high-volume video libraries.

The broader focus at re:Invent also highlighted the tension between proprietary innovation and multi-cloud resilience. Per Mondo (December 2025), while AWS is aggressively expanding its Nova family, many streaming enterprises are simultaneously recruiting talent fluent in Azure and GCP to maintain cross-platform architectures. This dual-track strategy ensures that while organizations leverage Amazon’s specific AI breakthroughs, they avoid total vendor lock-in for critical video delivery pipelines.