Azure Updates: Fireworks AI on Microsoft Foundry; Copilot migration agent; Agentic AI

Microsoft CVP Yina Arenas highlighted the introduction of Fireworks AI on Microsoft Foundry. Now in public preview, the offering brings high-performance open model inferencing to Azure. Developers are able to optimize deployments for custom weight models. After logging into Foundry they can access open models with Fireworks AI including DeepSeek v3.2, Kimi K2.4, OpenAI gpt-oss-120b, and MiniMax M2.5. 

According to Arenas, Fireworks AI operates on “internet scale processing,” handling over 13T tokens per day and about 180,000 requests per second. Developers will be able to optimize inferencing, run trusted models, and evaluate models more rapidly. 

“As open models mature, customization increasingly extends beyond training. Teams need consistent ways to configure, deploy, optimize, govern, and iterate on models in production without fragmenting tools or infrastructure,” Arenas wrote. “Microsoft Foundry provides the environment where these customization and operational workflows are standardized, while Fireworks AI supplies the performance and efficiency needed to run open models at scale. This means teams can move from experimentation to production using open models without stitching together separate tools, contracts, and deployment paths.”

Jeremy Winters, CVP and Chief Product Officer for Azure Platform looked into new agents that can be built for scalability and control. “[M]odernization is becoming one of the biggest opportunities in the AI era,” he wrote. Even after years of pushing for digital transformation, many organizations have a jumbled and disconnected data estate with tools that don’t talk to one another.