News
Microsoft Brings Production Ready AI Agents at GTC
Microsoft used Nvidia's GTC conference this week to roll out a series of enterprise AI announcements spanning agent infrastructure, real-time voice interactions and next-generation GPU deployments.
The highlight is the general availability of the latest Foundry Agent Service, Microsoft's developer platform for building, deploying and operating AI agents at enterprise scale. Built on top of the Responses API -- OpenAI's agentic API -- agents developed through the service are wire-compatible with existing Responses API implementations, meaning teams can migrate to the Foundry environment with minimal code changes while gaining access to enterprise security, private networking and observability features, according to Microsoft.
The service ships with production-ready SDKs in Python, JavaScript, Java and .NET, and supports both single- and multi-agent configurations. Microsoft said it also supports direct integration with LangGraph, allowing teams to compose complex multi-agent graphs while keeping individual agents running within Foundry and observable through the platform's Control Plane.
One early adopter making use of the updated platform is Corvus Energy, which operates battery systems across more than 1,500 maritime vessels. The company is using Foundry Agent Service alongside Microsoft Fabric to replace manual inspection workflows with AI-driven operational intelligence.
"The updated interface makes our services and workflows faster, more capable, and easier to use -- and we're now able to connect things that simply weren't possible before," said Andreas Hunderi, VP of Global IT at Corvus Energy.
Voice Comes to Foundry Agents
Among the most notable previews announced at GTC is
Voice Live API integration with Foundry Agent Service, now in public preview. The integration enables developers to build real-time, speech-to-speech agents for customer service, field support and other hands-free scenarios, with audio input and output handled natively and all interactions governed under Foundry's security and monitoring framework.
Gulf Air is already piloting the capability through its Falcon Eye platform, described as a real-time operational intelligence tool pulling live flight, passenger, crew and disruption data into a single command view.
"Executives -- from the CEO to operations leaders -- can speak naturally and receive immediate, accurate spoken answers grounded in live data," said Ahmed Naeemi, Chief Information Officer for Technology and Digital Services at Gulf Air. "Voice-driven navigation can also open dashboards and analyses, turning airline data into a conversational mission-control co-pilot."
Also reaching general availability is observability within Foundry Control Plane, giving engineering teams continuous evaluation of agent performance. Features include out-of-the-box evaluators for coherence, relevance and safety, custom LLM-as-a-judge evaluators for business-specific standards and tracing that captures end-to-end agent execution paths for debugging in production. Microsoft said tracing will formally reach GA later this month. The updated Microsoft Foundry portal at ai.azure.com also reached GA, consolidating agent building, model access and governance into a single interface.
Nvidia Nemotron, Vera Rubin and Physical AI
Beyond the Foundry Agent Service updates, Microsoft laid out several hardware and model-level advances tied to its broader Nvidia partnership.
NVIDIA Nemotron models are now available through Microsoft Foundry, expanding the platform's model catalog. Microsoft also highlighted a recent partnership with Fireworks AI, which allows customers to fine-tune open-weight models like Nemotron into low-latency assets deployable to the edge.
On the infrastructure side, Microsoft announced it is the first hyperscale cloud provider to use NVIDIA's new Vera Rubin NVL72 systems in its labs. The company said it has already deployed hundreds of thousands of liquid-cooled Grace Blackwell GPUs across its global datacenters over the past year.
Microsoft also expanded its Physical AI work at GTC, announcing a deeper integration between Microsoft Fabric and NVIDIA Omniverse libraries. The combination connects live operational data to physically accurate digital twins and simulation environments, a toolchain Microsoft said is targeted at manufacturing, energy and other industries looking to move from monitoring dashboards to AI-driven action across physical systems. A public Azure Physical AI Toolchain GitHub repository was also introduced as part of the announcement.