Microsoft broadens Foundry as rivals push harder on models and enterprise AI

Microsoft turned Azure AI Foundry into the clearest product story of the week by widening both its own model lineup and the menu around it. The platform added OpenAI’s GPT-chat-latest (GPT-5.5 Instant), brought in IBM Granite 4.1, NVIDIA Nemotron Nano Omni and Qwen 3.6-35B-A3B, and rolled out GPT-realtime-translate, GPT-realtime-whisper and GPT-realtime-2 through the Realtime API. That matters because Microsoft’s enterprise pitch increasingly depends less on one flagship model and more on being the place where companies can mix leading chat, speech, vision and realtime tools without leaving Azure.

The competitive edge here is flexibility, but it also suggests a continuing dependency. Several of the additions came from outside Microsoft, including OpenAI and third-party model providers, which is good for customer choice but means Foundry’s momentum still relies on keeping outside partners close and shipping their updates quickly.

OpenAI is moving beyond general chat and putting more pressure on Microsoft in high-value specialist work. GPT-5.5-Cyber was rolled out in a limited preview for vetted cybersecurity teams, with a design that is more permissive on tasks such as vulnerability identification, patch validation and malware analysis, and OpenAI also launched a Codex Chrome extension that lets the coding agent operate across browser tabs. The immediate significance is that the competitive fight is shifting from broad assistant claims to domain-specific tools that can sit directly inside security and developer workflows.

For Microsoft, this raises the bar for both Azure security offerings and developer tooling. GitHub Copilot and Microsoft’s broader Copilot stack now face a rival that is narrowing in on concrete use cases rather than just general capability, which usually makes enterprise buying decisions easier.

Cloud rivals are also tightening the enterprise AI race instead of leaving Microsoft to define it. Oracle launched OCI Enterprise AI with support for Grok 4.3 and Nemotron 3 Nano Omni, while AWS previewed Amazon Bedrock AgentCore Payments with Coinbase and Stripe so AI agents can pay for APIs, web content, MCP servers and other agents. These are different moves, but together they show competitors trying to make their clouds look like full AI operating environments rather than simple model hosting layers.

This matters for Microsoft because Azure now competes on workflow completeness as much as on raw model access. If Oracle can pair frontier models with its own enterprise stack, and AWS can make agents transact in real time, Microsoft cannot rely on Azure scale alone - it has to keep filling functional gaps around agents, tools and deployment patterns.

The broader model market stayed busy enough to reinforce that pace, not just capability, is becoming a competitive weapon. xAI released Grok 4.3 with agentic tool calling, non-reasoning mode support and a 1 million-token context window, then said older Grok models will be retired on May 15, 2026 in favor of the new version. In parallel, Anthropic doubled Claude Code rate limits, removed peak-hour caps, raised API rate limits for Claude Opus and secured over 300 MW of GPU capacity at Colossus 1.

The message for Microsoft is straightforward: enterprise AI customers are being trained to expect fast model turnover, rising service limits and visible compute backing. That helps explain Microsoft’s own emphasis this week on catalog breadth and deployment guidance, including its Global AI Diffusion Report for Q1 2026 and new digital sovereignty messaging, even if those policy-facing items are more context than fresh product news.