11 startups to watch at Microsoft Build 2026

From prototype to production: the new standard for AI-first startups

The conversation has shifted. A year ago, founders came to Microsoft asking whether to build with AI. Today at Microsoft for Startups, the question we hear most is how to make AI work in production, at scale, across real enterprise systems, and that shift shapes everything about this year's Build cohort.

For the first time in nearly a decade, Microsoft Build is leaving Seattle. The move to San Francisco is deliberate: this is where the AI infrastructure ecosystem is densest, where the startups redefining developer tooling, compute, observability, and data are doing their sharpest work, and where that work is closest to the enterprise buyers it needs to reach. This year's event reflects a sharper technical focus than ever, organized around the engineering decisions that actually determine whether AI ships, scales, and holds up under real conditions.

The startups featured here reflect that agenda. They're not building proofs of concept. They're working on the hard, unglamorous problems between a good idea and a system that actually runs: how to authenticate agents without exposing credentials, how to understand legacy code well enough to modernize it safely, how to know whether the AI tools your engineering team adopted are actually making them faster. These are the questions that matter in 2026.

The infrastructure moment

AI adoption has followed a predictable arc: early experiments and pilots gave way to a harder reality, where models hallucinate, agents drift, data is sensitive, and costs compound faster than anyone budgeted for. Enterprises learned quickly that shipping AI to production requires a fundamentally different class of tooling than shipping to a demo environment, and that gap has driven a surge of investment in the infrastructure layer.

The needs are specific, and they cut across the entire stack. On the data side, that means databases purpose-built for multimodal retrieval at scale and synthetic data platforms that give developers production-quality data without exposing anything sensitive. On the infrastructure side, it means compute frameworks that scale distributed AI workloads without a dedicated team, and observability platforms that bring the same visibility to language model behavior that engineering teams have always had into application performance.

Microsoft Build 2026 is organized around exactly this terrain, covering agentic retrieval-augmented generation (RAG) architectures, cost-optimized model deployment, Foundry IQ for agent-ready context, and go-to-market paths through Microsoft Marketplace, with the full arc from early traction to enterprise scale in view.

11 startups leading the way at Microsoft Build 2026

The startups joining us at Microsoft Build 2026 this year span developer tooling, AI infrastructure, and physical AI, and what they share is a bias toward solving real enterprise problems rather than interesting-sounding ones. Many are Microsoft for Startups Pegasus Program members or backed by M12, and all are available directly on e, making it straightforward for commercial customers to procure, deploy, and scale their solutions within the Azure ecosystem. Here's who to watch.

Engineering teams spend too much of their time responding to incidents instead of building. NeuBird's Hawkeye changes that equation: an agentic site reliability engineer (SRE) that interprets telemetry from across your observability stack, diagnoses issues, and drives resolution in minutes rather than hours. In 2025, customers used it to resolve 230,000 alerts and reclaim 12,000 engineering hours.¹ A Microsoft ISV Success Partner, Pegasus Program member, and M12-backed company, NeuBird was the first generative AI-powered SRE to land on Marketplace.

The premise behind Replit is straightforward: anyone in an organization, not just engineers, should be able to turn an idea into working software. With more than 500,000 business users on the platform, that bet is paying off.¹ Replit's partnership with Microsoft integrates its agentic development platform with Azure Container Apps, Azure Virtual Machines, and Neon Serverless Postgres, and makes it available directly through Marketplace. The result is enterprise-grade application creation without the traditional engineering bottleneck.

Ray is the open-source distributed compute framework powering AI workloads at Uber, Spotify, Canva, and dozens of others, with more than 27 million monthly downloads.¹ Anyscale, the company behind Ray, partnered with Microsoft to co-develop a fully managed, first-party Azure service built on it: a high-performance environment for training, inference, and data processing that runs natively inside your Azure Kubernetes Service (AKS) environment, with unified billing and up to 10 times faster performance compared to self-managed Ray.¹

Most AI coding tools help engineers write new code. Moderne solves a different problem: the millions of lines of existing code that still need to be understood, maintained, and modernized. Its platform automates large-scale refactoring across thousands of repositories simultaneously, built on OpenRewrite, the open-source framework Moderne's CEO originally developed at Netflix and now integrated into GitHub Copilot. Squarespace, Allstate, and five of North America's top banks are using it to eliminate technical debt at a scale that would otherwise require tens of thousands of manual engineering hours.

Before you can modernize legacy code, you have to understand it. CoreStory's Code-to-Spec platform uses agentic AI to analyze millions of lines of existing code and generate living documentation that captures business rules, system relationships, and developer intent automatically. What once required 18 months of manual review now takes days.¹ In research published jointly with Microsoft, using CoreStory's structured specifications inside AI software engineering agents improved accuracy by 51%,¹ a meaningful compression of modernization timelines for enterprise teams. If you're joining us in San Francisco, stop by the Marketplace experience in the Microsoft Showcase at the Festival Pavilion to see it in action.

AI coding tools are everywhere. Knowing whether they're moving the needle is a harder problem, and it's the one Faros AI is built to solve. Its engineering intelligence platform aggregates data from more than 100 tools, GitHub Copilot among them, giving engineering leaders a single source of truth on productivity, delivery, and AI ROI. Named the 2025 Microsoft for Startups Partner of the Year from a field of more than 4,600 nominations,¹ Faros is available on the Marketplace with Azure benefit-eligible procurement, so customers can apply 100% of the purchase toward their cloud spend commitment. The Faros team will also be at the Marketplace experience in the Microsoft Showcase, Festival Pavilion.

Most AI agents fail in production not because the models aren't capable, but because they can't securely act on the business systems they need to touch. Arcade.dev is the model context protocol (MCP) server runtime that solves this: providing the authorization, reliability, and governance infrastructure that lets agents take real actions across Microsoft 365, Github, Teams, Salesforce, Jira, and hundreds of other enterprise tools, without exposing credentials to the model. It's the control layer between capable agents and the systems they need to reach.

The gap between working prototype and production-ready robot remains a defining challenge as the tools, frameworks, and systems required to make robots intelligent were never built to work together. General Robotics is building a unified intelligence grid for physical AI: a cloud-native platform for composing AI skills for perception, planning, and action that can be simulated and deployed more quickly across any robot form factor. Their platform GRID makes robot intelligence more accessible, providing API-first access for developers and an agent-first experience for robot operators.

Multimodal AI applications need a database built for the job, and most weren't. LanceDB is an open-source, AI-native lakehouse built for billion-scale vector search across text, images, video, and audio, with a compute-storage separation architecture that runs natively on Azure Storage. It integrates with LangChain, LlamaIndex, and DuckDB, and companies like Midjourney, Runway, and Character.ai are running it in production.¹ LanceDB raised a $30 million Series A in 2025 to accelerate its enterprise platform.

Shipping an AI model is one thing. Knowing how it's actually behaving in production is another. Arize AI is the observability and evaluation platform built for that second problem, giving engineering teams the same visibility into language model behavior that they've always had into application performance. With more than two million monthly downloads of its open-source library Arize Phoenix¹ and deep integrations with Microsoft Foundry and Azure AI Studio, Arize is the monitoring layer enterprises are building on. M12, Microsoft's venture fund, participated in the company's $70 million Series C.

The data problem in enterprise AI is specific: the most useful data for training and testing is also the data that's most restricted. Tonic AI resolves that tension with high-fidelity synthetic data that mirrors production quality without touching anything sensitive, installing directly into a customer's Azure tenant with support for Azure OpenAI, Microsoft Fabric, and Azure SQL. A Pegasus Program member with Azure benefit-eligible procurement through Marketplace, Tonic is already helping teams at Comcast, eBay, and UnitedHealthcare move faster without compromising compliance.

The AI infrastructure layer is being built right now, and the startups featured here are doing some of the most important work in it. Whether you're joining us in San Francisco or streaming online, Microsoft Build 2026 is where those conversations happen.

Register for Microsoft Build 2026

Want to meet the startups behind the story? Join us at Dev Your Own Way, hosted by Microsoft for Startups on June 2, 2026, in San Francisco, California. You'll connect with founders, Microsoft Build attendees, technical experts, and leaders from across the startup and developer ecosystem.

Register for Dev Your Own Way

¹ Data reported by individual companies.