Microsoft used its Build 2026 developer conference to roll out seven in-house MAI models, the always-on assistant Scout, the Android-based agent operating system Project Solara, Microsoft Execution Containers, and the next-generation quantum chip Majorana 2. The announcements pointed in one direction: Microsoft is building a broader AI stack under its own control instead of relying mainly on OpenAI-powered services.
MAI family expands beyond a single flagship model
The headline model was MAI-Thinking-1, which Microsoft described as its first reasoning model. According to the company, it runs with 35 billion active parameters and supports a 128K context window. Microsoft also introduced related models for image understanding, speech synthesis, code generation, and automatic speech recognition, filling out a wider model family across input and output tasks.
That marks a visible shift from Microsoft’s earlier AI commercial approach, where Azure OpenAI Service was central to the offering. Rather than mainly licensing OpenAI models and packaging them through enterprise APIs, Microsoft is now presenting its own model layer as a core part of the product stack.
Surface RTX Spark Dev Box targets on-device AI work
On the developer hardware side, Microsoft introduced the Surface RTX Spark Dev Box, a mini PC designed to run AI models locally. It uses Nvidia’s new Arm-based Spark RTX chip and comes with 128GB of unified memory, with VS Code and GitHub Copilot preinstalled. Microsoft said the device is expected to launch in the United States later this year, though pricing and full specifications were not disclosed.
The company also paired the hardware with updates aimed at making Windows 11 more usable for AI developers. These include Coreutils, described by Microsoft as Linux-style command-line tools running natively on Windows 11, direct Linux container execution through WSL, and an Intelligent Terminal that feeds current context into developers’ preferred AI agents.
Scout and Solara split the agent strategy in two
Microsoft introduced two distinct agent paths. The first is Scout, an always-on personal assistant built on the open-source AI platform OpenClaw. It can work with Outlook, OneDrive, and Teams, handling tasks such as calendar organization, expense processing, and email drafting in the background. Microsoft placed it under its “Autopilot” line of agents, with a desktop preview currently open to Frontier customers in the United States.
The second is Project Solara, a larger operating system effort based on Android and designed for AI agents that move across devices. Microsoft said the project is being developed with Qualcomm and MediaTek. In the demo, Solara appeared as both a companion device for PCs and a system that can hand off tasks between endpoints, including a desktop hub and a digital badge form factor.
Security controls and quantum computing were part of the same push
As Microsoft pushes agents deeper into background workflows, it also highlighted security controls. Microsoft Execution Containers (MXC) are meant to run OpenClaw agents inside sandboxed environments, letting developers define which device resources, folders, and APIs each agent can access. Microsoft also introduced a companion OpenClaw app that allows users to build agents or connect to existing ones inside the same controlled setup.
On the hardware front, Microsoft announced Majorana 2, its next-generation quantum computing chip. The company said its qubit accuracy is 1,000 times higher than the previous generation, attributing that improvement to a new material stack using lead compounds. Microsoft said the current roadmap targets a “practical quantum computer” by 2029, while noting in effect that the performance claim has not yet been independently verified by a third party.
Build 2026 was less about a single product and more about how Microsoft wants its AI pieces to fit together. Models, developer machines, agent software, security controls, and quantum hardware were presented as parts of one vertically integrated system.