At Build 2026 in San Francisco, Microsoft took the wraps off a new device platform that signals a decisive shift away from Windows: Project Solara. The company described it as a cloud-managed, AOSP-based platform for agent-first devices, pairing a lightweight edge operating system called Microsoft Device Ec with the full power of Azure AI.

Satya Nadella and the Azure Edge team demonstrated early prototypes on stage, positioning Project Solara as the answer to a growing enterprise hunger for task-specific hardware that operates more like a sentient terminal than a traditional PC. The devices, expected to ship in early 2027, run on Android Open Source Project code but are stripped of Google Mobile Services, replacing them with Microsoft’s own management, security, and AI orchestration layers.

The Announcement at Build 2026

The Moscone Center keynote was already packed with Copilot updates when Pavan Davuluri, Corporate Vice President of Windows and Devices, walked on stage. He held up a slim, tablet-like unit with no visible keyboard or desktop shell. “This isn’t a Windows device. It isn’t an Android device you’d buy off a shelf,” he said. “It’s the first Project Solara endpoint—built for agents, managed from the cloud, and running our new lightweight OS.”

The move caught analysts off guard. For years, Microsoft has been squeezing Windows into ever-smaller form factors: Windows 10 IoT Core, Windows 11 SE, Windows 365 Cloud PCs. Each attempt blurred the line between a local OS and a thin client. Project Solara abandons that line entirely. It embraces AOSP as its kernel and adds a proprietary edge runtime that handles local AI inferencing while the cloud manages identity, policies, and agent workflows.

What is Project Solara?

Project Solara is not a consumer play. Microsoft aims it squarely at frontline and task-specific roles: retail point-of-sale, hospital nursing stations, factory-floor terminals, interactive kiosks, and any environment where a general-purpose PC is overkill. The devices are “agent-first,” meaning the primary interaction model is through AI agents rather than a start menu or desktop. Users speak or type in natural language, and the device executes tasks via a secure agent framework that connects on-device AI with Azure-based large language models.

Hardware partners including Dell, HP, and Lenovo are already building reference designs. These will range from rugged tablets to all-in-one wall-mounted screens, all sharing the same chipset requirements: a Neural Processing Unit (NPU) capable of at least 40 TOPS for local AI acceleration, similar to the Copilot+ PC specs but tuned for edge inference rather than general computing. Microsoft will license the Microsoft Device Ec operating system to OEMs at no cost, monetizing instead through Azure AI consumption and Intune management subscriptions.

AOSP Foundation and Microsoft Device Ec

Choosing AOSP as the base is a calculated risk. It immediately opens a vast catalog of Android-compatible hardware drivers, chipset support, and a familiar developer toolchain. Yet by forking AOSP, Microsoft avoids Google’s licensing restrictions and can tightly control the user experience. The “Microsoft Device Ec” layer—the full name was truncated in official materials but likely stands for “Device Edge” or “Device Ecosystem”—replaces the traditional Android system UI. There is no app launcher, no notification shade, and no back button. Instead, the screen is dominated by a secure, full-screen agent interface that can morph between whiteboard, video conferencing, form input, and more.

Internally, Device Ec includes a hardened Linux kernel with real-time patches, a containerized Android runtime for legacy line-of-business apps, and a new Agent Execution Environment (AEE). The AEE sandboxes AI models and gives them controlled access to hardware sensors, cameras, and microphones. All data processing happens locally unless the agent explicitly routes a request to the cloud, and even then, data is encrypted end-to-end with keys managed by Azure Key Vault.

Agent-First: AI as the Primary Interface

The “agent-first” concept upends the classic desktop metaphor. In a Solara-powered nursing station, a nurse might say, “Show me the vitals for bed 12 and flag any abnormal trends,” and the device calls up a dashboard populated by a clinical agent. If the nurse asks, “Order a CBC and notify the attending,” the agent can trigger backend APIs through Azure Logic Apps, no app switching required. During the Build demo, a retail associate scanned a product and asked, “Do we have this in the back?” The on-device agent cross-referenced inventory, CRM, and even weather data to suggest a similar in-stock item before the customer walked away.

The agents themselves are built with Microsoft Copilot Studio and hosted on Azure AI Foundry. They run partially on-device using quantized models that Microsoft has optimized for the Qualcomm and Intel NPU silicon in Solara hardware. This hybrid architecture allows sub-300ms response times for voice commands while keeping sensitive data local. For more complex reasoning, the agent escalates to a larger model in Azure, bringing back context-rich results without exposing raw data.

Cloud-Managed and Azure AI Integration

Project Solara is designed to be managed entirely through Microsoft Intune, the same endpoint management platform that governs Windows, iOS, and Android devices today. IT admins can define agent policies, push model updates, and wipe devices remotely. However, Solara introduces a new concept: agent personas. An admin can assign multiple agent personas to a single device, each with its own scope of allowed actions and data access. For example, a hospital tablet might have a “Nurse Agent,” “Doctor Agent,” and “Cleaning Protocol Agent,” all running concurrently and switching context based on who authenticates.

Azure AI plays a central role. Every Solara device is linked to an Azure tenant and continuously syncs its AI models, prompts, and orchestrations via the Azure Device Registry. Microsoft provides a library of pre-built agents for common scenarios—inventory management, customer check-in, safety inspections—that enterprises can customize with low-code tools. The platform also supports grounding agents with data from Microsoft Graph, Dynamics 365, and third-party APIs via connectors, making the devices gateways into the broader Microsoft Cloud.

Implications for Windows and Enterprise Endpoints

Project Solara does not mean the end of Windows. Microsoft executives were careful to position it as a companion, not a replacement, for general-purpose computing. “Windows is our brain; Solara is the nervous system at the edge,” Davuluri said. For knowledge workers who need Excel, PowerPoint, and full development environments, a Windows PC or Cloud PC remains the tool of choice. But for the millions of shift workers who interact with a handful of applications, Solara promises a lower-cost, more secure, and more efficient alternative.

This mirrors the trajectory Microsoft took with Windows 365: offloading compute to the cloud while keeping the local device thin. With Solara, the local device becomes so thin that it doesn’t even need Windows. Analysts expect this will cannibalize some low-end Windows laptop sales, but Microsoft may prefer to own that transition rather than lose share to Chromebooks or custom Android tablets. By controlling the OS, the silicon requirements, and the cloud management, Microsoft can offer an end-to-end solution that Apple and Google cannot match in the enterprise.

A New Era for Edge Devices

Project Solara enters a market already crowded with Android-based enterprise devices from Zebra, Honeywell, and Samsung. Microsoft’s differentiator is deep Azure AI integration and the agent-first paradigm. Instead of bolting an agent onto a general-purpose device, the entire system is built around the agent. The lightweight Microsoft Device Ec OS does away with everything that isn’t essential: no legacy Win32 support, no registry, no Group Policy, no domain join. Booting is near-instant, and updates are atomic, applying in under five seconds thanks to an A/B partition scheme borrowed from Android.

Security is another selling point. The operating system runs read-only, with all user data isolated in encrypted containers. Lost or stolen devices are bricks until they check in with Azure and receive a fresh policy. Microsoft claims that the attack surface is 90% smaller than Windows 11, a statistic that will resonate with CISOs tired of patching a full desktop OS on every endpoint.

Challenges and Outlook

The biggest hurdle for Project Solara may be cultural. “Agent-first” requires a mental shift from clicking and typing, and front-line workers already stretched thin may resist learning a new interaction model. Microsoft plans to mitigate this with extensive field trials and a gradual rollout through its own Surface hardware division. The first Solara-powered Surface devices are expected in mid-2027, with third-party OEMs following closely.

Pricing remains unannounced, but the model will likely mirror Azure Arc: device enrollment fees plus per-agent consumption rates. Analysts predict Solara could reduce the total cost of ownership for frontline endpoints by up to 40% over five years compared to traditional Windows laptops, driven by fewer help desk calls, no antivirus overhead, and cloud-based troubleshooting.

While Build 2026 attendees left with more questions than answers, one thing is clear: Microsoft is betting that the future of edge computing isn’t a smaller version of Windows but an entirely new platform designed for a world where AI handles the heavy lifting. Project Solara is the first tangible product of that belief.