At Build 2026, Microsoft doubles down on proprietary AI models to win every layer of the AI stack

During the opening day of its Build developer conference, Nadella previewed Microsoft’s first reasoning AI model, the inaugural model in a new family of MAI models. He also announced Project Soltera, an Android-based software platform that Microsoft describes as “a chip-to-cloud platform designed for an open, multi-agent world that expands how agents are built, deployed, and experienced.” Microsoft also unveiled Microsoft Scout, an “always-on” assistant designed to help users prepare for meetings, manage schedules, and draft emails. The assistant is based on OpenClaw.

Here’s everything Microsoft announced on the first day of Build, and why it matters to developers.

MAI-Code-1-Flash

At its Build developer conference, Microsoft unveiled MAI–Code-1-Flash, its first reasoning model that takes written response from humans and generate a source code for websites and applications. The AI system is very similar to models offered by OpenAI, Google, and Anthropic, and the company jumps on the trend of vibe coding in a big way where AI is being used to write a code.

MAI-Code-1-Flash is part of MAI-Thinking-1, Microsoft’s first reasoning model. The model is medium-sized (35 billion active parameter model with a 128K context window) and is “built for high efficiency and performance, but importantly, at a low token cost,” according to Kyle Daigle, Microsoft’s chief of developer marketing and chief operating officer of GitHub, in a blog post. MAI-Thinking-1 is available in Microsoft Foundry as a private preview.

Microsoft also unveiled other in-house models for generating images, transcribing audio, creating and synthetic voices.

What are AI tokens?

Tokens are the basic building blocks that AI models use to process and understand text. Think of them as invisible units that determine how AI systems read, process, and generate language. The number of tokens a model consumes directly affects the costs developers incur when using AI models.

Project Soltera

Perhaps the announcement that took everyone by surprise was the launch of Project Soltera, an Android-based operating system designed to run multiple AI agents in a secure environment. The platform is essentially designed to power a new generation of business-focused devices that users interact with primarily through artificial intelligence agents. Steven Bathiche, a technical fellow in Microsoft’s Applied Sciences Group, previewed a device resembling an employee badge, featuring Wi-Fi, Bluetooth, and wireless connectivity, along with a touchscreen display and a fingerprint reader.

A second device, a size of an Amazon Echo Show and Google Next Hub is a compact desktop unit with a built-in screen. Both devices are designed to showcase how AI agents could become the primary interface for future workplace computing.

Microsoft Scout

Nadella also unveiled Scout, an “always-on” assistant, during the keynote at the inaugural Build developer conference. Scout is essentially an AI agent designed to help users prepare for meetings, manage schedules, and draft emails. It is based on OpenClaw, the open-source software that surged in popularity late last year. For now, Scout is available only to a limited group of customers. The assistant bears a striking resemblance to Gemini Spark, Google’s own autonomous AI agent, which was unveiled last month at the I/O developer conference and is currently available only to premium subscribers in the United States.

Eyes quantaum computing

Microsoft also showed off Majorana 2, a next-generation topological quantum chip. The chip is way more capable and reliable than its predecessor, paving the way for a quantum computer that could solve commercially useful problems within three years. In fact, the company claims the new chip is 1,000 times more reliable than its predecessor. Microsoft says the qubits on Majorana 2 survive for an average of 20 seconds, compared with just milliseconds on Majorana 1. There is a global race to develop quantum computers, and the potential is enormous.

Microsoft also teased the Surface Pro Ultra, the high-end laptop powered by Nvidia’s first PC chip. (image credit: Microsoft)

Microsoft wants to own the ‘AI stack’

Microsoft, a heavyweight in the technology industry for decades, is doubling down on its proprietary AI models as it looks to lower costs and pass those savings on to developers. As the cost of using leading AI models continues to rise, Microsoft sees a financial advantage in building and running its own systems. By deploying these models on its Azure cloud infrastructure, the company can reduce its reliance on third parties such as OpenAI. The strategy mirrors what Google has done with its Gemini Flash models, which can write code, perform a range of tasks, and run on the company’s own infrastructure. More and more, technology giants such as Microsoft are thinking to own every layer of the AI stack as competition intensifies – not only from AI startups such as OpenAI and Anthropic, but also from established Silicon Valley rivals like Google.

Nvidia’s launch of an AI-focused PC chip this week is another sign of this broader industry trend. The world’s most valuable chipmaker is looking beyond the data center and toward edge devices such as personal computers, enabling them to run AI workloads locally rather than relying entirely on the cloud. It is a long-term strategy aimed at gaining control over, and ultimately competing across, every layer of the AI stack.