Majorana 2: Real Progress, But Still Temper Your Expectations

Microsoft's Majorana 2 quantum chip is the headline that matters long-term.

According to The Verge, Microsoft technical fellow and CVP Chetan Nayak says qubit lifetimes jumped from 1–12 milliseconds on the aluminum-based Majorana 1 to over 20 seconds on Majorana 2 — with some exceeding a full minute. That's the 1,000x stability improvement Microsoft is claiming.

The material change is specific and verifiable: aluminum superconductor swapped for lead, semiconductor active region updated to a combination of indium arsenide and indium arsenide antimonide. This is a testable engineering claim.

Microsoft says this progress is enough to accelerate its roadmap to a "scalable, practical quantum computer."

Context worth noting: physicists were openly skeptical of Majorana 1 last year. The Verge itself notes that physicists "immediately" questioned Microsoft's previous claims. Majorana 2 is genuinely impressive on paper, but the broader scientific community hasn't had a chance to independently verify these new numbers yet. Microsoft is asking the world to trust its own benchmarks — again. Progress is real. Independent validation is still pending.

Seven New AI Models — Including Microsoft's First In-House Reasoner

Microsoft dropped seven new AI models at Build, according to The Verge's Jay Peters. The flagship: MAI-Thinking-1, described as a "medium-sized" advanced reasoning model.

Microsoft says it "trained it from the ground up on clean data, without distillation from third-party models." That's a direct shot at rivals who build derivative models on top of existing ones. It also matters because Microsoft and OpenAI recently renegotiated their partnership to loosen ties — MAI-Thinking-1 is the most visible sign of that independence.

The other six models:

• MAI-Image 2.5 (and flash version): text-to-image and image editing
• MAI-Transcribe-1.5: claims to be five times faster than competing transcription models
• MAI-Voice-2: adds 15 new languages and new voice options
• MAI-Voice-2 Flash: listed as "coming soon"
• MAI-Code-1-Flash: inference-efficient coding model integrated into GitHub Copilot and Visual Studio Code

Seven models in one day is aggressive. Whether the benchmarks hold up against Google and Anthropic in real-world use — not just Microsoft's own "key" benchmarks — remains to be seen.

Scout: Microsoft's Always-On Personal Assistant

Microsoft Scout is the company's entry into the growing agentic AI assistant space.

Scout VP Omar Shahine described it to The Verge as "the first real personal assistant" Microsoft has offered customers — not a chatbot, not a toolbar feature. An always-on agent that reads your Teams threads, monitors traffic, manages your calendar, drafts emails, books travel, and fills out forms.

Three thousand Microsoft employees are already using the desktop preview internally, according to The Verge.

Right now it's limited to Frontier program customers in the US, requires a GitHub Copilot subscription, and only runs as a desktop app. Full cloud rollout comes later.

The privacy question deserves attention: this system reads your emails, calendar, and chat history in the background, continuously. TechCrunch notes Scout includes a built-in "policy conformance system" with audit trails. Microsoft adding audit trails is the right call. Users should understand exactly what they're opting into.

Project Solara: Microsoft Builds an Agent OS on Android, Not Windows

Microsoft announced Project Solara — a new operating system for AI agent hardware — and it runs on Android, not Windows. Specifically, the Microsoft Device Ecosystem Platform, a fork of Android optimized for low-power devices.

Microsoft showed off two concept devices: a desk unit (think Amazon Echo Show with facial recognition) and a wearable badge with a camera and fingerprint scanner. Companies including AccuWeather, Best Buy, CVS Healthcare, and Target are already lined up for pilots, according to The Verge.

Microsoft isn't shipping these devices itself. They're reference designs for other hardware makers. The competitive context: Google, Meta, and OpenAI (partnered with Jony Ive) are all building AI hardware. Microsoft is staking out enterprise turf before that race heats up.

The choice of Android over Windows for this platform signals that Windows is too heavy for the next category of AI devices. That's a significant strategic shift.

Surface RTX Spark Dev Box and the Linux-Friendly Windows

For developers, two concrete announcements landed June 2nd.

The Surface RTX Spark Dev Box — powered by Nvidia's Arm-based RTX Spark chip — packs 128GB of unified memory and a 100-watt thermal envelope into a chassis that looks like the top of an Xbox Series X. It can run models up to 120 billion parameters locally. Price and full specs aren't out yet; availability is "later this year" in the US, according to The Verge.

This is a direct replacement for Qualcomm's previously announced Snapdragon Dev Kit, which was delayed and ultimately canceled.

On the software side, Microsoft announced Coreutils for Windows — native Linux-style command-line utilities built in Rust from the uutils open-source project — plus deeper Windows Subsystem for Linux integration through WSL containers, and a new Intelligent Terminal that can interact with AI agents via the Agent Communication Protocol. Windows chief Pavan Davuluri framed it as making Windows a trusted development platform regardless of whether a developer lives in Linux, macOS, or cloud environments.

ASSERT: The Unglamorous Tool That Actually Matters for AI Safety

The least flashy announcement could shape enterprise AI deployment.

Microsoft released ASSERT — Adaptive Spec-driven Scoring for Evaluation and Regression Testing — an open-source framework that turns plain-language descriptions of how an AI should behave into automated test cases, according to TechCrunch.

Microsoft's chief product officer of Responsible AI, Sarah Bird, put it plainly: "If you don't understand the behavior of the AI system, it's really hard to know if it's meeting your organization's bar."

ASSERT lets developers define rules — say, "don't share confidential data outside the C-suite" — and continuously tests whether the deployed system actually follows them. That's not exciting to write about. It's exactly what enterprise AI deployment needs.

What Microsoft Showed at Build

Microsoft came to Build 2026 with hardware, shipping software, new models, and a quantum chip that — if the numbers are independently verified — represents a genuine leap forward.

The FTC probe hanging over the company didn't get a mention onstage. It never does. But the scale of what Microsoft is building — an always-on personal assistant reading your communications, an agent OS ecosystem, seven proprietary AI models — means the regulatory scrutiny isn't going away.

Microsoft is betting everything on AI infrastructure. The products are real. The risks are real too.