The Microsoft Build 2026

Sign Up here | Advertise with Us

Hey AI Geeks,

We weren't even done processing NVIDIA's GTC and Computex drops — and Microsoft just walked in with Build 2026 like it was nothing.

Satya Nadella basically declared that the "just chatting with AI" era is over. New chips, a new category of purpose-built agent devices, seven new models, and a quantum chip the size of a credit card. If you missed it, you're welcome.

Let's get into it. ⬇️

🏆 Before the tech — someone won $150,000

Microsoft opened with the Imagine Cup finals and quietly bumped the grand prize from $100K to $150K mid-event. The winner was Copy Flag, an AI tool that detects AI-modified copyright infringement and makes it visible to creators. Founder Patrick summed it up well: "turning an invisible problem into a visible one."

Other finalists worth knowing:

• Spoil Safe: predictive sensors that tell supply chains when food is about to go bad

• Revora Health: AI that matches patients to the right physical therapy program

💻 Windows is getting a brain of its own

Microsoft wants AI running locally on your device, not just in the cloud. Two big moves here:

• ION Models: two new small AI models built directly into Windows. One for reasoning tasks, one for running local agent loops. No internet required

• RTX Spark + Surface Ultra: built with NVIDIA, this is a single chip with a 20-core CPU, Blackwell GPU, and 128GB of unified memory delivering 1 petaflop of AI compute. Lands in the Surface Ultra laptop and a developer box this fall

🛠️ Windows for developers just got a serious upgrade

If you build software on Windows, this section is for you:

• Intelligent Terminal: your terminal now has a built-in agent that spots errors and suggests fixes in real time

• 70+ Linux utilities natively in Windows: grep, tail, Zshell, Homebrew, all of it, no workarounds

• WSL Containers: spin up and attach to native Linux containers without any third-party tools

• Git Worktrees: run multiple AI agents in parallel on the same codebase without them stepping on each other. This one is genuinely useful

📟 Project Solara: AI devices that aren't phones or laptops

This is the most "wait, what?" announcement of the event.

Microsoft is letting enterprises build purpose-built agent-first devices from scratch — not retrofitting AI into existing hardware. They showed two prototypes:

• A stationary desk companion (MediaTek chip) that handles your workday context

• A wearable digital badge (Qualcomm chip) for healthcare workers to dictate notes and scan vitals hands-free

The category didn't exist before this announcement.

☁️ Azure got significantly more serious

Agents query databases millions of times per second. Old infrastructure breaks. Microsoft's answer:

• Azure Cobalt 200 VMs: first-party ARM chips, 50% faster and 33% lower latency for agent calls

• Maya 200 GPUs: paired with Cobalt for heavy compute workloads

• Azure Horizon DB: a managed PostgreSQL service that goes up to 128TB and lets you call AI models directly via SQL. No separate pipeline needed

• Rayfin SDK: open-source, lets you vibe-code an app and immediately deploy it to an enterprise-grade backend

🔍 Microsoft IQ: Giving agents a sense of reality

Models hallucinate when they don't have context. Microsoft built a whole layer to fix that:

• Web IQ: ultra-fast search built specifically for grounding AI, not for humans to browse

• Fabric IQ: turns your business data into a live model that agents can reason over

• Work IQ: grounds agents in your internal documents, people, and company procedures

Think of it as giving your agents institutional memory.

🔒 Security for agents that actually do things

An agent that writes code and modifies files is a security nightmare if you don't think about it carefully. Microsoft did:

• MXC (Microsoft Execution Containers) - OS-level sandboxing so agents can't touch what they're not supposed to. OpenClaw runs natively inside it

• Agent 365 - a control panel for IT admins to monitor, govern, and approve what autonomous agents are doing across the company

• Mdash - 100+ specialized security agents that debate each other to find real vulnerabilities in your codebase before attackers do

🤖 GitHub Copilot is now basically a full IDE

The GitHub Copilot App is no longer just autocomplete, it's a standalone super-app with a terminal, browser, and canvas UI built in. The headline feature is Agent Merge, it watches your PRs through CI checks and merge conflicts and handles them autonomously.

And Autopilots are now a thing, long-running agents that manage your Outlook and Teams on your behalf. The first one, Scout, is live now.

🧠 Seven new MAI models, and one of them is a serious contender

Mustafa Suleiman's team dropped seven models. The one to pay attention to:

• MAI Thinking 1: 35B parameters, 256K context window, scores 53% on SWEBench Pro (in the same range as Claude Opus 4.6), and was trained without distillation. That last part matters, it learned from scratch, not by copying another model's outputs

• MAI Code 1 Flash: 5B parameters, optimised for VS Code, absurdly efficient

• MAI Voice 2: natural voice with emotional control built in

• MAI Transcribe 1.5: 43 languages, claims state-of-the-art accuracy

• MAI Image 2.5: high-fidelity image generation and editing

Frontier Tuning rounds it out, a platform that lets companies build reinforcement learning environments on their own private data, so a small 5B model can outperform much bigger generalist models on your specific tasks.

⚛️ Quantum computing just became an engineering problem

Two announcements that deserve separate attention:

• Microsoft Discovery: an agentic loop that automates the scientific method. Demoed discovering new proteins to recycle plastics indefinitely, treating physical lab experiments like software builds

• Majorana 2: a quantum chip the size of a credit card with a million qubits, 1000× longer qubit lifetime, and operations at 1 microsecond. Quantum computing stopped being a physics experiment and started being an engineering one

And that's just Day 1.

We'll be back tomorrow, same time, with everything from Day 2. Here's what to expect:

• Ollama's breakout session: Michael Chang and Parth from Ollama are doing live demos of models running agentic tasks in real time. If you want to see this stuff working, not just announced, this is the one to watch

• Fireworks AI: they're presenting alongside their team on Wednesday and it sounds like there's something to show, not just talk about

Day 1 was chips, models, OS rewrites, and a quantum breakthrough. Day 2 is where the builders show up. See you tomorrow. 👋

And that’s a wrap!

Rate this issue and reply to this email to tell us if you liked this and would want more such deep dives! (the more specific you are, the better we can deliver!)👇️

We’ll be back soon with more spicy takes on What’s Happening in AI, so stay tuned & share our newsletter with a friend! Stay Tuned! 🤖 

Team What’s Up in AI