Microsoft's AI Wrecking Ball: Scout, MAI Thinking 1, Chips, Quantum Claims—and Canada's $2B AI Plan

In this Project Synapse episode, Jim, Marcel Gagné, and John Pinard unpack Microsoft's sudden wave of AI announcements, including seven new in-house models such as MAI Thinking 1 (a 35B-parameter reasoning model trained from scratch on "clean" data), a Copilot replacement or reorganization called Scout, and Microsoft's Maia inference chips while still training on NVIDIA hardware. They debate whether Microsoft is finally moving beyond rebranding OpenAI/Anthropic models, discuss agent security concerns around Scout's OpenClaw/MCP foundations, and touch on the competitive push toward cheaper coding tools. The conversation broadens to quantum computing claims, data-center overbuild versus efficient small models and local inference, Anthropic's IPO valuation and Mythos/Glasswing security work, looming AI regulation challenges, robotics progress, and Canada's new AI strategy promising $2B, a supercomputer by 2031, and major adoption goals that they argue lacks implementation detail. 00:00 Cold Open Banter 00:19 Microsoft Drops New AI Stack 01:28 MAI Model And Clean Data 03:23 Copilot Confusion And Privacy 05:39 Maia Chips And Frontier Ambitions 12:14 Scout Agent And MCP Security 17:17 Microsoft Distribution And AI Economy 24:13 Apps Dying And Office Rivalry 27:20 Quantum Chip Shockwave 31:01 Data Centers Versus Small Models 35:51 NVIDIA RTX Spark Local AI 40:17 Software Overcapacity And CRM Threat 42:00 Why Software Gets Huge 44:41 DIY Simple Writer Demo 46:04 AI Note Taking Gadgets 48:47 Anthropic IPO Valuation 56:15 Mythos And Zero Days 59:25 Regulating AI Everywhere 01:04:24 Robots And China Scale 01:06:46 Canada AI Strategy Critique 01:15:56 Open Source Canada Plan 01:18:33 Hopeful Wrap Up