← Dailies

Daily · 2026-10-13 00:00 UTC

Microsoft's Maia chips go gigawatt; DeepMind's math proofs — 2026-10-13 00:00

Microsoft scales its Maia AI chips to gigawatt levels by 2027, DeepMind's AlphaProof Nexus passes peer review, Epic goes AI-first, Norway fines Microsoft's Xandr, and more.

Source ↗

1. Microsoft plans to scale its own Maia AI chips to gigawatt levels by 2027 — Microsoft will make its in-house Azure Maia chips its main AI inference system across products starting in 2027, deploying them at multi-gigawatt scale, executive Scott Guthrie wrote. Maia 200 already runs Copilot in the US, Europe and Australia, with GitHub Copilot and Foundry to follow by year's end; Microsoft says it pushes up to 1.7x more tokens per watt than leading GPUs in Teams. Successor Maia 300 ships in 2027 and Microsoft claims 40% more tokens per megawatt than leading GPU systems. The Decoder · Oct 12
2. Google DeepMind's AlphaProof Nexus solved open math problems — and got peer reviewed — Per a study in Science, AlphaProof Nexus cracked nine of 353 open Erdős problems (two unsolved for 56 years), proved 44 conjectures from the OEIS integer sequence database, solved an open question in algebraic geometry, and improved a known bound in min-max optimization. Built around Gemini 3.1 Pro, it drafts and revises proofs in Lean, checked after every edit. The peer-reviewed result lands as OpenAI faces a boycott, led by Terence Tao's Association for Human Mathematics, over dumping AI proofs on GitHub. The Decoder · Oct 12
3. AI is less regulated than milk, Microsoft president says — Microsoft president Brad Smith told the BBC that AI faces fewer safety rules than everyday goods: "Right now, your washing machine is subject to more safety rules than AI. Every carton of milk that you see in the grocery store is subject to safety rules and inspection." He called for urgent regulation including a legal obligation to a "victim" if their system is breached by rogue AI agents, and "emergency brakes" — kill switches — to pause or slow a model that grows out of control. The Guardian · Oct 12
4. Epic reportedly delays its Disney game to use more AI — Epic Games has delayed its Disney project and adopted an "AI first" production plan, per Bloomberg-cited sources: one director told the team "No more hand-written code," and ordered all art generated with "full AI workflows." Epic's communications director said the company is "embracing AI-accelerated development." The title had already slipped from November to March. In GDC's annual survey, 52% of developers said generative AI is bad for the industry, up from 30% in 2025. Engadget · Oct 12
5. WindBorne wins a US Air Force contract worth nearly $50 million — The US Air Force will buy data from WindBorne Systems' fleet of roughly 700 autonomous weather balloons under a contract worth up to $49 million — the single-largest in the Redwood City company's history. The Air Force pays only as it orders data; it has so far ordered $1.7 million. WindBorne feeds balloon sensor readings into its own deep-learning weather model, WeatherMesh 6. TechCrunch · Oct 12
6. IonQ to buy cybersecurity startup Elisity for $1.3B — Quantum-computer maker IonQ announced plans to acquire Elisity for $1.3 billion, including $150 million in performance incentives tied to business milestones; the all-stock deal is set to close by year's end. Elisity maps devices on a corporate network and generates safer access policies, with a simulation tool to test them first. IonQ, whose flagship is a 256-qubit machine called Superion 256, plans to add "quantum-related security risk" detection after close. SiliconANGLE · Oct 12
7. This startup says it can generate permit-ready blueprints in minutes — Blueprints AI, a bootstrapped Santa Monica startup co-founded by architect Al Dram and Coco Cao, uses AI to produce code-compliant, permit-ready architectural blueprints from a 2D sketch, photo or PDF. Dram first proposed the idea in a college thesis two decades ago, then launched the concept within six months of ChatGPT's 2022 release. The company was selected for TechCrunch Disrupt's 2026 Startup Battlefield 200. TechCrunch · Oct 12
8. Norway fines Microsoft's Xandr ~$9.6M over Grindr data — The Norwegian Data Protection Authority (NDPA) fined Xandr, Microsoft's adtech subsidiary, 92 million Norwegian kroner — around $9.6 million — its largest fine to date, for allegedly using Grindr users' sensitive data for advertising without obtaining "explicit consent." The NDPA said data was relayed through Xandr to potentially thousands of actors in the ad ecosystem, letting advertisers bid for targeted ads "completely hidden from the individual user." The infringements occurred between July 2018 and April 2020; Xandr has three weeks to appeal. Engadget · Oct 12
9. Both Russia and Ukraine are using AI-powered drones on the battlefield — NPR reports a Ukrainian company that built smartphone-controlled pet cameras now outfits first-person-view drones with an AI image-recognition system, and that both sides in the war are fielding AI-enabled drones. NPR · Oct 12
10. As US medical debt grows, AI will be used more to collect it — The Guardian reports AI is moving deeper into US medical billing and debt collection: taking notes in exam rooms, coding visits, sending billing messages, and powering agents that text patients in billing portals, as medical debt is expected to grow under Republican-led health cuts. More than one in four US adults reported owing medical debt in 2025, per an Urban Institute/Robert Wood Johnson Foundation analysis; analysts at JPMorgan have called AI in coding, billing and collections "early wins." The Guardian · Oct 12

🧪 Try This Next

• 🔧 Look at Microsoft's Maia 300 numbers — "tokens per megawatt" is becoming the metric that matters for AI inference economics.
• 📄 Read the AlphaProof Nexus study in Science — AI solved real open math problems and went through peer review; the process contrast with OpenAI's GitHub dump is the story.
• 📝 Save the Blueprints AI workflow — a template for AI eating document-heavy, regulation-heavy work (permits, compliance, filings).

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.