Microsoft makes ‘biggest bet yet’ on own silicon

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

💻 Microsoft makes ‘biggest bet yet’ on its own silicon

🛩️ Russian bot traffic collapses after drone strikes

🛑 Microsoft CEO wants a human brake on AI

📈 IMF credits AI spending for near-3% growth

📚 Publishers quietly use AI for book covers

🧊 Cyberattack disables supermarket fridges across Europe

Plus: 🎁 12 more stories you might like, 🧰 6 tools, and 📚 5 papers.

💻 Microsoft makes ‘biggest bet yet’ on its own silicon LINK

• Microsoft has unveiled Maia 300, the third version of its own AI chip, and plans large-scale rollouts in 2027 as it aims to run most of its AI services on homegrown silicon rather than Nvidia hardware.
• The company says Maia 300 produces 40% more AI output for the same power as leading Nvidia GPU systems, and a full rack delivers up to seven times the output of the current generation on tasks like AI agents and long documents.
• Microsoft designs the chips and has TSMC make them, calling Maia its "primary AI inference system," though unlike Google and Amazon it won't rent them to outside customers, and analysts expect Maia to cover only about 10% of its AI chip spending by early 2027.

🛩️ Russian bot traffic collapses after drone strikes LINK

• Russian bot traffic has fallen sharply in recent days after Ukraine destroyed Yandex's two largest data centers, according to Cloudflare, which tracked bot requests from the Yandex network dropping to just 0.4% of the recent peak.
• The decline began on October 8 when Ukraine struck a Yandex site in Sasovo holding two supercomputers, falling to around 57%, before a bigger campus in Kaluga was knocked offline yesterday, leaving at least 84 Yandex Cloud services down.
• NetBlocks found Yandex network traffic dropped to 30% for Yandex Cloud and 63% for Yandex LLC, likely slowing or cutting internet service for ordinary Russians, who cannot fall back on Starlink since it does not work in Russia.

🛑 Microsoft CEO wants a human brake on AI LINK

• Microsoft CEO Satya Nadella argued in an essay published on Saturday that AI models should run under outside controls, including an emergency brake that an authorized person could pull to pause or shut down a model mid-task.
• Nadella's proposed design separates the model from the software coordinating its work and the actions it can take, so an organization sets permissions independently and keeps tamper-proof records, though he gave no rollout timetable.
• His essay follows Microsoft's October 7 release of Microsoft Execution Containers, software that limits an agent's file, network, and interface access from outside, so the model or its code cannot grant itself more access.

📈 IMF credits AI spending for near-3% growth LINK

• IMF chief Kristalina Georgieva said that spending on AI infrastructure is pushing the world economy to grow about 3% a year, even with high energy prices weighing on many countries.
• She said Asian makers of AI hardware gain the most: advanced economies could add 0.5 to 1 percentage point of growth, while emerging ones might pick up 0.2 to 0.8 points, with Thailand among the winners.
• Major tech companies plan to spend between $660 billion and $690 billion on AI this year, and such hardware and related products now make up more than a tenth of global trade in goods, Georgieva said.

📚 Publishers quietly use AI for book covers LINK

• A new Wired report, based on more than two dozen anonymous staff, says three of the big five publishers are quietly using AI for cover art and publicity copy without telling authors or the public.
• Publishers are reportedly generating PR material and cover copy with ChatGPT and Claude, while image generators produce book covers, and executives are exploring AI for translation, internal documents, and tracking employee productivity.
• One agent said they now make publishers agree not to feed entire unpublished manuscripts into language models for promotional text, a safeguard meant to shield authors from both tech firms and their own publishers.

🧊 Cyberattack disables supermarket fridges across Europe LINK

• A cyberattack has hit major European supermarket chains, breaking refrigeration controls by forcing fridges and freezers into a nonstop defrost cycle set to a sweltering 50 degrees Celsius, La Dépêche reports.
• The attack targeted Danfoss-made refrigeration systems at French chains Intermarché, Leclerc, and Coopérative U, whose president called it "extremely serious," with early losses nearing 300,000 euros and possibly reaching several million nationwide.
• A separate incident on Sunday hit Dutch chains Albert Heijn, Etos, and Gall & Gall, taking down websites, disrupting deliveries, and tampering with electronic shelf labels and prices; it is unclear whether the two attacks are linked.

Other news you might like

• Console market craters as Xbox sales crash 33% to the lowest since launch 25 years ago, PS5 sales drop 25% LINK

• Mt. Gox’s Disgraced Ex-CEO Finds Redemption Exposing ‘Spy Implants’ in Ledger Devices LINK

• IKEA turns to driverless semis with nobody in the cab LINK

• Fingerprints are the new cookies LINK

• if your team is too busy doing their 'normal job' to experiment with AI, you're preparing them to be replaced LINK

• Batteries are now cheaper than natural gas turbines used at many data centers LINK

• Japan declares cyber ‘emergency’ after millions affected in hacks LINK

• Google announces the Fitbit Edge, a middle ground between Pixel Watch 5 and Fitbit Air LINK

• The U.K. won back access to pre-release OpenAI and Anthropic models LINK

• Creator of free Adobe clones unveils open-source Microsoft Office Excel, PowerPoint, and Word replications LINK

• Report: Amazon in talks with AI startup Decart over possible $7B acquisition LINK

• Decade-old RAM is making a comeback LINK

🧰 Trending tools

ExcaliClaw: a collaborative workspace for hand-drawn diagrams with real-time editing, instant search, version history, and MCP integration so Claude, Cursor, or VS Code can draw for you. LINK

NewMinecraftServers: browse Java and Bedrock server listings with player stats and trends, while owners promote servers and track uptime, peak activity, and audience analytics. LINK

Kooler: helps remote teams build culture by sharing company values, sparking organic discussions, tracking engagement, and rewarding employees who embody them. LINK

Cube Forms: lets AI agents create and fill forms over MCP or HTTP without a browser, while people can review or edit the same link anytime. LINK

Strawberry: an AI companion that works inside your existing accounts, handling tasks like replying to leads and logging them to your CRM automatically. LINK

SalesRep: a virtual sales rep that pulls live pricing and stock from your store, answers across WhatsApp, Instagram, email and SMS, and follows up until deals close. LINK

📚 Trending papers & reports

Feature cleanup for recommenders gives companies a reliable way to spot and safely remove unused data signals from recommendation systems without hurting performance, validated on two large systems and flagging up to 59 removable features in one case. LINK

Graph generators secretly memorize their training data in a way standard quality checks miss entirely, meaning businesses using these tools to create synthetic networks risk leaking the original data they trained on. LINK

Confidence-aware training teaches a neural network to be both accurate and honest about how sure it is in a single training pass, removing the usual extra calibration step and its overhead. LINK

What-if predictions can be made directly from raw data without first mapping out cause-and-effect, cutting error by over 20x on a loan benchmark and handling bigger problems than it was trained on. LINK

Team-safety credit assignment pinpoints which cooperating robots or vehicles most cause a shared safety breach, boosting safe task completion up to ~5x in transport and lifting traffic throughput by ~50 to 68%. LINK

See you tomorrow for a new dose of ☕️ Techpresso!