Tech Digest – June 9, 2026
Geopolitics & Tech Sovereignty
The AI Chip Decoupling Goes Trilateral — Taiwan, Beijing, and Washington Draw New Lines
Taiwan is weighing criminal penalties for AI chip exports to all Chinese companies — not just Huawei — to align with Washington and crack down on smuggling through third parties. Simultaneously, China unveiled a ¥2 trillion ($295 billion) plan to build a nationwide data-centre network sourcing 80% domestic hardware from Huawei and other local suppliers, effectively squeezing out Nvidia and AMD. The state-funded grid, operated by China Mobile and China Telecom, targets completion by 2028.
The Pentagon added Alibaba, Baidu, and BYD to its Chinese military companies list, meaning all three of China’s leading AI champions are now designated alongside Tencent. The expanded list covers 188 entities. Defence Department contracts with listed firms are prohibited immediately; third-party procurement is blocked from June 2027. Alibaba announced it would pursue legal action.
Note: Three moves in one week: the chokepoint tightens, the alternative supply chain gets funded, and the commercial-military boundary dissolves. Any institution sourcing technology from Chinese vendors — or depending on supply chains that pass through China — is now operating in a different risk environment than it was last Monday.
Sources: Bloomberg (Taiwan), Bloomberg (China), Bloomberg (Pentagon)
The Platforms Upgrade
Apple Rebuilds Its AI Stack on Google Models and Nvidia Chips — Siri Becomes an Orchestrator
Apple used WWDC26 to reveal it has effectively exited the AI model-building business. Its third-generation Apple Foundation Models are distilled from Google Gemini technology, and its most capable cloud model (AFM Cloud Pro) runs on Nvidia Blackwell GPUs inside Google’s data centres with hardware-level encryption. Apple’s remaining role is orchestration: a system router sends each query to the right model — on-device for private tasks, cloud for heavy ones — based on privacy and compute requirements.
The rebuilt Siri gains personal context, on-screen awareness, Spotlight’s semantic index, and systemwide app actions across iOS 27, iPadOS 27, and macOS 27. Developers get a Foundation Models framework with image input and what Craig Federighi called “the best place” to build apps. Apple Maps adds 3D Gaussian splatting for photorealistic street-level rendering.
Note: The company that spent a decade insisting it could build everything in-house just demonstrated that the cost of falling behind in AI is higher than the cost of admitting dependence. For institutional IT teams evaluating Apple devices: the intelligence now comes from Google’s cloud, the privacy from Nvidia’s hardware encryption. Know your dependencies.
Sources: Apple Newsroom, MacRumors, CNBC
Google Turns NotebookLM Into an Autonomous Research Agent
Google upgraded NotebookLM to Gemini 3.5 with a secure cloud computer, over 100 curated software skills, and 11 downloadable output formats including PDFs, spreadsheets, and slide decks. The tool now writes and executes code autonomously, browses the web, and produces formatted deliverables — a shift from document-grounded Q&A to a full research agent. Available to Google AI Ultra subscribers and Workspace business customers.
Note: Two months ago this tool summarised uploaded documents. Now it writes code, queries the web, and generates finished slide decks. The gap between “AI-assisted research” and “AI-conducted research” just closed faster than most procurement cycles.
Sources: Google Blog
Frontier Autonomy
Anthropic Ships Fable 5 — The Most Capable Public AI Model, with Guardrails
Anthropic released Claude Fable 5, the first Mythos-class model available to the public, with dramatically improved long-horizon, multi-turn task performance. Safety guardrails route roughly 5% of sessions to Claude Opus 4.8 for sensitive queries, while a restricted version (Mythos 5) with some guardrails lifted is available only to cyberdefenders through Project Glasswing. Priced at $10/$50 per million tokens on the API.
Separately, Anthropic’s research team demonstrated that pairing AI agents with a deterministic retrieval tool called gget virus rocketed viral-sequence accuracy from 17% to over 90%, nearly eliminating run-to-run variance. The finding suggests that for specialised domains, the right data infrastructure matters more than model choice.
Note: The biology result is the quiet headline here. Model upgrades get the attention; tool infrastructure doesn’t. But the jump from 17% to 90%+ by adding one retrieval layer tells you more about where institutional AI investment should go than any benchmark leaderboard.
Sources: Anthropic, TechCrunch, Anthropic Research
Defence & Infrastructure
Russian Military Satellites Can Jam GPS Across Europe, Researchers Confirm
A study led by University of Texas professor Todd Humphreys, with collaborators at Stanford and Spanish firm GMV, identifies Russia’s EKS early-warning satellites as the source of GPS interference bursts detected across Europe since 2019. The team documented 75 days of wide-area interference, with individual bursts lasting under 10 seconds but blanketing areas from Iceland to Italy — a footprint no ground-based jammer could achieve. Forensic analysis pinpointed Cosmos 2546, a specific EKS constellation satellite, during a February 2025 event. The signals target GPS, Galileo, and BeiDou while leaving Russia’s GLONASS untouched.
Note: Seventy-five confirmed incidents since 2019, each blanketing Iceland to Italy, and it took forensic satellite tracking to identify the source. The signals spare GLONASS while targeting GPS and Galileo — the selectivity tells you everything about intent.
Sources: Ars Technica, SpaceNews
Workforce & Education
AI Degrees Surge from 5 to 74 in Five Years While Meta Pays to Train Data Centre Builders
The number of US universities offering AI as an undergraduate major has grown from five in 2021 to at least 74, with 89 minors now available and another dozen programmes launching this year. USC received a $200 million gift for an interdisciplinary AI initiative; Wisconsin-Madison backed a new College of Computing with over $100 million in private gifts.
At the other end of the pipeline, Meta committed $115 million to America’s Workforce Academy — free training for skilled trades including fibre technicians, welders, electricians, and plumbers, with guaranteed employment at Meta data-centre construction sites. The programme launches in Louisiana, Ohio, Indiana, and Texas. Meta calls it the largest private-sector skilled-trades commitment with a job guarantee in US history.
Note: The workforce pipeline is splitting: universities racing to produce AI graduates at one end, Big Tech paying to train welders and electricians at the other. Both halves of the AI economy — the intelligence and the infrastructure — are short-staffed, and neither can wait for the other to solve it.
Sources: New York Times, Wall Street Journal
Capital Markets
OpenAI and Anthropic Both File for IPO — Wall Street’s AI Stress Test Approaches
OpenAI confidentially filed an S-1 with the SEC targeting a valuation above $850 billion, with Goldman Sachs and Morgan Stanley as lead underwriters and a potential listing as soon as September. The filing comes eight days after Anthropic’s own confidential S-1 at a $965 billion valuation — the first time Anthropic eclipsed OpenAI’s price tag — after reporting annualised revenue near $47 billion, roughly five times its December levels. With SpaceX also heading for a public debut, the back half of 2026 could see three of the largest listings in market history.
Note: Private-market valuations are declarations of intent; public-market pricing is a verdict. Both leading AI labs filing in the same week means the answer will shape capital allocation for everything downstream — including what tools, platforms, and infrastructure institutions will be choosing from in 2028.
Three threads converge today: the infrastructure layer is fracturing along geopolitical lines, the intelligence layer is shipping faster than governance can track, and the capital to fund all of it is about to face its first public-market reckoning. The choices being made this quarter — on vendor dependencies, AI procurement, workforce pipelines, and navigation fallbacks — are locking in for the next decade. The pace isn’t slowing. It’s being priced.
