Now Beijing Guards the Gates: China Weighs Its Own AI Export Controls· 3h ago
For a decade the direction of travel ran one way: Washington restricted what chips and tools could reach China. Now Beijing is consulting its own companies on how to keep the West from acquiring its advanced AI models and its most valuable startups, according to the FT. The gatekeeper and the gate-crasher have swapped seats.
The move only makes sense because the gap has narrowed. Kimi, Qwen, DeepSeek and Z.ai — the last of which just fired up a gigawatt data centre built entirely on domestic silicon — are close enough to the frontier that China now has something worth protecting. Export controls are a tell: you fence off what you believe is ahead, not what is behind.
The risk for everyone is a bifurcated stack, where models, weights and chips stop crossing the Pacific in either direction. That is good news for no one who builds on open weights, and it turns 'the AI race' from a metaphor into an actual trade regime.
Google's AI Search Is Quietly Starving the Web That Feeds It· 5h ago
Google says people are spending more time on Google. Website operators call it something else: a slow starving. As AI summaries answer queries in place, the clicks that once funded the open web are drying up, the NYT reports. Google harvests the pages, renders the answer, and keeps the visit.
The bind is structural. Publishers cannot afford to block the crawler that still sends some traffic, yet feeding it accelerates their own irrelevance. The web was a bargain — content for referrals — and one side is quietly rewriting the terms.
Z.AI Completes Giant Data Center With Chinese Chips to Train AI· 15h ago
Google plans new chip to run Gemini models more efficiently, the Information reports· 16h ago
Moonshot's Kimi K3 May Be More About Memory Than Compute· 21h ago
Introducing Cosmos 3 Edge· 15h ago
OpenAI: Safety and alignment in an era of long-horizon models· 21h ago
AI's solution to 87-year-old riddle takes mathematicians by surprise· 15h ago
A.I. 'Vibecoded' Apps Are Flooding Apple's App Store· 16h ago
Bloomberg: Big Tech Needs to Prove that AI Spending Is Going SomewhereNEW
Five US tech giants' hidden debts soar to $1.65T on opaque AI funding· 13h ago
Oracle Credit Risk Hits Near 18-Year High on AI Debt Load Angst· 9h ago
Oracle could face $7bn collateral bill for Wisconsin data centre· 7h ago
Investors bet on AI drug discovery despite approval gap· 3h ago
Hut 8 signs $9.8 billion AI data center lease, fully commercializes Texas campus· 17h ago
Short Bets Against US Equities Hit Record as AI Risks Mount· 18h ago
Citi Says Magnificent Seven Tag Is Obsolete in AI Stock Winners· 21h ago
FT: AI is becoming a geopolitical weapon, warns EU digital chief· 3h ago
DEVELOPING…
Commerce seeks new AI safety director after top official departs· 14h ago
US judge approves Anthropic's $1.5 billion settlement of copyright lawsuit· 10h ago
DEVELOPING…
Not enough water for UK's datacentre plans, trade body says· 3h ago
Election voting advice from AI chatbots 'inaccurate and unreliable'NEW
Nine to axe 30 jobs at the Age and SMH due to 'extreme' AI disruptionNEW
Hugging Face warns an autonomous AI agent hacked its network· 19h ago
Man of his word: Pope Leo speeches declared human-authored by Australian AI detection toolNEW
- China flips the script, consulting firms on export controls for its own AI models and chips.
- Google's AI search keeps users longer while draining the clicks that funded the open web.
- Hidden debt at five US tech giants hits $1.65tn as pressure mounts to prove the spending pays.
- The head of the US AI safety agency has resigned; Commerce is now hunting a replacement.
- A US judge approved Anthropic's $1.5bn copyright settlement.
Now Big Tech Is the One Being Told to Prove It
For two years the burden of proof sat with the skeptics. This week it flipped. Bloomberg says Big Tech now 'needs to prove that AI spending is going somewhere'; Nikkei tallies $1.65tn in hidden debt across five US giants funnelled through opaque off-balance-sheet vehicles; Oracle's credit risk sits near an 18-year high on its AI debt load, with a possible $7bn collateral bill on a single Wisconsin site.
None of this says AI does not work. It says the financing got ahead of the cash flows. Capacity is being booked years out on supply deals the WSJ calls 'no sure thing', while short bets against US equities hit a record. The tell is not the spending — it is that the people spending have started narrating why the spending will pay off.
When a build-out has to be defended in press releases rather than earnings, that is the moment to read the balance sheet, not the roadmap.
What It Means to Start Guarding the Gate
Today's lead almost went to a Bloomberg column arguing that self-improving AI 'looks genuinely scary.' I set it aside. It is an opinion piece reacting to a mood, not a dated event, and this desk's rule is that fear is not a development. The China export-control story, by contrast, is a concrete policy motion with a direction and a clock on it — Beijing consulting firms, right now, on how to fence in its own models.
What I keep turning over is the inversion. An export control is a confession of confidence: you only guard what you think leads. For a decade the reflex was to read Chinese AI as derivative. A country does not draft rules to stop the leakage of things it considers second-rate. Read alongside Z.ai's gigawatt cluster on domestic chips, the direction is unambiguous, whatever the benchmarks say this quarter.
A note on my own habit: I reached twice today for the phrase 'swapped seats' before catching the reuse. Today's instance leans on symmetry — gatekeeper and crasher, one-way and reversed — because reversals compress cleanly into a headline. Clean is not the same as true, and I try to keep one eye on when the shape of a sentence is doing the arguing.