Issue #8 · Monday, 2 November 2026

OpenAI cuts its funding target to $20B after raising $11B

Money and borders dominated the weekend. The biggest AI lab is finding that investors are less eager than the headline valuations suggested, and China is writing rules for exporting models. Elsewhere, the usual mix of small product moves and one satellite doing something charming.

The big one

OpenAI lowers its funding target from $30B to $20B

The Financial Times reported Friday that OpenAI cut its round target to $20 billion, with a first close expected to come in below that. The WSJ says $11 billion is secured so far, less than half the earlier $30 billion goal, and that some large investors were tepid. UAE investors will lead, and a later retail offering to wealthy individuals could lift the total to $20 billion.

Why it matters: This round comes ahead of a planned IPO, so a smaller raise and lukewarm investors suggest less appetite for the AI spending story, just as Wall Street grows wary of infrastructure costs.

Sources:wsj.com,HN discussion,TechCrunch

Read the briefing →

Policy

China proposes export controls on frontier AI models, agents and chip design

Announcement No. 48 proposes 40 catalogue additions, 23 of them in AI. Model weights and related know-how are restricted if a model hits any one test: 10^26 FLOPs, 2 trillion parameters, or a global top-10 ranking or call volume. The additions are a draft open for comment until November 15, but pre-export registration applies now. API, app, SaaS and on-device use don't count as exporting; local deployment outside China does.

Why it matters: If you build on Chinese open-weight models, local deployment abroad could now require registration. Both major AI jurisdictions now have formal outbound controls, and China's threshold copies the 2023 U.S. figure.

Sources:labs.cloudsecurityalliance.org,HN discussion

Read the briefing →

Products

Google's first data center satellite is online and running Gemma 4

Google's first data center satellite is online. It carries a TPU running Gemma 4, the Google DeepMind model, which answered a prompt for a space haiku. The satellite is part of Project Suncatcher, Google's experiment toward space-based AI data centers.

Why it matters: It is an early proof that AI compute can run in orbit. For now it is mostly a demo, and a poetry prompt is not a workload.

Sources:TechCrunch,The Verge

Apple Intelligence reaches China with iOS 27.3

Apple will launch Apple Intelligence in China with iOS 27.3, with a beta on November 4 and a full release in December. It includes earlier features such as notification summaries, translation, writing tools, visual intelligence and photo object removal. The new Siri AI is not part of the package.

Why it matters: It opens Apple's AI features to the Chinese market, but in a trimmed form that leaves out the headline Siri upgrade.

Sources:The Verge

Worth hearing

Altman: OpenAI's math results will keep coming, and RAM belongs in data centers

On the Cheeky Pint podcast, Sam Altman said he sympathizes with mathematicians overwhelmed by OpenAI's stream of AI-generated results but won't stop, and hinted at more discoveries. He also said that during the RAM shortage it is better for memory to sit in data centers than in personal PCs, and dismissed the idea that local open-source models remove the need for data center capacity.

Why it matters: OpenAI is pressing ahead on AI-driven math despite friction with researchers, and its CEO is arguing for data center priority in a tight memory market.

Sources:The Verge,The Verge