This week, AI left the ground, crossed a border, and plugged into the power grid. Google put a data center in orbit and asked it to write a poem. Apple finally found its way into China for AI, by partnering with Alibaba. And OpenAI promised that its biggest data centers can help rescue the electricity grid instead of only straining it. Here is what happened, with every claim verified against primary or reputable sources.
Key Takeaways
Google's Project Suncatcher prototype satellite is online, carrying 4 TPUs and generating up to 1 kW, but it can run only one TPU at a time for about 20 minutes, and its first prompt produced a haiku about space. (TechCrunch)
Apple will integrate Alibaba's Qwen models into Apple Intelligence in China, with testing from November 4 in the latest iOS developer beta and a broader rollout in December with iOS 27.3, and Alibaba says requests and responses will not be used to train its models. (MCI Group)
OpenAI launched its Grid Protectors Program on November 2, committing 1,000 megawatts, or 31 percent of peak load, at its planned 3,200-megawatt Project Camellia campus in Georgia for grid emergencies, after a UK trial cut data center power by up to 40 percent. (Tech Times)
1. Google's Data Center Goes to Space, and Its First Job Is Poetry
Google's first data center satellite is online and communicating with the ground after launching earlier this month alongside more than 100 other spacecraft on a SpaceX Falcon 9 rocket, reports TechCrunch. The satellite, part of Google's Project Suncatcher, carries four Google Tensor Processing Units (TPUs) and generates up to 1 kW of electricity. It is intended to pave the way for future networks of close-flying satellites performing large inference workloads in orbit.
The limits are as interesting as the launch. The power demands of the TPU mean the satellite can only turn one on for about 20 minutes at a time, which complicates the already tricky business of space-to-ground communications. After a commissioning process that tested the satellite's subsystems over about three weeks, engineers sent their first prompt: they asked the model pre-loaded onto the satellite, Google DeepMind's Gemma 4, to write a haiku about being in space. It replied: "Blue marble below / Data streams in silent grace / Starlight feeds my core."
The poetry is a demo, but the real test is radiation. The ongoing test campaign will examine how the chips perform under thermal stress and in the space radiation environment. Radiation can cause "bit flips", changing a stored value in computer memory and corrupting data without triggering typical monitoring tools. Testing on the ground suggested the impact would be minimal for inference workloads and more of a problem for training a model. Now the Google team will get to assess those dynamics in reality, not a simulation.
If the TPUs onboard prove capable, the next step will be launching two more satellites equipped with the advanced chips in 2027. Those spacecraft will support higher power generation and demonstrate inter-satellite optical links. Suncatcher is a classic Google moonshot: the focus is on designing infrastructure that will be economical in the 2030s. Getting there may depend on rockets too, since both Google and SpaceX will require Starship, the still-in-development heavy-lift rocket, to be fully and rapidly reusable before orbital data centers can scale.
2. Apple Finally Brings AI to China, With Alibaba's Qwen
Apple will integrate Alibaba Group's Qwen artificial intelligence models into Apple Intelligence in China, bringing the United States technology giant's long-awaited AI services to Chinese iPhone users, reports MCI Group. Qwen-powered features will be available for testing from November 4 in the latest iOS developer beta and will be rolled out more broadly in December with the release of iOS 27.3, Apple said on October 31.
The partnership will allow Chinese users to access Qwen through Siri, Apple's writing tools, and its visual intelligence feature, enabling text and image understanding, content generation, and analysis of documents and photos. For some requests, Siri will be able to call on Qwen's broader knowledge to answer directly. Users will also be able to ask questions about selected photos and documents, including PDF files and presentations. Qwen will also be integrated into Apple's writing tools through a "Compose with Qwen" function that generates original content from a prompt, including text and images, and into visual intelligence, where users can select "Ask Qwen" to identify or obtain more information about objects and places.
Privacy is part of the pitch. Apple said users will be asked for permission before a question, document, or image is sent to Qwen. Users can choose whether to enable the Qwen integration and can sign in through Apple or an existing Qwen account. Alibaba said requests and responses generated through the integration will not be used to train its models. Apple also said users will remain in control of when Qwen is invoked and what information is shared.
The tie-up marks a significant step in Apple's efforts to bring generative AI services to China, more than two years after Apple Intelligence was first unveiled. China's cyberspace regulator registered Apple's on-device generative AI service for use on iPhones in July, clearing a major regulatory hurdle. The December update will not include the more advanced Siri AI capabilities or other next-generation Apple Intelligence features unveiled at Apple's Worldwide Developers Conference this year. The first batch of features will be available on supported iPhones, including the iPhone 15 Pro and Pro Max, iPhone 16 and later models, iPhone Air and iPhone Duo, with device and Siri languages set to English or Simplified Chinese.
3. OpenAI's Data Centers Offer to Rescue the Grid
OpenAI on November 2 launched the Grid Protectors Program, a national initiative to embed electricity curtailment commitments into every compute campus the company directly develops, reports Tech Times. The program's pilot is OpenAI's Project Camellia, a planned 3,200-megawatt campus in Effingham County, Georgia, being developed under a 25-year power agreement with Georgia Power. Under the terms, OpenAI will make 1,000 megawatts, 31 percent of the campus's peak load, available for curtailment when the regional grid faces extreme weather or surging demand. That is equivalent to roughly 700,000 average US households.
The idea passed a real test before it became a promise. In December 2025, National Grid ran a controlled trial with Emerald AI, energy research consortium EPRI, cloud provider Nebius, and NVIDIA at a London-area GPU cluster running NVIDIA Blackwell Ultra processors. Over five days, the trial simulated more than 200 grid emergency scenarios. The facility cut power consumption by up to 40 percent across the full battery of simulated events, with a 100 percent compliance rate. In the fastest emergency scenario, simulating the sudden loss of a large generating plant, the cluster shed 30 percent of its load in under 40 seconds. In a sustained stress test, it held a 10 percent reduction for 10 consecutive hours without disrupting a single critical workload.
The trick is separating work that can wait from work that cannot. Emerald AI's platform separates latency-critical inference tasks, queries that users are actively waiting for, from deferrable workloads such as batch training runs, prefetch operations, and model evaluation jobs. When a grid signal arrives, deferrable tasks are paused or throttled, GPU clock states are reduced, and instantaneous power draw drops. When the event clears, deferred work resumes automatically.
OpenAI itself is careful about the limits. The company stated explicitly that campus flexibility cannot replace investment in new energy supply and transmission. The program is designed as a complement to grid expansion, not a substitute for it. Critics of voluntary approaches also argue that the utility of demand response commitments depends entirely on their enforceability, and bilateral private agreements are not subject to the same regulatory scrutiny as supervised wholesale market programs. Still, the technical question now has a credible answer: AI workloads can flex, by up to 40 percent, in under a minute. The policy question, whether that flexibility will be required, measured, and compensated under binding rules, remains open.
How the Week's Stories Stack Up
Aspect | Story | Key Number | Source |
|---|---|---|---|
Space infrastructure | Google's Suncatcher satellite online | 4 TPUs, up to 1 kW, one TPU for about 20 minutes | TechCrunch |
AI in China | Apple integrates Alibaba's Qwen | Testing from November 4, rollout in December with iOS 27.3 | MCI Group |
Energy | OpenAI Grid Protectors Program | 1,000 megawatts, 31 percent of peak load, 3,200-megawatt campus | Tech Times |
The thread across all three: AI is no longer just software running somewhere else. It is hardware in orbit, a partnership that had to cross a regulatory border, and a power consumer large enough to act as a grid resource. Last week was about rules, money, and enforcement. This week is about where AI physically lives, who it partners with to reach people, and what it gives back to the grid that powers it.
If you like quick video takes on how new open AI projects work, browse the repo signals for 30-second overviews. For last week's roundup, see This Week in AI: November 3, 2026.
Frequently Asked Questions
What is Google's Project Suncatcher?
Project Suncatcher is Google's prototype space data center program. Its first satellite, launched on a SpaceX Falcon 9 alongside more than 100 other spacecraft, carries 4 TPUs, generates up to 1 kW, and can run only one TPU at a time for about 20 minutes. Google plans two more satellites in 2027 with higher power generation and inter-satellite optical links, to test whether inference workloads can run economically in orbit in the 2030s.
When will Apple Intelligence with Qwen arrive in China?
Qwen-powered features will be available for testing from November 4, 2026 in the latest iOS developer beta, with a broader rollout in December with iOS 27.3. Qwen will be accessible through Siri, Writing Tools, and Visual Intelligence on supported iPhones, including iPhone 15 Pro and Pro Max, iPhone 16 and later models, iPhone Air, and iPhone Duo, with device and Siri languages set to English or Simplified Chinese. Alibaba says requests and responses will not be used to train its models.
What is OpenAI's Grid Protectors Program?
Launched November 2, 2026, the Grid Protectors Program embeds electricity curtailment commitments into every compute campus OpenAI directly develops. Its pilot, the planned 3,200-megawatt Project Camellia campus in Georgia, will make 1,000 megawatts, 31 percent of peak load, available for curtailment during grid emergencies under a 25-year agreement with Georgia Power. A December 2025 UK trial showed AI workloads can cut power by up to 40 percent, shedding 30 percent of load in under 40 seconds in the fastest scenario.
Conclusion
A poem from orbit will not pay for a data center in space, but it proves the hardware works where it matters. A partnership in China will not settle the global AI race, but it finally brings Apple Intelligence to hundreds of millions of iPhone users who waited more than two years. And a curtailment promise will not build a new power plant, but it shows AI data centers can flex fast enough to help the grid in an emergency. The technology keeps advancing. This week, the interesting part was where it lives, who it works with, and what it gives back.
Read this article in Bahasa Indonesia: Berita AI: 04 November 2026.