1Google backstops a $19B lease for Anthropic's Kentucky data center
TeraWulf, a data-center developer, announced Tuesday that Google will provide credit support for the project financing behind a roughly $19 billion, 20-year lease it signed with Anthropic for 401 megawatts of computing capacity at TeraWulf's campus in Hawesville, Kentucky. TeraWulf also said it has signed a definitive agreement with engineering firm Fluor to build the facility.
In plain English: AI companies like Anthropic need enormous buildings full of computers, but they don't want to build them. Specialists like TeraWulf do - if they can borrow the money. Google's "credit backstop" means Google effectively vouches for the loan, making it cheaper and easier to finance. Anthropic rents the finished capacity.
Why it matters: the AI race is now a construction and finance race, and Big Tech balance sheets are being pledged directly against it. Google guaranteeing an Anthropic data center is also a sign of how intertwined the rivals have become: Google is both a major Anthropic investor and the owner of its competitor, Gemini.
2ChatGPT can now make animated GIFs
OpenAI's new Images 2.6 model lets ChatGPT create animated GIFs from a text prompt or an uploaded photo, The Verge reported. Users can adjust the animation speed, edit individual frames, and share the result - something ChatGPT previously couldn't do in a single step. The feature is rolling out now.
In plain English: a GIF is a short, looping animation - the internet's favorite reaction format. Until now, making one meant stitching frames together in separate tools. Now you can just describe it.
Why it matters: this is a small feature with a big audience. GIFs are how people talk in group chats, and putting one-step animation in front of hundreds of millions of ChatGPT users is the kind of everyday capability that keeps a chatbot a daily habit rather than an occasional tool.
3Moonshot AI releases Kimi K3.1 Pro's weights for anyone to download
Moonshot AI has published the weights of Kimi K3.1 Pro, its most capable model, on Hugging Face. The model card describes it as an open-weight, natively multimodal "agentic" model with 2.8 trillion total parameters, of which 104 billion are active on any given task - meaning it can work with images as well as text, and is built to carry out multi-step work.
In plain English: yesterday's briefing covered Tencent's Hy4, which used Kimi K3.1 Pro as the benchmark to beat. Now that benchmark is itself fully open - anyone with enough hardware can download and run it, inspect it, or build on it. "Parameters" are the model's learned settings; only a slice of them are used per answer, which keeps running costs down.
Why it matters: the strongest Chinese labs - DeepSeek, Alibaba, Moonshot, now Tencent - keep releasing their best work openly, while the top American models stay closed. For startups, researchers, and governments wary of renting AI from US providers, the open tier keeps getting more capable.
Checked against original announcements and named press reports on the morning of November 5, 2026. TeraWulf's figures come from its press release; Moonshot's parameter counts and model description come from its own model card.
