Riot just leased its Texas campus for $28B
The biggest story today is about capacity, not models. Riot Platforms is leasing its entire Texas campus for 15 years, and the tenant pays rent on all 910 megawatts whether it uses them or not. If you're building an AI company, the lesson is simple: power and chips are being locked up years ahead, so plan your compute before you need it.
- Riot's $28B Texas power lease
- Nvidia's China chip orders outpace supply
- Render gives each agent its own computer
- Quick hits: agents, models, and roleplay
- Today's deals and the Lounge AI Index
Top stories
All news →What happened: Riot Platforms signed a 15-year lease on its whole Corsicana, Texas campus with a frontier AI lab, which a 10-Q filing reportedly names as Anthropic.
- The deal is worth about $28B over the initial term, with average rent near $1.8B a year, and roughly $54B if both five-year extensions are exercised.
- Rent is payable on all 910 MW whether or not it is used. Delivery comes in three phases: 400 MW by September 11, 2027, 200 MW by January 28, 2028, and 310 MW by July 30, 2028.
- Riot sold 10,599 BTC in the third quarter to help fund the buildout, cutting its holdings from 11,380 BTC to 2,418 BTC.
Why it matters: The biggest buyers are locking up land and power for years, and that shapes what compute costs everyone else. If you rent compute, expect longer commitments, and budget for them now.
What happened: Chinese AI and cloud firms have reportedly ordered about 6 million Nvidia RTX Pro 5500 workstation cards, and Nvidia can fill only a small fraction of requests.
- Bloomberg puts the order at about 6 million cards, worth roughly $70B at more than $11,000 each.
- Buyers expect only some tens of thousands of units within four months, about 1% of the total. Nvidia reportedly told partners it could fill only 5% to 10% of requests.
- Reuters reported in October that ByteDance, Alibaba and Tencent alone ordered more than 3 million cards at about $11,400 each. Nvidia has not confirmed the totals.
Why it matters: Chip supply is the bottleneck, and it sets the price of compute for everyone, including you. If your roadmap depends on GPUs, treat them as scarce and secure commitments early.
What happened: Render Sandboxes gives each AI agent its own isolated Linux computer that starts in under a second.
- Agents can run tools like Claude Code or Codex, and can run your app with Docker Compose, a tool for running multi-part apps, to test their own changes.
- You choose which domains each sandbox can reach. When an agent goes idle, the sandbox suspends and compute charges stop, and it wakes with files and memory intact.
- You can run thousands at once on the same platform as the rest of your app. The launch drew 236 upvotes on Product Hunt.
Why it matters: Agents need a safe place to work, and billing only while they are active keeps that cost down. If you sell agents, check whether your pricing tracks active time or idle time.
Quick hits
- Kimi posted downloadable files for its K3.1 Pro model on Hugging Face, and our full brief is on the way. huggingface.co →
- Hacker News users are sharing which AI subscriptions and agents they use day to day. news.ycombinator.com →
- Wanaka Studio lets an AI agent build a 3D game with no code, and says it uses about 80% fewer tokens. Read →
- Prelulu offers chat and video roleplay with AI characters, using real-time avatar technology. Product Hunt →
- OpenMail gives each agent its own email inbox on your domain, free for three inboxes. Product Hunt →


