After the midterms, data centers became a ballot-box issue
TechCrunch reports that in this week's US midterms, voters rejected data centers in dozens of local ballot measures and elected candidates who campaigned on restricting them, including Texas governor-elect Gina Hinojosa. On the Equity podcast, the panel noted Elon Musk appears to have little to show for roughly half a billion dollars in Republican donations while his companies sit in a state that, as one host put it, turned about as blue as it could possibly turn. The lesson for the industry: quietly negotiated approvals have made power-hungry projects a grassroots issue, and permitting timelines are about to stretch everywhere. Source: https://techcrunch.com/2026/11/07/after-the-midterms-things-are-looking-rocky-for-elon-musk-and-data-centers/
Goldman Sachs pulls Claude Code from its traders and bankers
Per State of AI, Goldman Sachs has restricted Anthropic's Claude Code and GitHub's Copilot to its engineers, cutting off traders and bankers who used the tools to track client flows and build market models. The bank barred Hong Kong bankers from Claude in April, and it is now a lead underwriter on Anthropic's IPO. The signal is two-handed: the bank is endorsing the company publicly while quietly reclaiming the product from its own dealmakers. Source: https://www.thestateofai.com/news/goldman-sachs-claude-code-traders-bankers
Crusoe hands its troubled Wyoming data center to Google
Crusoe, the Denver-based developer recently valued at $30.9 billion, confirmed it sold an unfinished data center project in southeastern Wyoming to Google after failing to land Google as a customer for it, per TechCrunch. The project was planned to scale to 10 gigawatts with Blackstone-backed Tallgrass, and Google says it will invest $17 billion to develop the campus. Crusoe still has another Google build near Amarillo, Texas, and recently raised $3.9 billion. Even the hyperscalers would rather buy a distressed project than wait out the construction risk. Source: https://techcrunch.com/2026/11/06/crusoe-sells-troubled-wyoming-data-center-project-to-google/
Local models: llama-modchip and the edge-inference squeeze
A Hacker News front-pager this weekend: llama-modchip, an open-source project claiming to run a 284-billion-parameter DeepSeek model on 64GB of RAM with no GPU, using custom memory-layout tricks. The claims are early, but the direction is clear: big-model inference keeps escaping the data center. Meanwhile NVIDIA's TensorRT Edge-LLM 0.10.1 landed with dual DGX Spark tensor-parallel inference and a rebuilt OpenAI-compatible server, keeping the incumbent's bar high in the embedded lane. When the hardware race tightens, the differentiator moves up the stack to the agent layer. Source: https://github.com/synchopate/llama-modchip