Grid engineers are the new scarce talent. • HVDC + transmission • grid planning + interconnection • protection + substations • nuclear project delivery Retirements accelerating. Graduates declining. By 2030 we need ~2× the workforce—engineers who can bring 10–30 GW online

✨

Grid engineers are the new scarce talent. • HVDC + transmission • grid planning + interconnection • protection + substations • nuclear project delivery Retirements accelerating. Graduates declining. By 2030 we need ~2× the workforce—engineers who can bring 10–30 GW online

🚨BREAKING: Anthropic just ordered $21 BILLION in TPU racks from Broadcom $AVGO CEO: “But. That does not mean our other customers are using TPUs. Infact they prefer to control their own destiny by continuing to drive their multi year journey to create their own custom AI

It's funny to me that GPT 5.2 is available on Perplexity but not on the ChatGPT App. 😭

🚨 Why is nobody talking about $AVGO? $1.9T market cap. ~$408. Up ~73% YTD. Earnings TODAY. Broadcom is quietly becoming the backbone of the entire AI buildout. 1/ What Broadcom actually is: SEMICONDUCTOR SOLUTIONS (~60% rev) - Data center / AI / cloud networking chips -

7/ What’s Next for @Starcloud_Inc_ 🛰️☀️ Starcloud-1 was the lab demo. The roadmap looks like this: 2026 – Starcloud-2: first commercial satellite, with a small GPU cluster and persistent storage in sun-synchronous orbit, processing raw space data in orbit and sending down only

6/ Dyson-Sphere Angle Instead of building more cooling towers around our star, we start putting compute nodes into orbits bathed in its light, turning sunlight directly into FLOPs. SpaceX, Google and Blue Origin’s reported work on orbital AI data centers show that Starcloud

5/ The Earth-Side Problem: Energy and Land On Earth, AI is starting to hit hard physical limits: Global data-center electricity demand is expected to more than double by 2030, driven largely by AI. Data centers fight communities over water use, grid strain, and land for new

4/ What This Experiment Actually Proves This mission doesn’t solve every problem—but it does demonstrate that: - A modern data-center GPU can survive launch and operate in orbit. - It can run non-trivial LLM training and inference end-to-end. - The pattern “ship job → train in

3/ Why People Were Skeptical When Starcloud first talked about H100s in orbit and a long-term vision of multi-gigawatt space data centers, a lot of people put it in the “ambitious concept art” bucket and asked: - Can GPUs survive radiation and stay reliable? - How do you dump

2/ How Inference Proved It Worked 1. Inference on the trained nanoGPT: - Sample from the model in orbit to generate Shakespeare-style text. 2. Inference on Gemma/Gemini: - Run larger, pre-trained models on the same H100 for chat-like responses tied to the satellite’s position

1/ What Starcloud-1 Actually Did Hardware: a single NVIDIA H100 mounted in a small satellite (roughly mini-fridge size) in low Earth orbit. Models: - nanoGPT – compact GPT-style model (tens–hundreds of millions of parameters class) - Gemma (and a version of Gemini) – larger,

Starcloud-1 and the First LLM in Space A small satellite, Starcloud-1, just became the first spacecraft to train and run an LLM entirely in orbit on an NVIDIA H100 It trained @karpathy’s nanoGPT on the complete works of Shakespeare, then ran inference on that model from space

🚨 BREAKING: Elon musk confirms SpaceX IPO incoming SpaceX will build DATA CENTERS IN SPACE

🚨 BREAKING: $ORCL crashes -11% after earnings On the call, Oracle management signaled $15B increase in CapEx to scale cloud and AI infrastructure. That means lower near-term free cash flow and higher leverage, in exchange for greater capacity to support growing cloud and AI

the web should work for you

What the H200 Deal Really Means Export controls rested on a simple premise: starve China of advanced compute and you slow its AI. That only works if China stays dependent on Nvidia; with domestic chips and software maturing, it doesn’t. Letting Nvidia sell freely would undercut

The CUDA Clock From Nvidia's vantage, this is existential. If China decouples from CUDA, the company loses leverage over a massive slice of global compute. No export control brings that back. Adding urgency: early CUDA patents, filed in the late 2000s, begin expiring around