😺 Your AI bill is creeping up. Here's why.

Content length and AI-generated slop

High-status tech people now post 3,000-word slop articles with zero shame; being concise takes real work and longer content no longer signals quality or effort.

AI compute demand as limitless as energy demand

Demand for AI intelligence may be the closest thing the economy has seen to limitless energy demand, making today's data center spending look like laying rails for the next industrial revolution rather than a bubble.

Cerebras IPO timing

Cerebras is perfectly timed for today's answer inference market, though the coming agentic inference market will look different and favor slower, cheaper compute.

Orbital Data Centers as AI Compute Infrastructure

Cowboy Space raised $275M to build data centers in orbit, betting that the bottleneck for agentic AI inference is electricity cost, not engineering — and that orbital compute solves this.

Positions space-based infrastructure as a viable solution to terrestrial energy constraints for AI compute, with rocket capacity as the only real bottleneck.

Agentic Inference Runs on Cheaper, Slower, or Orbital Compute

Ben Thompson argues that agentic inference (AI doing overnight work with no human watching) doesn't need fast GPUs — it can run on slower, cheaper, or even orbital compute wherever electricity is cheapest, making today's GPU-centric infrastructure potentially obsolete for future AI workloads.

Challenges the assumption that all AI inference requires expensive, low-latency GPU clusters like NVIDIA or Cerebras chips.

OpenCode replacing Claude Code

OpenCode is the free open-source Claude Code alternative that everyone's switching to, with 150K+ GitHub stars and 6.5M monthly developers, running with 75+ model providers at zero API cost.

Engineers switching from Claude Code to OpenCode

Mira Murati (Ex-OpenAI) building in stealth

Mira Murati's Thinking Machines unveiled a new real-time way to interact with AI.

Former CTO at OpenAI now exploring Real-time AI interaction