OpenAI closes $40B at $340B valuation — the largest private round in history.

SoftBank leads with $30B. The round exceeds Stripe's record by 4×. OpenAI is now valued higher than Intel, IBM, and AMD combined. The AI hardware bottleneck is now the most valuable problem in tech.

Harvard study finds AI intensifies work — it doesn't reduce it.

An 8-month enterprise case study found employees using AI tools worked faster but logged more hours. "More output per person" becomes a retention problem when it means "more work with no relief."

Anthropic and OpenAI release competing models on the same day.

Both labs ran dueling Super Bowl campaigns simultaneously. These labs are fighting on three fronts — model capability, developer tools, and consumer brand.

Circular Financing

When Nvidia invests in OpenAI, and OpenAI spends on cloud compute that buys Nvidia chips, money doesn't flow — it loops. This loop explains why AI valuations feel disconnected from revenue.

AI value is being captured by platforms, not users.

OpenAI's ad launch and the Harvard work study point the same direction. Expect this debate to intensify through 2026 as labor markets respond.

Chinese AI labs are closing the gap across modalities.

ByteDance's Seedance 2.0 and Kuaishou's Kling 3.0 arriving days apart mirror last week's Kimi K2.5. The pressure is no longer quality — it's release velocity.

Build a Custom Skill in Claude Code

Simon Willison demonstrated how to create reusable skills for Claude Code — saved prompt templates that turn repetitive tasks into one-command workflows. Instead of re-explaining your constraints every session, a skill file encodes them once.

Create a SKILL.md in your project root. Define three sections: the task it handles, the rules it follows, and the output format.

Then run claude "review src/auth.py" and the skill loads automatically. Willison reports cutting his review workflow from 15 minutes of prompt