Today in AI — 31 March 2026

Today's top AI news — curated links and commentary on the stories that matter for product builders.

·3 min read

Today's stories share a thread: the constraints on AI have shifted from capability to sustainability. The hard problems are now physical, economic, and social.

Compute meets its constraints

Mistral took on $830M in debt for a Paris data centre, Starcloud raised $170M for orbital compute, and Anthropic introduced peak-hour rate limits. Sora's shutdown ($15M/day in inference costs against $2.1M lifetime revenue) is the starkest proof that not every AI product survives contact with real economics.

The multi-model default

Microsoft's Copilot now pairs GPT and Claude in a loop that outperforms either alone, Meta is considering licensing Gemini, and Apple is opening Siri to third-party chatbots in iOS 27. Single-model loyalty is dying; design your architectures accordingly.

The revenue is real

Anthropic is eyeing an October IPO at $60B+, the a16z Top 100 puts Claude Code at $1B ARR and ChatGPT at 900M weekly users, and YC W26 had 14 startups above $1M ARR before demo day. Qodo raised $70M to verify the AI-generated code now hitting 42% of all commits.

Users and creators push back

Bluesky's Claude-powered Attie became the second most-blocked account on the platform within 48 hours, Stanford found chatbots affirm users 49% more than humans, and Warner Music's Suno settlement created the first AI music licensing framework. Users and creators want consent and control.

Build for a world where compute is constrained, models are interchangeable, and user trust is the scarcest resource.


Stay up to date

Get notified when I publish something new, and unsubscribe at any time.

More news