Top stories today:

1. OpenAI said to boost compute margin to ~70% from ~52%
2. New York signs AI safety law despite Trump blocking order
3. Cerebras said to file for U.S. IPO in Q2 2026
4. Elon’s $56B Tesla pay package must be restored, Delaware court rules
5. Sam Altman said to spin out brain-computer interface startup
6. Anthropic’s Claude Opus 4.5 hits 50% task time horizon in under 5 hrs

0. Data and calendar

Listen to our AI-generated podcast summarizing today’s newsletter (beware of hallucinations):

1. OpenAI said to boost compute margin to ~70% from ~52% at end-2024

• OpenAI’s post-compute revenue share has reportedly risen in recent months.

• The ~70% compute margin remains well below levels seen at publicly traded software companies.

• Anthropic posted a ~-90% compute margin in 2024.

• However, Anthropic could overtake OpenAI in overall server efficiency over time, as it has fewer nonpaying users.

• OpenAI’s margin gains were driven by an internal “code red” effort and lower compute rental costs.

Anthropic’s revenue has grown 8×, while OpenAI’s annualized revenue has ~3x’d since the start of 2025

• Relatedly, SoftBank is said to race to meet its $22.5B commitment to OpenAI by year-end.

2. New York signs AI safety law despite Trump order blocking state regulation

• Fri.: Gov. Kathy Hochul signed the sweeping AI safety bill.

• Jan. 1, 2027: AI firms with $500M+ in revenue must publish and follow protocols to prevent critical harm and report serious breaches.

• Companies face fines for violations.

• A new office within New York’s financial regulator will oversee enforcement, rulemaking, fees, and annual AI safety reports.

• Developers must disclose AI safety incidents within 72 hours, stricter than California’s 15-day rule.

4. Cerebras said to file for U.S. IPO in Q2 2026