Nathan Lambert hosts the AI Daily Brief, covering major developments in artificial intelligence policy, business, and technology. The show is sponsored by KPMG, Granola, Robots and Pencils, and Zencoder, with ad-free versions available through Patreon.

This episode focuses on Anthropic's controversial pricing changes that have sparked developer backlash, alongside coverage of the US AI envoy's Beijing visit with Jensen Huang, Cerebris's massive IPO, and polling data showing strong American opposition to data center construction.

The discussion explores how the shift from 'assisted AI to agentic AI' is forcing companies to move from seat-based to token-based pricing models, marking what Lambert calls the end of the 'AI subsidy era' due to compute constraints and exploding enterprise demand.

Anthropic's Pricing Overhaul Sparks Developer Revolt

Starting June 15th, Anthropic will separate Claude usage into 'interactive' (human-in-the-loop) and 'programmatic' (API-based) categories, with separate billing for each type of usage.

Power users have been consuming 10x to 20x more tokens than their subscription value suggests - one session used '$31 worth of API tokens, counting as 7% usage over five hours' for a $100/month plan.

Developer Theo from T3 wrote: 'I can't help but feel personally burned by the cloud code changes... Now our users are getting their rate limits cut by 40x despite us doing everything right.'

The change effectively ends token subsidies for third-party tools like T3 Code, Conductor, and Claude-P while maintaining them for Anthropic's own interfaces.

American Opposition to Data Centers Exceeds Nuclear Power

A new Gallup poll found 70% of Americans oppose data center construction in their local area, with 48% strongly opposed - higher than the all-time nuclear power opposition peak of 63%.

Environmental concerns dominate opposition: 46% worry 'a great deal' about environmental impact, with 50% citing excessive resource use including water and electricity consumption.

Only 12% cited job displacement fears and 14% mentioned generic AI concerns, suggesting opposition stems from infrastructure rather than AI-specific anxieties.

Mads Campbell argues data centers face 'fundamentally a marketing problem' and suggests 'no one would care about data centers in their neighborhood if it came with free electricity and Wi-Fi.'

Cerebris IPO Success and OpenAI Policy Shift

Cerebris raised over $5 billion at $185 per share, significantly above the $150-160 range, with orders for '20 times more than the available shares' and a $40 billion market valuation.

ARM and SoftBank made a last-minute acquisition bid that Cerebris rejected, preferring to go public in what became one of the largest IPOs in recent years.

OpenAI is now supporting AI regulation including the Kids Online Safety Act, marking a shift from their previous Industrial Policy for the Intelligence Age white paper that was 'poorly received.'

OpenAI's Chris Lahane acknowledged AI companies will be 'crushed by public sentiment' unless they redistribute AI wealth, citing Alaska's oil revenue sharing as a model.

The End of AI Token Subsidies

Lambert argues this reflects 'the end of the AI subsidy era' as demand for tokens vastly exceeds supply, forcing a shift from seat-based to token-based pricing models.

Anthropic estimated average enterprise developer costs at '$13 per day, or between $150 and $250 a month' just for Claude Code usage in their own harness.

The compute constraint is fundamental: 'Even Colossus One is a drop in the hat' when dealing with 'stratospherically growing demand' and training next-generation models.

Lambert predicts OpenAI 'will be forced to make a very similar decision before very long' as enterprise customers pay six and seven-figure bills while consumer subsidies become unsustainable.