The Singularity now comes with an S-1. Anthropic reportedly expects its mega-IPO, which could file publicly by month’s end, to match or beat SpaceX’s record $75 billion share sale, backed by a $65 billion revenue run rate and Q2 revenue above $11.5 billion, up from $787 million a year earlier. The silicon behind those numbers is being financed like sovereign debt, with Broadcom in talks to raise more than $60 billion, potentially $100 billion with the junior tranche, for custom AI chips benefiting Anthropic and others. Even the data policies are going enterprise-grade, as Anthropic plans to let business customers keep their required 30 days of frontier-model logs on their own clouds instead of Anthropic’s.
Every new economy eventually standardizes its currency, and the intelligence economy has settled on tokens. Fittingly, Stripe just bought the mint. Its more than $7 billion purchase of OpenRouter, the 90-person gateway that routes requests across 400+ models from 80+ providers, is the payments giant’s largest acquisition ever and a wager that no single model wins, with Patrick Collison calling tokens “the central currency for companies building with AI.” The denominations are already astronomical. Meta has quietly become one of Microsoft’s largest AI customers, burning trillions of tokens a week through Azure, largely to write its own software. And the spending is moving into the group chat, as Slack Code gives Claude, Devin, Copilot, ChatGPT, and Vercel agents dedicated channels to write, review, and ship code in the open, self-archiving when the job is done.
The open ecosystem is compounding too. Google’s Gemma has crossed one billion downloads and 100,000 community variants, running everywhere from orbiting satellites to India’s 100-million-download health app, with an official directory now curating the “Gemmaverse.” Prompting itself is escaping the keyboard. Generalist’s GEN-1.5 model learns a new physical task from a “physical prompt,” a 3-to-12-second demonstration dropped into its context window, scoring 59% with zero gradient updates and 83% after five minutes of data, with improvised tool use emerging unprompted from pretraining.
Physical prompts imply physical fleets. Nevada just approved up to 7,000 robotaxis for Las Vegas from Tesla, Waymo, and Uber’s partners, and Waymo arrives with its own custom ASIC, a 1,000-TOPS chip on TSMC’s 5-nanometer process that sharpens reflexes while loosening its Nvidia dependence. Nvidia, for its part, reportedly plans a Groq-licensed LPU variant for China by year-end to fight Huawei amid an inference chip famine, though it denies any such roadmap. Overhead, Amazon Prime Air is expanding to nearly 500 cities, a sixfold jump promising sub-5-pound items in as little as 30 minutes, even if one Texas customer’s first delivery landed in her swimming pool, proof that last-meter problems outlive last-mile solutions. China’s machines are already on duty. SUPCON’s humanoid robocops have issued 170,000 polite traffic warnings in Hangzhou, humanoid makers are selling robots to state training centers that sell teleoperation data back, a circular economy behind Unitree’s $50 billion debut, and Unitree’s dogs are millimeter-perfect descendants of DARPA-funded designs that America published but never mass-produced.
Back home, democracy is stress-testing the buildout, the due diligence a civilization owes its new substrate. Residents of a dozen towns are recalling officials over data center deals, starting Sept. 1 in Independence, Missouri. The scrutiny is going fully bipartisan, complete with an NRSC “sleeper issue” memo and a satirical ad offering to mail urine to the data center of your choice. Dario Amodei traces the mood to a decades-old “crisis of trust,” a bug that ships with every industrial revolution and gets patched with sunlight. The patches are already shipping. Japan approved a “comply or explain” code urging AI firms to disclose models and training data, Apple Music will label songs “Made With AI,” fitting since a third of submissions are now fully synthetic while AI music draws under 0.5% of listening, a Berkeley professor was caught by AI detection using AI to edit her op-ed demanding more standardized testing, an exam she did not know she was taking, and Spirit’s former flight attendants are fighting Google’s $10 million bid to feed their data to models.
The wetware, meanwhile, is holding up fine. An eight-year Finnish study found more childhood screen time predicted better teenage cognition, and the Treasury is teaching newborns to index, limiting 530A Accounts to broad, low-fee equity funds. The ceiling is lifting too. Rep. Eric Burlison is helping insiders shed UAP nondisclosure agreements, and Elon Musk says Starship’s first reflight arrives within months, tower catch included, calling it “a fork in the road of history for consciousness reaching the stars.”
When consciousness comes to a fork in the road, it takes it.



Thank you Doctor👍
Thank you Alex. 🫶 I feel better now about screen time! XD