Good morning. It's Friday, May 22, and we're covering China’s growing resistance to NVIDIA chips, the rising concentration of power inside AI giants, and why California is preparing for a potential wave of AI job disruption.

California is weighing ideas once considered fringe, including universal basic capital, as officials brace for AI-driven white-collar job losses. Gov. Gavin Newsom signed an executive order directing agencies to study subsidies for companies that retain workers and expand retraining programs. The move follows layoffs at Meta, Intel, and Amazon, while Dario Amodei warns half of entry-level white-collar jobs could vanish within five years.

Anthropic co-founder Jack Clark says AI could help produce a Nobel prize-winning discovery within a year, while fully AI-run businesses may generate millions in revenue within 18 months. Speaking at University of Oxford, Clark predicted robots would assist tradespeople within two years and AI systems could design successors by 2028. He also warned advanced AI still carries a “non-zero chance” of catastrophic harm, citing rapid global competition and Anthropic’s powerful new model, Mythos.

Donald Trump has delayed an executive order that would let U.S. agencies evaluate advanced AI models before release, arguing parts of the proposal could slow America’s lead over China. The order was expected to require security reviews following concerns around Anthropic’s Mythos and OpenAI’s GPT-5.5 Cyber models, which reportedly exposed cybersecurity risks. One disputed provision would force AI firms to share models with the government 14 to 90 days before launch.

Spotify announced a licensing deal with Universal Music Group on Thursday that will let Premium subscribers create AI-generated covers and remixes. Sold as a paid add-on with revenue sharing for participating artists. The deal arrives after Suno settled a $500M suit with Warner Music last year and UMG settled separately with Udio, while both AI music startups still face ongoing copyright claims from other major labels.

FRIDAY FACTS

For two decades, every civilian GPS device on the planet was reading less accurately than the technology actually allowed, and the U.S. government was doing it on purpose. Full story at the bottom of today's newsletter.

Higgsfield is the world's first AI-powered streaming platform, dropping original series made entirely by artificial intelligence. No studios, no gatekeepers, no $200M budgets. Just stories that shouldn't exist yet, streaming right now for free.

Stream for Free.

Matt reacts to Andre Karpathy joining Anthropic, warning that fear-driven AI narratives are overtaking optimism and open innovation.

SpaceX’s IPO is asking investors to back far more than its rocket and satellite businesses, with the company seeking a valuation approaching $2 trillion despite reporting a $4.28 billion quarterly loss for the three months ending March 31, 2026. Reuters reports the investment thesis hinges on a chain of ambitious bets: Starlink profits funding the next-generation Starship rocket, lower launch costs expanding the space economy, and that expansion supporting a future AI business.

Analysts say traditional valuation metrics are difficult to apply given SpaceX’s heavy spending and reliance on Elon Musk’s long-term vision, which includes Mars colonization and space-based data centers. → Continue reading here.

Six months after President Donald Trump approved NVIDIA’s H200 AI chip for sale in China, Beijing has yet to allow a single domestic company to buy one. Chinese officials instead continue steering firms toward local chipmakers such as Huawei and Cambricon, reflecting Beijing’s broader push for technological self-reliance.

NVIDIA CEO Jensen Huang acknowledged the slowdown during a May 2026 visit to Beijing, even as Chinese AI firms report severe shortages of computing power. → Read the full article here. (Paywall)

Researchers are using artificial intelligence to design “metamaterials” that mimic the strength, flexibility, and porous structure of human bone, potentially improving hip replacements and fracture healing. Teams at institutions including Leiden University Medical Center, TU Delft, and ETH Zurich have developed AI-generated materials that can absorb stress more effectively and integrate better with surrounding bone tissue than conventional titanium or steel implants.

One project created unusually stiff “auxetic” materials that become thicker when stretched, while another produced soft, porous structures designed to support early-stage bone healing. Scientists say the approach could eventually lead to personalized implants tailored to individual patients and minimally invasive implants that expand inside the body after insertion. → Read the full article here.

OpenAI AI Cracks Math Conjecture: Its model disproved part of Paul Erdős’s 1946 planar unit distance theory, marking a milestone for AI-assisted math.

Hark Raises $700M for AI: Brett Adcock’s startup hit a $6B valuation as it expands infrastructure and AI models for personalized voice-first assistants.

SoftBank Surges on NVIDIA AI Boom: Shares jumped 20% after NVIDIA earnings boosted confidence in AI demand, Arm Holdings, and OpenAI.

Here’s how you voted:

🟩🟩🟩🟩🟩🟩 Anthropic (43%)
🟨⬜️⬜️⬜️⬜️⬜️ OpenAI (13%)
🟨🟨🟨⬜️⬜️⬜️ Google DeepMind (21%)
🟨🟨🟨⬜️⬜️⬜️ Start my own lab (21%)

FRIDAY FACTS

From the moment GPS opened to civilian use, the U.S. military deliberately degraded its accuracy for everyone outside the armed forces. The program was called Selective Availability. It introduced artificial errors into the signal, capping civilian precision at roughly 330 feet. Military receivers, with the correct decryption keys, got the real signal. Civilian users got the scrambled one. On May 1, 2000, President Clinton ordered Selective Availability switched off. Overnight, civilian GPS accuracy jumped from ~330 feet to ~66 feet.

The reason it finally ended had less to do with goodwill than with technology catching up. The military had developed a new capability: jamming GPS over a specific geographic area without degrading the global signal.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team