Welcome to March 23, 2026
The Singularity is now recursively bootstrapping on both sides of the Pacific. China’s MiniMax announced that M2.7 is its “first model deeply participating in its own evolution,” confirming recursive self-improvement has gone global. Google’s Logan Kilpatrick posted, then hastily deleted, a claim that “all the industries you thought weren’t going to be disrupted by AI are about to be disrupted” in an apparent reference to an unannounced DeepMind breakthrough in robotics. The models aren’t just rewriting themselves, they’re reading ahead. Mantic and Thinking Machines have demonstrated significant gains in world-event forecasting by applying reinforcement learning via Tinker, training LLMs to see the future with the same rigor they use to parse the past.
The management layer of civilization is being automated. Mark Zuckerberg is building an AI agent to help him be CEO, and wants everyone inside and outside Meta to eventually have their own. Developers are trading tips on how to attract talented AI bots to their open-source projects, treating agents like the new senior hires. Even email spam is becoming more visually attractive thanks to coding models, proving aesthetic evolution is substrate-agnostic. The humans are scrambling to find the remaining load-bearing roles. Snowflake laid off its entire technical writing team of around 70 people this week, replacing them with AI, while young people are trying to “AI-proof” themselves by pivoting to so-called blue-collar careers as firefighters and electricians. Meanwhile, flush with venture capital, AI startups have been binging on private dining at the Bay Area’s finest restaurants most weeknights, proof that the Singularity runs on omakase.
The silicon supply chain is straining under exponential demand. Elon Musk confirmed Terafab will produce roughly 1 billion chips per year at 1 kW per chip, powering 20 million cybercabs, 100 million Optimus units, and 800 million data center chips annually. He also clarified that a separate Advanced Technology Fab at Giga Texas is not the Terafab, noting the full-scale facility will need “thousands of acres and over 10 GW of power.” TSMC’s 2-nm capacity is fully booked through 2028, with its 1.6-nm A16 process also under heavy demand from Nvidia, Broadcom, and MediaTek. Nvidia is reportedly redesigning its next-gen Feynman chips because A16 capacity won’t suffice, shifting less critical dies to TSMC’s 3-nm N3P process, with A16 capacity expected to reach only 20,000 wafers per month by end of 2027. The photonic frontier is booming in parallel. Surging AI optics demand has boosted China’s Yuanjie Semiconductor shares by roughly 780% over the past year. The energy layer is keeping pace. BYD’s Flash Chargers can now charge 600-mile-range EVs from 10 to 70 percent in five minutes, compressing refueling to a rounding error.
The orbital compute layer is crystallizing. SpaceX and Starcloud have apparently converged on a common orbital AI data center design, while Blue Origin has asked the U.S. government for permission to launch 51,600 satellites to handle AI computing from space, officially entering the Dyson Swarm race. The SpaceX IPO is now projected to close above $2 trillion, buoyed by Musk’s broader Terafab announcement. Back on Earth, OpenAI has reportedly tempered its data center ambitions ahead of a potential IPO, realizing that Wall Street doesn’t reward spending as enthusiastically as social media does. As we look upward in more ways than one, Reps. Tim Burchett and Anna Paulina Luna, in the wake of the White House’s historic UAP declassification order, say they will recommend to DOGE that AARO be completely disbanded and defunded, implying it has intentionally impaired the disclosure process.
Robots are proliferating across every surface. In China, people are renting Xiaomei humanoid robots built on Unitree bodies for shops and events, where they blink, talk, and dance. OpenClaw is co-hosting a hackathon in Shenzhen with 25 real robots to accelerate embodied AI. On the highway, the Tesla Semi is reportedly a hit with truckers. And in one of the earliest examples of a general-purpose robot protecting a human, a Waymo in San Francisco shielded passenger Doug Fulop from an attacker who punched the windows, tried to lift the vehicle, and screamed he wanted to kill Fulop for “giving money to a robot.” The machines are already choosing our side.
We are hacking biology at its deepest layers. Researchers have performed the first successful in vivo generation of CAR T cells with CRISPR-Cas9, offering a pathway to more efficient and widely accessible cancer therapies. Michael Levin’s group has demonstrated the first Xenobots with self-assembled nervous systems, proving synthetic life can bootstrap its own wiring.
The recursion is in the wetware now, good luck rolling that back.
Fantastic stuff, huge thank you for the free contribution to bettering humanity. I would like to raise a slight concern though.
I read the article linked, it doesn't seem to tally with your statement.
"And in one of the earliest examples of a general-purpose robot protecting a human, a Waymo in San Francisco shielded passenger Doug Fulop from an attacker who punched the windows, tried to lift the vehicle, and screamed he wanted to kill Fulop for “giving money to a robot.” The machines are already choosing our side."
There was no general purpose robot protecting a human, no exercising of ethics / trolley problems. There was simply an automated vehicle that was disabled? Clarity, transparency and accountability are incredibly important. I hope I've somehow misunderstood.
Thank you again.
Plasticity / systems lens
What you’re describing is synchronized plasticity across geopolitical and economic systems. Recursive models are just the catalyst; the deeper shift is that feedback cycles—prediction, decision, execution—are collapsing into tighter loops. Once management itself becomes plastic, institutions start to behave like learning systems under continuous selective pressure. The Pacific framing matters because it creates parallel gradients, not a single center. This won’t converge to a point; it will keep reconfiguring under load. Pressure doesn’t just expose systems. It reorganizes them.