Welcome to June 2, 2026
The Singularity has stopped being a finish line and become a leaderboard. Opus 4.8 just posted a “breathtaking” state-of-the-art 1.5% of human efficiency on ARC-AGI-3, tripling GPT-5.5’s score and proving the gap between frontier models is now measured in multiples per quarter. The open-weight pack is sprinting to close it. Nvidia’s Nemotron 3 Ultra became the smartest open US model at roughly 550B parameters and 48 points of AAII intelligence, though China’s Kimi K2.6 still leads at 54. MiniMax’s new M3 claims to be the only open-weight model fusing frontier coding, a million-token context, and native multimodality, while undercutting Opus 4.7 at $0.12 per million input tokens against $5. As raw capability commoditizes, the real frontier becomes control. Anthropic is giving the EU’s ENISA access to Mythos, making it the first European agency in Project Glasswing to wield a model officials quietly fear could be turned on the vulnerabilities of critical infrastructure, while the Linux Foundation’s new DNS-AID teaches agents to find each other safely over plain old DNS. The cautionary tale already landed. Hackers seized the Instagram accounts of the Obama White House, a Space Force chief, and Sephora just by politely asking Meta’s support bot to swap the email, the first social engineering attack where the social was optional.
When the model writes the code, the institution rewrites itself. Microsoft is reportedly stitching GitHub Copilot, chat, Cowork, and a new “Autopilot” agent into one Copilot super app, collapsing the entire toolchain into a single surface. But consolidation has a cost. A historian-turned-engineer warns that “the oral tradition that built software may not survive AI,” as the mentor-to-junior handoff of institutional memory dissolves into model weights. The next cohort may never learn it the old way anyway, since the American Federation of Teachers, the country’s second-largest teachers’ union, just released a ten-point plan to fence off classrooms, capping AI use and banning screens through second grade “unless there is a compelling reason.”
Silicon is being recut for every altitude of the stack. Nvidia named Anthropic, OpenAI, and SpaceX as launch customers for its Vera CPU, its first standalone data center microprocessor, then shrank that ambition onto a desk with the DGX Station, a “deskside AI supercomputer” packing 748 GB and 20 petaFLOPS for trillion-parameter models. It shrank further still into the RTX Spark, an Arm consumer chip for laptops built on TSMC 3 with MediaTek that Jensen Huang calls “the first completely reengineered, reinvented line of PCs that has happened in 40 years,” shipping this fall from Dell to Lenovo. The demand is already booked, as HPE’s revenue jumped 40% to $10.7B, well past the $9.74B forecast, on a 33% surge in server sales. And the stack now ends in a body, with Nvidia’s open Isaac GR00T reference humanoid fusing a Unitree chassis, Sharpa hands, and Jetson Thor into a blueprint anyone can build. One architecture now spans rack, desk, laptop, and body.
All of this runs on capital and copper. Alphabet is raising $80B in equity, including a $10B vote of confidence from Berkshire Hathaway, to feed its infrastructure buildout, even as Ohio suspends the tax break that made it a data center magnet, an early sign the industry must finally pay the full cost of its own warehouses. That freight is fragile, so the US, UK, and Australia are developing AUKUS undersea drones to guard the cables and pipelines carrying the world’s data and power. Africa, meanwhile, is leapfrogging straight to clean, with 173 of its 322 new energy projects in 2025 being solar, trailed by hydropower, wind, and gas, as renewables decisively overtake the old power plants on a continent skipping the fossil era altogether.
The money is rotating into a new physics of value. SoftBank just passed Toyota to become Japan’s most valuable company, ending the automaker’s twenty-year reign and crowning Masayoshi Son’s AI bet over the combustion engine as Tokyo stocks hit an all-time high. The public markets are bracing for the flood, as SpaceX’s IPO, expected to more than double any in history, forces index providers to rewrite their own admission rules, while Honeywell-backed Quantinuum lifted its quantum IPO to a $14.3B valuation. Not everyone made the cut, as hundreds of pre-ChatGPT startups now sit stranded, more than 220 of them fallen unicorns, too inflated for fresh venture money and too unprofitable for Wall Street. The reckoning is going political too, with Bernie Sanders proposing an AI Sovereign Wealth Fund that hands the public “a direct ownership stake” via a one-time 50% tax paid in frontier lab stock, not profits. And the rotation spares no one, as even Michael Saylor’s Strategy net-sold bitcoin for the very first time.
Diamond hands, it turns out, are not forever.



Thanks for the update! Re watched the Nvidia presentation a couple of times to wrap my head around what that was going to mean for the x86. I want one of those laptops.
Teachers want to fence off classrooms, but they won't teach history. Time to fence off teachers...