The month the pieces stopped waiting for each other

Every week a dozen developments break and none of them make sense alone. A grid number here, a robot demo there, a paper, a bond sale, a Slack handle. Read separately, they're noise. Read together, they point somewhere specific.

This week the through-line is a threshold. Across energy, robotics, and AI, a bunch of things that lived in the future tense quietly moved into the present. The forecast became a measurement. The clip became a shift. The tool became a colleague. Here's the pattern.

Renewables stopped being the supplement

For the first year on record, US wind and solar generated more electricity than coal. Coal built the American grid and was the default source for a century. In 2025 it slipped behind two sources that barely registered fifteen years ago.

The same shift shows up across the Atlantic. Ember's data on Britain has power prices increasingly decoupling from gas. For decades the gas price set the electricity price, because gas plants were the ones on the margin. That link is loosening as wind and solar carry more of the load.

Two markets, one move: renewables stop supplementing the system and start setting it. We keep talking about the energy transition as a forecast. This week it read as a scoreboard.

The humanoid stack is maturing one subsystem at a time

Watch the robotics demos and the same story tells itself in parts. Wuji Tech's Hand 2 moves metal fingers with the speed and precision of a human hand. Unitree's G1 takes a spoken command in plain language and turns it into physical action in real time. Dexterity here, speech-to-action there, locomotion already solved elsewhere. The pieces are showing up separately, on different bodies, from teams that don't work together.

Then there's the piece nobody sells. NRE-Skin is a neuromorphic electronic skin that copies the human reflex arc. Touch a hot surface and your hand pulls back before your brain decides anything. The skin senses contact, damage, or heat and fires a reaction locally, without waiting for a central model. Every humanoid demo sells the legs and the hands. Skin is the sensor surface that decides whether a robot is safe to stand next to.

And the scoreboard has quietly changed. Figure's F.03 humanoid is standing on the line at BMW's plant in Spartanburg. Not a rendered demo, not a staged warehouse. BMW's biggest US plant, a named customer, a robot on the actual floor. For years the question was: can it walk. Now it's: who's paying for one to work a shift. That bar means uptime, safety around people, a task that survives a full production day, and a cost a plant manager can defend.

Worth noticing where the other robots go. DEEP Robotics' Pulse walks into burning buildings. REACCH was tested aboard the ISS to grab defunct satellites before they turn one piece of junk into a thousand. The robot jobs arriving first are the ones humans physically can't or shouldn't do. Fire, vacuum, debris at 28,000 km/h. Watch which names show up on which floors.

AI crossed from answering our questions to producing new ones

An OpenAI model just disproved a math conjecture that stood for eighty years. The question goes back to Paul Erdős in 1946, about how points can be arranged in a plane. For decades the best constructions looked like tidy square grids, and most assumed that was the ceiling. The model found a different family that beats it.

Here's what separates this from the usual benchmark headline. Until now these systems climbed tests we designed, answering questions we already had answers to. This time a model worked on a prominent open problem at the center of a field and produced a construction nobody had written down.

That reframes what the field even is. Microsoft's chief scientist Eric Horvitz says he never liked the term artificial intelligence, preferring computational intelligence, because in his view the same class of process describes a biological nervous system and a machine. Sit with that and the map rearranges: the biggest wins won't stay inside chatbots and code. Horvitz expects the breakthroughs we remember from this decade to land in biology and medicine, filed under AI.

Someone already ran that experiment without meaning to. A tech entrepreneur in Australia with no biology background sequenced his dying rescue dog's tumour for about $1,000, ran the DNA through ChatGPT and AlphaFold to find the mutated proteins, and designed a personalized mRNA vaccine from scratch. The genomics professor who reviewed it was, in his own word, gobsmacked. One tumour has since roughly halved. Look past the tool: what changed is the distance between a hard problem and a person willing to attack it.

Cheap competence doesn't shrink teams. It stretches demand.

METR tracks how long a software task frontier models can finish at least half the time. In the GPT-4 era that was a few minutes. Today it's multi-hour work, and the curve doubles roughly every seven months. Dwarkesh Patel points at the cap on the next leap: continual learning. A model starts every session from zero. A new hire is slow in month one and carries context by month six. Once models keep what they learn on the job, the whole measure changes.

While that horizon stretches, the tools are moving into the room where work already happens. Anthropic's Claude Tag drops the model straight into Slack as a team member. It sits in chosen channels, gets tagged like a colleague, picks up delegated tasks in the same threads as your humans. Using AI used to mean leaving: open a tab, type, copy the answer back. Now it's inline with everything else.

And the headcount math is the part most plans miss. Dan Shipper automated every task he could with AI agents. His company went from 4 people to 30. Every workflow handed to agents freed people for work they couldn't reach before, so human work grew. When expert competence gets cheap, you find more places to put experts. Cheaper legal review means more contracts reviewed. Cheaper analysis means more questions asked. Demand stretches to fill the capacity. Most plans budget for headcount coming out. The teams pulling ahead budget for what to do with the room that opens up.

The constraints nobody put on the slide

Every threshold this week hides a bottleneck. By 2028, AI will want nearly five gigawatts of power for every one gigawatt of new data center capacity North America can build. In 2023 that ratio was 0.4x; supply comfortably covered demand. By 2028 the chart puts it at 4.8x. Everyone tracks compute doubling. Almost nobody tracks whether the grid can feed it. Silicon scales in a couple of years. A high-voltage transformer has a multi-year lead time, a substation needs permits, transmission lines need land and a decade of patience.

The money already knows. Twice in four months, Amazon went to the bond market to fund its AI buildout: $37 billion in March, at least $25 billion more this month. Meanwhile Meta started a cloud business to rent out its excess compute. For years the capex came from pocket money. Now it comes from debt and rented GPUs.

And the materials squeeze looks like a price chart before it looks like a headline. Tungsten sat below $0.5K per DMTU for over a decade. Then the chart went vertical. It's the quiet input behind armor-piercing rounds, cutting tools, and semiconductor contacts, twice as dense as steel, melting at 3,400°C, and China controls the majority of supply. Everyone watches chips and rare earths. Tungsten is what a critical-materials squeeze actually looks like once it hits the market.

One more constraint, human this time. Nature Medicine has a name for what AI is doing to medical training: never-skilling. We've worried about de-skilling, the surgeon or coder whose edge dulls from leaning on the machine. Never-skilling is quieter: a trainee who reads every scan alongside an AI from day one may never build the internal model to read one without it. A large study in China found the mechanism precisely: as students used AI to cut homework time, their scores dropped in lockstep. AI that removes the struggle deskills. AI that scaffolds the struggle doesn't. Same tool, opposite outcome. The thing that predicts whether AI helps is whether the effort survives contact with it.

The close

String the week together and one shape emerges. Things that were forecasts became measurements. Renewables set the price. A model produced new math. A robot clocked in at BMW. A dog got better. Each on its own is a headline. Together they mark the same crossing, from something coming to something here.

The interesting question isn't whether the capability arrives. This week says it already has, in pieces. The question is what feeds it and who forms the judgment to use it well: the grid, the transformers, the tungsten, and the people who still have to learn the work before the machine does it for them. The abundance and the bottleneck are arriving on the same week.

I do this every week in Navigating Exponential Change: take the developments that actually matter and read them as one story, so you leave with the pattern instead of the pile. If that's useful, you're welcome to subscribe.