Musk's trillion-dollar antimatter daydream is the wrong question to ask

Cold open

It was, by any reasonable measure, a passing remark. On 23 June 2026, Elon Musk weighed in under a thread pondering how much solar generation would be required to feed long-term AI ambitions — and floated the figure of a trillion dollars in antimatter, because of course he did. The exchange was surfaced by the market-newsletter account Unusual Whales, and the internet did what the internet does: it laughed, it memed, and it moved on.

That is the wrong reaction. The remark is silly, and the framing around it is even sillier. The serious question is not whether antimatter is a viable substrate for civilisation-scale compute — it obviously is not — but why a man whose companies now sit on roughly a third of the world's active machine-learning compute keeps getting treated as the right person to ask about the energy future.

The thesis

The bottleneck on the next decade of AI is not generation. It is interconnection, capital allocation, and who owns the electrons that already flow. Treating the energy question as a generation question — solar panels, fusion reactors, antimatter, whatever the month's exoticism happens to be — is a category error that flatters the people selling the exoticism and obscures the policy fights that will actually determine the outcome.

The Unusual Whales thread is the smallest possible data point. It still points at the largest possible misunderstanding.

Generation is the easy part

The US grid already added more than 30 gigawatts of utility-scale solar in 2024, according to the Energy Information Administration's monthly electric power inventory — a record, and one comfortably outpacing coal retirements. The pipeline of projects awaiting connection is larger still. Generation capacity, in other words, is no longer the binding constraint at the margin. It is the precondition, and it is being met.

The binding constraint sits two steps downstream. Permitting queues at the major regional transmission organisations now stretch four to seven years for large interconnections, a duration the Lawrence Berkeley National Laboratory has documented in successive interconnection queue reports. A project that has its panels in the ground cannot sell a kilowatt until a transmission planner has finished studying it, and the planner is studying roughly two thousand other projects ahead of it. That is the queue Musk did not mention.

The capital question underneath the capital question

The other thread from the same day is more revealing than the antimatter one. Wells Fargo lifted its 2027 S&P 500 earnings forecast on 23 June, and the accompanying note — again surfaced by Unusual Whales — made a quietly important point: almost the entire upgrade came from higher expected profits, not from investors paying a richer multiple for the same profits. The analyst was saying, in plain English, that the bull case is now about cash flow, not about sentiment.

This matters for the energy-AI story because the same discipline applies. The capital markets are no longer rewarding stories. They are rewarding delivered electrons, delivered tokens per second, and delivered inference margins. A trillion-dollar antimatter facility is a story; a 230-kV line from a West Texas substation to a hyperscale data centre campus is electrons. One of those two things the bond market will underwrite at investment-grade spreads. The other it will not.

What the elite-fitness framing gets wrong about AI

A second item circulating on 23 June, from the Epoch Times' research desk, argued that aligning workouts with circadian rhythm improves strength, recovery and focus. The link to AI is not the exercise physiology; it is the underlying assumption that timing, not magnitude, is the limiting variable. That assumption is correct for individual recovery and broadly wrong for civilisational-scale compute.

The AI build-out is not a recovery problem. It is a logistics problem. The constraint is not when the electrons arrive at the GPU; it is whether the electrons arrive at all, on a schedule that lets a hyperscaler sign a power-purchase agreement at a price a treasury team will accept. Treating AI energy policy as a biological-rhythm problem — feel the timing, get the cadence right — is a comforting metaphor for executives who do not want to think about transmission reform.

The serious paragraph

Here is what is actually at stake over the next thirty-six months. Roughly sixty per cent of new US data-centre demand is expected to land in three states — Virginia, Texas and Ohio — that already sit on congested transmission backbones. The projects will get built, because the capital is committed. The question is who eats the cost of the grid upgrades: ratepayers, taxpayers, or the hyperscalers themselves, in the form of larger interconnection fees and longer offtake commitments. That is a state-level politics story, fought in front of public-service commissions, not a moonshot story about antimatter.

If the AI-energy debate stays in the moonshot register, three things will follow. First, the utilities will continue to capture the rents. Second, the hyperscalers will continue to frame themselves as civilisation-builders, when they are in fact the largest new industrial electricity buyers in a generation and should be treated as such. Third, the actual policy fight — over who finances the lines, who owns the surplus capacity, and how the costs are socialised — will be settled by default rather than by design.

Kicker

Musk is welcome to spend his Saturday evenings replying to antimatter prompts. The rest of us should spend the week reading interconnection-queue reports, not X threads. The future of AI energy will be decided by line-rating studies and FERC dockets, and the people who understand that are not the ones tweeting about trillion-dollar physics experiments. They are the ones filing comments in Washington and Austin and Columbus, and they are the ones the markets should be listening to.