The grid publishes everything about itself. It just found the one thing it can't see.

On 10 July 2024, in the stretch of Virginia that carries more data-center load than most countries carry in total, roughly 1,500 megawatts left the grid in under a minute and a half. Not a power station tripping. The load — the data centers themselves — disconnecting. And here is the detail that should stop any operator, engineer or marketer cold: the grid operator did not order it, and largely could not see it coming. As NERC's own review of the event puts it, verbatim: "None of this load was disconnected from the system by utility equipment; rather, the load was disconnected on the customer side by customer protection and controls."

The machines guarding the data centers made the call. The grid found out afterwards.

That one sentence is the whole of Edition #4. Because it exposes a gap nobody has priced — and, underneath it, an inversion of everything you think you know about who owns valuable energy data.

Here's what six weeks of signals, and one very uncomfortable regulator document, taught us.

1. What the machine heard: energy was hiding in plain sight

A reminder of how our engine works, because it matters for what follows: we call it signal-first. We never start from an opinion about where the data is. We start from facts — dated, sourced, external. A grid-connection queue. A storage tender. A regulator opening a consultation. A datacenter interconnection fight. Each signal is a real article from the trade press, with a URL and a date, or it doesn't exist for us.

Between 28 May and 22 July 2026, our map surfaced 721 unique press signals across the physical economy. Sorted by trade-press family, one filière came back on top — and it wasn't the one that made Edition #3:

Energy: 214 signals — 29.7% of everything we collected, ahead of freight and supply chain (201), and well ahead of automotive (82). Behind those 214 headlines sit 135 data holders — 28.2% of our published map — the largest single cluster we've mapped, with the highest median opportunity score of any press family.

Now the uncomfortable, honest part — and it's the reason this edition exists. Energy is not a sector in our own labels. Those 135 holders were filed by our own classifier under two catch-all buckets — "industrial" and "other." No sector cut we could run would ever have surfaced them; they were dissolved into an umbrella of hundreds. It took an exogenous source — the trade press, which we don't control — to reveal energy as the first cluster on the map. When your own taxonomy hides the biggest thing on it, you learn to trust the outside signal over the inside label.

And note what the 214 headlines are about, because it tells you this is not Edition #2 in a new coat: solar and wind (20.1%), grid and connection (12.6%), storage and batteries (11.7%), datacenter and AI load (7.5%), regulation (7.0%), nuclear (6.5%), tariffs and markets (6.1%). Maintenance — the theme of Edition #2 — appears in five headlines out of 214. The real economy is not talking about broken machines here. It's talking about connection queues, storage, tariffs, and a new kind of customer that can move the whole system.

2. The obvious story — and why it's exactly backwards

The easy pitch writes itself: energy is the biggest cluster, the grid is being rebuilt around AI, so the grid operators must be sitting on the most valuable data in the economy — help them sell it.

We went to check. It is precisely wrong, and the way it's wrong is the point.

Grid data is the most published data estate in the economy — and it is free by law. In Europe, Regulation (EU) No 543/2013 has required transmission operators to submit generation, load and transmission data to the ENTSO-E Transparency Platform, which has published it openly since 2015. In the United States, the EIA and the system operators — PJM, CAISO, ERCOT — publish operational data as a matter of course. Britain's regulator states its default in two words: "presumed open."

So we did the obvious thing and went looking for the price. We searched our map for a single transmission or distribution operator, or a single system operator, selling or licensing its grid data. We found none — no tariff, no licence, no transaction. Grid data isn't a moat. It's a public utility. Anyone pitching a grid operator on "monetise your data" is trying to sell water to a river.

Which raises the only interesting question left: if the grid's own data is free and abundant, then what, exactly, is scarce?

3. The inversion: the grid is the buyer now

The scarce thing is the one piece of data the grid doesn't have — the behaviour of the enormous new loads plugging into it. And this summer, the reliability regulator did something that gives the whole game away.

NERC issued a Level 3 alert — its most serious tier — on large computational loads. Read what it actually asks for. Recommendation EA #6, verbatim:

"TOs should install and utilize dynamic fault recording devices to capture and share computational load facility electrical performance during system disturbances."

Three things about that sentence, because the easy reading gets each one wrong:

• It says should, not must. The alert disarms itself in its own text: it "does not create a mandatory obligation," and non-compliance "will not be subject to penalties." The "Computational Load Entity" that would one day carry a real duty is proposed — not created.
• It doesn't ask anyone to invent new data. It asks operators "to install, or grant access to existing, dynamic fault recording devices." The data very often already exists — on the customer side. It simply isn't captured centrally, and it isn't shared.
• The party that needs the data is the grid. The party that holds it is the operator of the load. The buyer and the seller have swapped ends from where your intuition puts them.

That is the inversion, and it's the piece that makes this a new edition and not a rerun. In every prior edition, an operator held scarce data that some outside buyer wanted. Here, the operator — the data center, the electrolyser, the industrial site — holds data that the grid itself needs: to forecast, to model, to keep a system standing when 1,500 megawatts can decide, on their own protection logic, to leave in under two minutes. The grid became the customer.

4. The tell nobody picked up: "open" energy data is quietly narrowing — and AI is the stated reason

Here is the freshest signal in the whole file, and as far as we can tell no one has connected it to the story above.

The most open energy regulator in the world is narrowing what "open" means — and it names AI as the reason. On 29 May 2026, Ofgem opened a consultation, Securing Open Data in Energy. In it, the regulator proposes to "increase the proportion of data being triaged as 'shared' rather than 'open'," and writes, verbatim: "The increased use of AI has also changed how Open Data might be used…"

Sit with that. An "open by default" regime, in Europe, in the sector that publishes more data than any other, is adding a layer of friction — and pointing at AI as the cause.

But we owe you the sentence it would be convenient to leave out, because leaving it out would be dishonest and, frankly, attackable. Ofgem also states, verbatim: "This does not represent any retreat from Open Data, or the principle of 'presumed open.'" Take both sentences together and you get the true signal — not a wall going up, but a value being noticed. The moment data becomes worth training a model on, "free and open" stops being a reflex and becomes a decision. (The consultation closed on 14 July 2026. This is live, not history.)

5. Who actually pays for grid data — and why the honest answer is "no one, yet"

Here we have to be more honest than a marketing story usually allows, because the honesty is the finding.

We looked hard for the price of grid data. There isn't one — as we said, the grid's own data is free by mandate. So we looked for the price of the scarce data, the load-side behaviour the grid now needs. And across our entire map we found no tariff, no licence, no clearing transaction — at any operator, anywhere. The market that ought to exist doesn't yet. "Who pays for this data" has, today, no observed answer. That is not a hole in our research. It is the result.

And resist two tempting overstatements, because our own checking killed both:

• "The forecasting models are starving for volume." They're not. What the evidence actually shows is a quality problem, not a quantity one — regulators have documented large loads being double-counted as the same demand is submitted to several utilities at once. The scarce thing isn't more data; it's trustworthy, reconciled data. If you've followed editions #2 and #3, that's the whole d-nvest argument restated: the value was never volume. It was provenance and fit.
• "The regulator is ordering this into existence." It isn't — yet. It published a should, a proposal, and a list of what to ask for. The obligation is a trajectory, not a fact: an alert now, a register proposed, a standard likely by the end of 2026.

Strip the overstatements away and what's left is cleaner, and more valuable, than the hype: a sector that gives its own data away for free, sitting next to a class of data it urgently needs and cannot see, with no market, no price, and no trusted intermediary between the operator who holds it and the grid that needs it. That missing layer — qualification, provenance, matching — is the entire thing we build.

The bottom line

Editions #1–3 found operators sitting on scarce data that outside buyers wanted. Edition #4 is the mirror image: an entire sector — energy — whose own data is free by law, and whose most valuable data isn't the grid's at all. It's the load's. And the buyer isn't a frontier lab or an insurer. It's the grid itself.

One date makes this urgent rather than theoretical. NERC's Level 3 alert carries a response deadline of 3 August 2026 — a few days from now. We're publishing while that window is still open, before the "should" hardens — as it likely will by year-end — into a register, and then a standard.

If you operate a large load — a data center, an electrolyser, a heavy industrial site — you are sitting on performance data the grid now needs to model a system you have become big enough to move. If you build AI for the energy transition — forecasting, grid modelling, flexibility, interconnection — the data you need isn't on the web and isn't for sale yet; it lives with those operators. Between the two there is no market. That's the layer we're building.

One honest note on geography, because it's the discipline this edition demands: our energy holders skew European — the UK, Germany, France, the Netherlands — while the sharpest regulatory action this summer is American. Read that as the preview it is. The load-visibility problem arrives everywhere the grid meets AI — and Europe's own most-open regulator is already the one narrowing the door.

— Salim Labriki, d-nvest

Methodology note: a "signal" is a dated external fact — a press article with a URL and a date — never an inference. Energy did not appear as a sector in our own taxonomy; it was surfaced entirely by third-party trade press, which is why we treat it as an exogenous finding rather than a reflection of our own labels. Regulatory quotes are string-matched against the primary PDFs (NERC, Ofgem); where a primary site blocks automated access (EUR-Lex; and the NERC PDF, which requires a browser user-agent) we describe the regime rather than re-quote it. We publish no price for grid data because none was observed: transmission, distribution and system-operator grid data is published free by mandate, and no market for load-side data yet exists — an absence we report as a result. Our own platform figures are dated and describe the holders our engine reaches, not the world.