Analyst: I'm going to skip the CV if that's alright, we've got fifty minutes and I've got about forty questions. Start me at the worst month.
Expert: October 2024. Or November. They've run together.
Analyst: Go into it.
Expert: We'd signed for a block meant to land in September and it hadn't landed, and there was a training run scheduled that we had already — and this is the part that actually stings — that we'd already put in front of the board. You can absorb a slip. You can't absorb one you've already socialised. And I'd socialised it, so that one's mine. Then I spent something like six weeks doing nothing but managing the fallout, which meant I wasn't doing the job, which meant a second thing broke while I wasn't looking at it. Nobody writes that part up.
Analyst: Was the slip on silicon or on the site?
Expert: Site. It was always the site by then. In 2023 you genuinely couldn't get parts, that was real. Somewhere around the second quarter of 2024 the binding thing moved to power and to the building, and plenty of buyers never updated. Still negotiating like the chip was the scarce object. The chip wasn't scarce. The energised, cooled, cabled megawatt was scarce.
Analyst: Practically, what does that change?
Expert: If chips are scarce you're bidding against other buyers and money works. If the megawatt is scarce you're bidding against an interconnection queue and a transformer with a sixty-three-week lead time, and money doesn't work. We offered an embarrassing amount to jump a queue once and the answer was no. Not "no, pay more." Just no. There was nothing there to buy.
Analyst: How embarrassing?
Expert: I'm not giving you that number.
Analyst: Fine. So what changed in how you bought?
Expert: We stopped buying capacity and started buying dates. Which is a slogan, sorry. Concretely, we rewrote the term sheets so the committed object was an energisation milestone with damages hanging off it, rather than a quantity of accelerators. A contract for ten thousand units with no enforceable date is a wish. The first three we signed, late 2023, were wishes. Expensive ones.
Analyst: Did counterparties wear damages?
Expert: The small ones did. The big platform clouds didn't, and they were right not to, we were a rounding error against their capex. With the specialists — the people putting sheds of accelerators on cheap power — we had real leverage, because at one point we were forty percent of one provider's book. Sorry, thirty-seven. I've seen that number recently enough that I should say it properly. So we got damages, we got a step-down schedule, and in two of them we got a generation-swap right, which turned out to be worth more than everything else in the document put together, and nobody in that room cared at the time.
Analyst: Hold on. Thirty-seven percent of their book. How do you actually know that?