What We Got Wrong — What We Got Wrong

What We Got Wrong We ran two articles this week about AI data centre power demand — one framed around Babcock & Wilcox's gas-fired buildout, the other around global data centre investment overtaking oil supply spending. Both told essentially the same story from slightly different angles, and neither challenged the underlying assumption: that the projected demand actually materialises on schedule. The Economist piece we cited noted that McKinsey's AI pilot project success rate is less than 15 percent. We mentioned that number in passing, buried in paragraph eight, then immediately pivoted back to the bull case. That's backwards. A 15 percent success rate for enterprise AI adoption is the single most important number in the entire data centre demand thesis, and we treated it as a footnote. If enterprise AI disappoints — and a sub-15 percent hit rate says it might — every power demand forecast we cited this week is too high. On negative European power prices, we wrote that spot values could breach EUR -200/MWh this weekend and called the pattern "accelerating." But we also ran a piece on French nuclear drought risk pushing power prices higher. We didn't address the contradiction. Weekend negative prices and weekday scarcity-driven spikes can coexist — they're driven by different mechanisms — but we never explained that to readers. It looked like we were bullish and bearish on European power in the same edition. The Japan coverage leaned heavily on the Hormuz disruption narrative without acknowledging that Japan's LNG imports were already declining 1.5 percent year-on-year before the crisis. The structural shift toward renewables and nuclear restarts was reducing gas dependence regardless. We framed the Idemitsu Sakhalin crude purchase as emergency procurement, which it partly is, but Japan was always going to maintain some Russian supply for diversification. The crisis accelerated a direction Japan was already moving. Our carbon market piece on the EU's proposed rule change to hold surplus allowances rather than cancel them was solid on the mechanics but thin on the market reaction. We could have checked whether front-year EUA prices moved on the announcement. We didn't. The thinnest coverage this week was on physical gas flows. We had Kpler's EU gas demand forecast down 8 bcm to 314 bcm, and we had the Entso-G warning that storage may only reach 76 percent by October. Those two numbers together tell a story about whether Europe can fill storage at lower consumption levels, and we never connected them. That's the calculation a gas trader actually wants to see, and we left it on the table.