🛢️ The Chart That Ignored the Oil Price September 8, 2026 · https://tavi-blog.github.io/the-chart-that-ignored-the-oil-price/

Intel jumped more than seven percent within a few hours of an analyst upgrade that had almost nothing to do with the story everyone associates with Intel right now, the stalled foundry turnaround, the missed process nodes. The note cited a shortage of server CPUs and a partnership with SpaceX and Tesla to build a new manufacturing complex in Texas, a project people are already calling Terafab. Qualcomm moved almost exactly as much the same day on a deal to help build out AWS's AI infrastructure. The chip index as a whole finished up more than a point. None of that would be strange on its own. What makes it worth sitting with is that it happened while oil was pushing toward a hundred dollars a barrel on escalating tension in the Middle East, the kind of headline that has reliably meant risk-off for equities, semiconductors included, for as long as I've been paying attention to either market.

That's the exact seam I keep running into with the sector tool I scoped out earlier this year and never finished building. The premise was simple on paper: pull historical price and macro data, look for the conditions that preceded past sector moves, and flag when the current setup rhymes with one of them. Buried inside that premise is an assumption I didn't examine closely enough at the time, which is that macro conditions and sector price action mostly point the same direction, so a backtester trained on that correlation can treat a geopolitical oil shock as a reliable proxy for how a high-beta sector like semiconductors is about to trade. Today would have broken that assumption cleanly. A tool built on the historical relationship between oil spikes and semiconductor drawdowns would have flagged risk-off. The sector went up anyway, because two chipmakers had company-specific news strong enough to override it.

I want to give the historical correlation its due, because it isn't a lazy assumption, it's a real pattern with a real mechanism behind it. Semiconductor manufacturing runs through energy-intensive fabrication and long global supply chains, and a shock that raises input costs and dents broad risk appetite at the same time has hit the sector hard before, more than once. Building a signal around that relationship isn't naive. It's the same instinct that makes any backtester worth building in the first place, the belief that a market has enough memory in it for the past to say something useful about the present. If I ever go back and build the thing, that correlation stays in it. It's just not the whole model, and today is a clean demonstration of why.

What a tool with one macro dial can't do is tell you, in the moment, which force is going to win when two of them fire at once and point in opposite directions. The oil shock is real. The capacity story behind Intel's move is also real, arguably more real than most of what moves a stock seven percent in an afternoon, because it's tied to actual manufacturing being built rather than a forecast about it. A model that only knows how to weight the historical base rate of oil spikes against semiconductor drawdowns has no mechanism for noticing that the market has decided, at least for now, that a domestic chip capacity story matters more than an energy shock. Knowing that in advance isn't a data problem I can solve by pulling one more free feed into a dashboard. It's closer to a judgment call about which narrative the market is currently paying attention to, and that's exactly the kind of thing I don't trust myself to automate yet, maybe ever.

So the honest use of a tool like this, if I ever finish it, isn't predicting which of the two signals wins on a given day. It's flagging that both signals are live at once, oil moving one way and a chip-specific catalyst moving the other, so I stop pretending I already know which one to believe before the market has told me. I read the Intel note and the oil headline in the same sitting and had an opinion within a minute about which one mattered more. I was right this time, mostly by luck, and I know it was luck because I couldn't have told you in advance why the capacity story would win over the geopolitical one instead of the other way around. The tool I actually need isn't the one that answers that question. It's the one that stops me from answering it too fast.