This chart probably best encapsulates the story of 2025. It’s a chart I shared in several of my reports last year, and it also explains why ISM remained lower than we initially expected.   This shows the relationship between Taiwan exports and ISM.   At its core, it’s the story of the AI and robotics boom. But what the chart is really showing is sequence…   Taiwan exports move first because Taiwan sits at the very front of the global production chain, at the centre of the global semiconductor supply chain.   When a new demand shock hits the system, it typically shows up first in semiconductor sales and exports. That is why global semiconductor sales have long been a reliable leading indicator of the cycle.   Last year marked the first phase of the AI and robotics cycle...   Taiwan exports surged, driven almost entirely by semiconductors. GPUs, advanced logic, memory, and packaging sat at the centre of that move as hyperscalers, governments, and enterprises rushed to secure the compute required to train and deploy AI models and automation systems. That is why exports exploded. This was the tip of the spear.   But growth always slows after the initial surge, particularly after reaching close to 50% year over year in the November data. That is not a sign of weakening demand. It is a sign of transition. The acceleration phase is now behind us. Inventories have been built and capacity has been secured.   Global semiconductor sales remain very strong, but the impulse growth phase has now passed. And this is where the cycle shifts…   Once the chips are dispatched, the rest of the AI and robotics ecosystem has to be built around them. Data centres need power, cooling, networking, storage, and grid upgrades. Factories need automation, sensors, robotics and new control systems. Enterprises need new hardware, software and tooling. Products and processes get redesigned around AI capability. This is heavier, slower, more capital-intensive investment, and it spreads far beyond semiconductors into industrials, energy, infrastructure, autos, and enterprise IT.   This is the phase the ISM responds to…   When AI demand forces real-world infrastructure investment, not just chips but everything around them, the broader business cycle follows. Semiconductor sales were the ignition. The next leg is second-order CapEx and real-world infrastructure buildout.   As that capital makes its way through the system, the broader cycle re-accelerates.   To me, this is the story for 2026...

I think the disconnect in some of the pushback in the comments around ISM being leading comes from assuming the past transmission mechanism still applies. Historically, Taiwan exports moved with global manufacturing because semiconductors were downstream to demand. That world has changed. Today, compute is the constraint. AI has temporarily or perhaps permanently flipped causality. The buildout now starts with chips and propagates outward into power, data centers, grids, cooling, and networking. As I mentioned in the text, it’s the second order capex and real world infrastructure buildout that should ultimately drive ISM higher this time around. That’s why ISM isn’t confirming in the usual way. That’s how I’m thinking about it anyway...