Why a map of bottlenecks, and not a map of winners

Most maps of technology are maps of winners. They list the companies with the largest market capitalisations, the models with the highest benchmark scores, the factories with the newest equipment. Those maps are easy to build, because the subjects publish their own numbers, and they are almost useless for deciding what to do next, because everybody already has them.

This is a map of the opposite thing: what the winners are waiting for.

The claim underneath it

Progress in a technology is set by the slowest input that cannot be substituted. Not the average input, and not the most expensive one. If a datacentre can be designed in months, financed in weeks, and filled with accelerators in a quarter, but cannot be connected to the electricity grid for four years, then the grid connection is the technology's speed. Everything faster than that is, for the moment, free.

That is an unglamorous thing to say about a field that mostly talks about model architectures. It is also testable, which is the reason to say it.

Why the constraint is usually invisible

Three things push the binding constraint out of view.

It is upstream. A buyer sees their supplier. They rarely see their supplier's supplier, and almost never the step where a material is upgraded from ordinary to qualified. Tin is not scarce. Tin at seven nines, qualified for a lithography light source, is made by a handful of firms, and that sentence is invisible from anywhere downstream of the wafer.

It is boring. Grain-oriented electrical steel gates how fast datacentres can be energised, and nobody has ever written a headline about it. The steel is not interesting. What it holds up is.

It is administrative. An interconnection queue is not a factory, a material, or a machine. It is a waiting list. It cannot be relieved by capital and it does not appear on any balance sheet, which is exactly why it is so often the thing that binds.

What this map does about it

Every row here answers four questions and refuses the rest: why it holds things up, who makes it, what has happened to it lately, and what would remove it. Each answer is either backed by a source you can open or marked as unverified, and the site shows which. The front page prints how many maker rows are sourced as a fraction rather than rounding it up.

A map of this kind is only worth reading if it is honest about what it has

not checked.

That is also the reason there are no prices here, and no recommendations about what to buy. Both would require confidence this research has not yet earned.

What would show this is the wrong map

If constraints resolved at the visible layer, so that the amount of compute the world could deploy tracked chip design announcements rather than packaging, power and tooling, then this map would be describing a chain that no longer gates anything. That is the falsifier, and it is worth watching for.

Until then, the interesting question is not who is winning. It is what they are queuing for.