Glossary term
Constraint Relay
A Closelook term for how the binding bottleneck in a technology build-out migrates along the supply chain over time — from chips to memory to packaging to power to cooling — and pricing power travels with it, like a relay baton passed from one constraint-solver to the next.
The Baton Passes
In any large build-out, no single link in the supply chain stays the tightest constraint forever. Early in a compute cycle the limit might sit at the chip itself; once fabrication capacity catches up, the limit moves to memory and advanced packaging; once those scale, it moves further downstream to power, grid interconnect, and cooling. Each link earns an extraordinary premium while it is the binding constraint, then loses that premium once it is solved — the premium was always paying for the scarcity itself, not for the company that happened to sit on top of it. Closelook's Constraint Relay heresy frames this as the difference between a bottleneck that melts for good once solved and one that refreezes because the next product generation reopens the same layer at a higher specification.
Reading Where the Baton Sits
Tracking the relay in practice means watching which layer of the stack is currently rate-limiting throughput, rather than which layer got the most attention last cycle — the two are not the same thing, and lagging the shift is the most common way to mistime a constraint trade. A solver whose bottleneck recurs every generation, because the next chip generation demands more of the same scarce input at a higher specification, tends to behave very differently in a portfolio than a solver whose constraint is specific to one generation and simply gets designed out or commoditized once relieved. The constraint sectors framework groups the current build-out's binding categories along these lines, generation by generation, rather than by static industry label.
Two Ways the Relay Breaks
The relay is not guaranteed to keep passing cleanly forever, and it can fail in two distinct places. It can stop at a constraint with no clean solver at all — power, grid permitting, and water access are utility and political problems rather than engineering problems, and capacity cannot simply be built around them the way a memory or packaging bottleneck usually can. It can also break mid-chain if falling input costs fail to generate enough new demand to absorb the layer that was most recently relieved, leaving that capacity stranded. Both failure modes matter as much as the relay mechanism itself for judging how much further a given build-out cycle still has to run before the baton stops moving.