Historical hypothesis, now retired from live research: Agent and reinforcement-learning workloads may add server CPU demand, constraining advanced substrates and packaging. Merchant CPUs, captive processors and substrate suppliers can capture different economics.

Why now

Retire the broad CPU-shortage basket until CPU allocation and substrate utilization establish a distinct constraint. MPS power growth is retained in power delivery, not promoted into proof of a CPU shortage.

In scope

Agent and reinforcement-learning workloads may add server CPU demand, constraining advanced substrates and packaging. Merchant CPUs, captive processors and substrate suppliers can capture different economics.

Out of scope

No current campaign-level investment recommendation · No reconstruction of past returns or reassignment of original calls

Expected horizon

Two reporting cycles after verified supply evidence; presently closed.

What supports it

Require AMD/Intel allocation and pricing evidence plus ABF/OSAT utilization and multi-quarter constraints; compare merchant versus substrate performance. Capacity ramps or captive substitution test different hypotheses.

What would break it

The closure is a research-priority decision, not a new market falsification verdict. Original evidence gaps and outcomes remain as recorded.

Next review On condition · Primary CPU allocation/pricing evidence plus ABF or OSAT utilization and a named beneficiary would justify a separate successor.

Closed 2026-09-06 · Retire the broad CPU-shortage basket until CPU allocation and substrate utilization establish a distinct constraint. MPS power growth is retained in power delivery, not promoted into proof of a CPU shortage.