Structural shift

OPEN :: structural shift :: tier HIGH

Registered claim

Through 2026-09-30, the binding bottleneck on AI compute deployment shifts away from GPU compute itself (NVDA/AMD/AVGO supply) toward four named adjacent constraints (memory: HBM3E/HBM4 capacity; networking: optical interconnect / scale-up fabric; power: chip-to-grid + behind-the-meter gas + rack power; cooling: liquid-cooling adoption), and equity outperformance follows, with these four pure-play constraint stacks meaningfully outperforming pure-GPU-exposure on a 6-month basis.

Registered 2026-05-10.

Resolution spec

Spec registered 2026-06-12.

• [A] CORE :: At least one of the four named constraint domains (memory / networking / power / cooling) is explicitly cited as a binding constraint on AI compute deployment in hyperscaler or apex-supplier primary commentary during Q3 2026.
  kind: dated binary event :: event date 2026-09-30 :: evidence: primary source
• [B] confirm :: Four-domain constraint-stack pure-play basket (equal weight: MU, LITE, COHR, MRVL, ETN, VRT, CEG, VST, MOD, CC - the falsifiability cohort) outperforms the pure-GPU basket (equal weight: NVDA, AMD, AVGO) by >= 500 bps over the spec window (registration to horizon). Basket-performance template instantiated from publication-policy.yaml STRUCTURAL_SHIFT.outperformance for the claim's 'meaningfully outperforming on a 6-month basis' predicate; constituents frozen at spec registration.
  kind: basket performance :: spread threshold 500 bps
  basket (equal): MU, LITE, COHR, MRVL, ETN, VRT, CEG, VST, MOD, CC
  benchmark (equal): NVDA, AMD, AVGO
  window: 2026-06-12 to 2026-09-30

Verdict rule: HIT if leg A and leg B. PARTIAL if leg A and not leg B. MISS if not leg A.

As frozen: HIT = A && B; PARTIAL = A && !B; MISS = !A

Tier history

• 2026-05-15: HIGH (initial). Trigger: Hold tier=high within photonics frame. The argument review capex-cascade-physical-constraints reinforced (90.26 weight, zero contradicts), optical-architecture-transition reinforced with falsifiability inversion (60.22 vs 1.75 weight; Hock Tan superseded by NVDA SpectrumX in production). NVDA-hyperscaler-GPU-not-bottleneck (May 2026) is the named narrative-attestation. Bottleneck-shift framing structurally validated in photonics frame. The stage review photonics thesis-level advanced cohort cadence → consensus building today.

Evaluated daily since registration: 191 evaluations. The 1 that moved the tier is listed above.