Daily Alpha · Substack
· Post-Market Alpha · by Buzzberg Research
The strongest specialist work mapped AI constraints by physical layer: grid and power first, then HBM and data movement, with architecture risk inside optics.
Themes on this desk
Power before compute
Power and grid equipment remain the dominant constraint on new AI capacity through 2027, favoring scarce equipment over fragmented construction exposure.
Memory and data movement
HBM supply is expected to remain tight even after 50-60% bit growth, while larger clusters raise networking's share of system economics; CPO remains a potential offset to optical content.
Distributed sites expand interconnect demand
If power and permitting force compute across more locations, geographic dispersion can enlarge the optical opportunity even as within-rack architectures evolve.
Power and grid as primary AI capacity constraints
Power and grid infrastructure are the dominant constraints on the turn-on timing of new AI capacity through 2027, with gas turbine orders facing multi-year lead times and 2030 slots largely booked.
Investors should prioritize companies with pricing power in power/grid equipment over those in fragmented construction markets, as scarcity in the former is more durable.
Watch Material improvement in transformer lead times or interconnection wait times for two consecutive quarters.
Source →Data center dispersion as an optical catalyst
If power and site constraints force hyperscalers to spread compute across multiple locations, the total optical opportunity may expand.
Geographic dispersion of data centers increases the need for high-speed interconnects, potentially offsetting risks from architecture changes like CPO.
Watch Trends in data center site selection and interconnect requirements.
Source →