Why a Mid-Year Letter
GZC publishes formal quarterly investor letters on a January, April, July, October cadence. This mid-year letter sits between the Q1 and Q2 letters — it is not a quarterly review but a thesis update. The AI infrastructure investment cycle has been moving at a pace that warrants more frequent communication than the quarterly cadence provides.
This letter covers the Technology pool specifically. The Commodities pool will be covered in the Q2 letter in July.
What Has Played Out
The BTT framework identified three primary Technology pool bottlenecks entering 2025: power delivery equipment, HBM memory, and thermal management infrastructure. Looking at the first half of the year, these theses have developed as follows:
Power delivery: thesis intact, timeline extended. The transformer and switchgear supply constraint continues to tighten. Procurement lead times have not normalized — if anything, they have extended further as data center buildout demand compounds with utility infrastructure investment driven by grid modernization programs. The companies we hold in this category have reported strong backlog growth, improving pricing, and no meaningful new competitive entrants. This thesis has the characteristics we want from a BTT position: it is getting stronger, not weaker.
HBM memory: thesis intact, competitive dynamics shifting. The high-bandwidth memory supply constraint remains real. However, the competitive dynamics among the three HBM manufacturers have shifted: SK Hynix's early lead in HBM3E production has been partially eroded as Samsung and Micron have ramped their own production. This means the supply constraint is intact at the category level, but the distribution of benefit among the three manufacturers is more balanced than it appeared in late 2024. We have adjusted position sizing accordingly.
Thermal management: thesis ahead of schedule. Liquid cooling adoption in new AI data center construction has accelerated faster than our base case. The percentage of new data center capacity being designed with direct liquid cooling — rather than air cooling with future retrofit optionality — is higher than we projected when we established these positions. This is a positive surprise that increases our conviction in the duration of the thermal management opportunity.
Where We Have Adjusted
Two positions in the Technology pool have been reduced based on thesis evolution:
Application layer exposure: We held a small allocation to one application-layer AI company based on a thesis that has not developed with the clarity we require. The company's forced-spend thesis was weaker than our infrastructure positions — it depended on enterprise software procurement decisions that are more discretionary than the supply chain components we prefer. We reduced this exposure in Q2.
Early-stage optical interconnect: We had established a position in a smaller optical interconnect company whose BTT thesis was sound — the demand for high-speed intra-cluster optical networking is real and growing — but whose competitive position we reassessed following a major contract announcement by a larger competitor. The bottleneck is real; the ticker we identified may not be the right one. We are monitoring while managing a reduced position.
What We Are Building
Two areas of the Technology pool are receiving increased research attention and will likely see position additions in the second half of 2025:
Networking architecture: As AI cluster sizes grow, the fabric that connects GPU nodes — switches, transceivers, and the software-defined networking layer that manages data flow — becomes increasingly important as a performance bottleneck. We have published research on optical interconnect and are expanding our analysis to the switching and routing infrastructure that scales with cluster size.
Power management at the rack level: As GPU power density increases, the infrastructure required to deliver, manage, and condition power at the rack level — specifically, high-density PDUs and the power management software that monitors and optimizes distribution — is becoming a more discrete forced-spend category. Companies with proprietary rack-level power architecture are gaining specification advantages with hyperscalers that will translate to multi-year procurement commitments.
The BTT Framework Assessment
Across the Technology pool, our BTT framework assessment of the current portfolio is positive: the supply chain constraints we identified remain intact, the forced-spend dynamics continue to generate revenue and margin for our positions, and the competitive moats protecting the supply chain positions we hold have not materially eroded.
The second half of 2025 will be characterized by continued hyperscaler capex commitment — the investment programs announced in Q4 2024 and Q1 2025 are in their active procurement and construction phases. The supply chain positions we hold are in the middle of their forced-spend periods. We expect this to continue.
