Q3 2025 Review
The third quarter gave us one of the cleaner validations of our Bottleneck-to-Ticker process that we have seen since the firm's founding. Two thesis areas — HBM memory supply constraints and liquid cooling infrastructure — moved from GZC's internal research watch list to mainstream financial media coverage during the quarter. In both cases, we had been building positions and refining our analysis for months before the broader market recognized the investment opportunity. That is the BTT framework working as designed.
HBM Memory: The Constraint Finally Visible
We published our internal analysis on high-bandwidth memory supply dynamics in early 2025. The core thesis was straightforward: GPU compute scaling was progressing faster than HBM manufacturing capacity could follow. SK Hynix, Micron, and Samsung collectively control effectively all HBM production, and each of their expansion timelines was measured in years, not quarters.
In Q3, that constraint became impossible to ignore. Multiple AI system providers publicly acknowledged extended lead times on HBM allocations. One major hyperscaler disclosed in an earnings call that HBM supply was a gating factor on their AI compute roadmap. When a constraint that was previously discussed in supply chain forums starts appearing in Fortune 500 earnings disclosures, it confirms that the bottleneck is real and that the investment thesis has legs.
We maintained our position throughout the quarter. We did not add materially — the entry point had passed and we do not chase. What we did was monitor the thesis for signs of deterioration, found none, and held with conviction.
Liquid Cooling: The Next Bottleneck
If HBM was the story the market caught up to in Q3, liquid cooling was the story beginning to set up for Q4 and into 2026. As AI compute density increases — and it is increasing rapidly — air cooling hits its physical limits. The thermal output of modern GPU clusters running at scale cannot be managed with air. Direct liquid cooling (DLC) and immersion cooling solutions are moving from niche deployments to standard infrastructure requirements.
The constraint is not in the technology itself — liquid cooling works and the physics are well understood. The constraint is in qualified installation capacity, in the supply of specialized rack and manifold components, and in the integration expertise required to deploy these systems at the scale hyperscalers need. We identified three companies whose positioning in this transition creates the kind of durable, forced-spend dynamic that GZC's process is built to capture. We initiated positions in two of them during Q3.
BTT in Practice: How We Got Ahead of This
Our research process for both of these positions followed the same five-step sequence we use for every investment:
- We mapped the supply chain for AI compute infrastructure and identified the nodes with limited substitution options.
- We found the publicly traded companies that own or control those nodes.
- We validated pricing power, procurement visibility, and demand durability.
- We sized the positions to reflect our conviction and the quality of the entry point.
- We established monitoring criteria to track when the thesis might change.
Neither position was built on a hot take from a financial media article. Both were built from the bottom up — from teardown analysis, supply chain interviews, and capacity tracking across the semiconductor and infrastructure supply chain.
Commodities Pool: A Quiet Quarter
Pool Two was relatively quiet in Q3. We maintained our oil and gas positions through a period of price volatility driven by OPEC+ production disputes and demand uncertainty from China. Our commodities posture is fundamentals-first: we hold energy positions when the underlying supply-demand dynamics support them, and we are patient when near-term price action is noisy.
Gold and silver continued to perform as macro insurance assets, with gold in particular supported by continued central bank buying from emerging market reserve managers. We made no changes to our precious metals exposure during the quarter.
Looking Ahead
As we enter Q4, our focus is on three catalysts: AI capex guidance from hyperscalers in earnings season, HBM allocation announcements from memory manufacturers, and grid infrastructure commentary from utilities and data center developers. Each of these has the potential to either validate or challenge current positions — and we will be watching closely.
Thank you, as always, for your continued partnership.
— Ahijah Ireland, Founder & CIO, Green Zone Capital
