Twelve Months of Signal and Noise
The AI infrastructure investment theme has now been the dominant narrative in technology equity markets for more than a year. In that time, an enormous amount of capital has been deployed — both into the actual buildout of AI compute infrastructure and into the equity securities of companies tangentially related to that buildout. This combination of real capital cycle and enthusiastic market narrative creates the conditions where distinguishing signal from noise becomes the most important analytical task.
GZC's Bottleneck-to-Ticker framework is explicitly designed for this distinction. The question is not whether AI infrastructure investment is a real theme — it clearly is — but which specific supply chain positions represent durable forced-spend dynamics versus which are narrative-driven investments with more precarious demand foundations.
What Has Been Verified
A useful exercise, twelve months into any major capital cycle, is to list the claims that have been validated by actual procurement behavior versus those that remain largely narrative.
Verified: Hyperscaler capital expenditure is real and accelerating. The four major hyperscalers — Microsoft, Google, Amazon, and Meta — collectively announced capital expenditure budgets for AI infrastructure in 2024 that substantially exceeded analyst expectations at the start of the year. These are not press releases — they are capital allocation decisions reflected in financial statements, balance sheet drawdowns, and equipment procurement contracts. The money is being spent.
Verified: Power delivery is a genuine bottleneck. Transformer and switchgear lead times, which we identified as a critical supply chain constraint through our BTT analysis, have extended materially over the past 18 months. Data center developers are publicly discussing power delivery timelines as the primary constraint on deployment schedules. This is verified by procurement disclosures, utility interconnection queues, and the financial results of equipment manufacturers.
Verified: HBM supply is constrained. High-bandwidth memory — the memory architecture required for GPU-based AI accelerators — has been in tight supply. The three manufacturers capable of producing it at commercial scale cannot rapidly expand production, and demand from GPU manufacturers has consistently exceeded available supply. This is verified by memory pricing trends, GPU allocation dynamics, and manufacturer capacity announcements.
What Remains Narrative
Unverified: The AI revenue cycle will arrive on the timelines implied by current infrastructure investment. The physical infrastructure is being built. The business model applications that will generate the revenue to justify the investment are still emerging. This does not invalidate the infrastructure investment thesis — forced-spend dynamics do not require the end-market application to be profitable immediately — but it means that the application layer is more speculative than the infrastructure layer.
Unverified: Every announced data center project will be completed on schedule. A significant number of data center announcements, particularly from second-tier operators and speculative developers, have not been backed by committed capital, signed power agreements, or verified procurement of long-lead-time equipment. The gap between announced projects and completed ones will become apparent over the next two years.
The BTT Implication
For GZC's portfolio, the distinction between verified and narrative elements of the AI infrastructure cycle shapes position sizing. We are most heavily concentrated in positions where the supply chain constraint is verified — where procurement lead times, manufacturer backlog data, and customer commitment documentation confirm the forced-spend dynamic. These positions have the characteristics BTT requires: non-discretionary demand, limited supply, and multi-year procurement visibility.
We hold lighter positions in areas where the forced-spend thesis is logically sound but not yet confirmed by supply chain behavior. These positions can grow as the verification accumulates, or they can be eliminated if the verification does not arrive.
Twelve Months Forward
The next twelve months will determine whether the AI infrastructure capital cycle is multi-year and durable or concentrated in a single investment cycle. The indicators we watch are: hyperscaler capex guidance in quarterly earnings, the pace of grid interconnection approvals for large data center loads, and the behavior of procurement lead times across the power delivery supply chain.
Our current view is that the forced-spend dynamics we identified through BTT analysis are real, multi-year, and still in early innings. The infrastructure required to support the AI compute workloads that have been committed to does not yet exist at the required scale. Building it is not optional — and every non-optional buildout cycle eventually produces the companies that BTT is designed to find.
