Innovation's Second-Order Winners: Where AI Economics May Accrue Beyond the Obvious
The largest investment opportunities in a technology cycle are not always located in the most visible product. Bottlenecks, complements, and enabling infrastructure can capture a disproportionate share of economics.
By Manish Sharma, CFA · Founder & Managing Principal
Central Thesis
“The AI cycle should be analyzed as an industrial system involving compute, memory, networking, power, cooling, data, security, and workflow redesign.”
Key Takeaways
- AI demand is constrained by physical infrastructure as much as by software capability.
- Second-order beneficiaries often have clearer monetization but may also face cyclical overbuild risk.
- The durable winners will convert bottleneck status into returns on invested capital rather than temporary revenue acceleration.
From Product Narrative to System Economics
Technology cycles are initially described through the most visible product. The internet was discussed through portals and browsers; mobile through handsets and applications. Yet large pools of value also accrued to semiconductors, networks, payment systems, logistics, and enterprise software.
Artificial intelligence is following a similar pattern. The model is only one layer. The economic system includes data preparation, accelerators, memory, networking, power generation, transmission, cooling, cybersecurity, and the redesign of business processes.
Bottlenecks Create Pricing Power—Temporarily
Scarcity can create exceptional margins, but scarcity also attracts capital. Investors must distinguish a structural bottleneck from a cyclical shortage. The relevant questions are how quickly supply can respond, whether customers can redesign around the constraint, and whether the provider controls a proprietary standard or simply owns current capacity.
The strongest businesses combine a bottleneck with switching costs, ecosystem control, or recurring service revenue. Capacity alone is less durable.
Power and Cooling Become Computing Inputs
Data-center economics increasingly depend on power availability, grid interconnection, thermal management, and site selection. This shifts part of the AI opportunity toward utilities, electrical equipment, power semiconductors, backup systems, engineering firms, and specialized cooling technologies.
These areas may offer long backlogs and visible demand, but investors should examine contract quality, customer concentration, working-capital needs, and whether capacity expansion can erode incremental returns.
Cybersecurity and Governance as Mandatory Complements
More automated workflows expand the attack surface and increase the value of identity, data governance, model monitoring, and secure access. Security spending is therefore not merely a defensive afterthought; it is part of the adoption architecture.
The beneficiaries are likely to be platforms that reduce operational complexity rather than add another isolated tool. Consolidation, integration, and measurable risk reduction matter more than feature count.
Underwriting the Cycle
A thematic allocation should not be a basket of narratives. It should be a map of value pools, capital intensity, competitive advantage, and adoption risk. The investment committee should track order growth, backlog conversion, customer concentration, free-cash-flow quality, and the pace at which supply is entering each layer.
The goal is to participate in a structural cycle while maintaining the discipline to recognize when a structural thesis has been overcapitalized.
