Artificial intelligence is one of the most important investment themes in global markets. It is also one of the easiest to misuse.
The technology is real. The potential productivity gains are real. The ability of AI to change how software is built, how customers are served, how factories operate, how risk is priced, and how content is created is significant. But a real technology is not automatically a good investment. Investor expectations can move faster than business economics. Valuations can discount the future before it arrives.
Our job is not to chase the loudest AI story. It is to identify where AI can create measurable economic value and to insist that the price leaves room for imperfect outcomes.
A 3-part framework
We think about AI and scout for opportunities across three connected layers: infrastructure, platforms, and adoption.
Infrastructure: scarcity matters
At the foundation of AI is physical capacity. Leading-edge semiconductors, high-bandwidth memory, advanced packaging, data centers, power, cooling, and networks are all required to train and run models. These are not frictionless digital assets. They require capital, technical capability, supply chains, permits, and time.
That matters because scarcity can shape returns. When demand rises faster than supply can respond, the companies controlling constrained parts of the stack can earn attractive returns. But infrastructure is also cyclical. Memory, foundries, and hardware supply chains can shift from shortage to surplus. A good AI infrastructure investment still requires normal-cycle earnings analysis, balance-sheet discipline, and a clear view of what could happen when capacity catches up.
Platforms: distribution advantage
The second layer is where AI becomes usable at scale. Cloud platforms, software ecosystems, commerce networks, payment platforms, and developer tools can turn model capabilities into products that businesses and consumers actually use.
Distribution matters because enterprises rarely adopt new technology in a vacuum. They adopt through tools they already trust, integrate, and budget for. A company already embedded in enterprise workflows or large digital ecosystems can commercialize AI more efficiently than a standalone model provider with limited distribution.
This does not mean every large platform will win. The question is whether AI improves the product, increases customer value, reduces churn, expands monetization, or lowers the cost to serve. We look for evidence in usage, pricing, margins, product velocity, and customer behavior, not just in management commentary.
Adoption: quiet efficiency gains
The third layer may be the least glamorous but the most durable. Many AI benefits will appear as small improvements within existing businesses. Better recommendations. Better fraud detection. Better credit decisioning. Better customer support. Better logistics routing. Better software productivity. Better marketing efficiency.
These gains may not look dramatic in isolation. Over time, they can matter. A business that grows while support costs rise more slowly can expand its margins. A lender that prices risk more effectively can grow faster with lower losses. A marketplace that improves search and matching can increase conversion. A payments network that detects fraud earlier can improve trust and reduce leakage.
The investment discipline
AI can tempt us to suspend valuation discipline. We resist that. A company can be important to the AI ecosystem and still be a poor investment if the valuation already assumes flawless execution. Conversely, a company may benefit from AI without branding itself as an AI company.
We ask a few basic questions. Where does the economic value accrue? Is it captured by the model provider, the chip supplier, the cloud platform, the application layer, or the customer? Which part of the stack is scarce? Which is likely to commoditize? Does AI strengthen the moat or merely raise the cost of staying competitive? Are we underwriting normalized economics or peak enthusiasm?
Why a global lens helps
A global lens is useful because AI is not only a U.S. mega-cap story. The infrastructure layer spans Taiwan, Korea, Japan, the United States, Europe, and China. The platform layer includes Western cloud and software companies, Chinese digital ecosystems, and regional fintech and commerce platforms. The adoption layer reaches banks, retailers, telecom operators, manufacturers, and marketplaces across both emerging and developed markets.
This broader view helps us avoid relying on a single narrative. AI may create value in many places, but not equally, and not always where market attention is highest.
The bottom line
AI is powerful, but power alone is not enough. We want to own businesses where AI tightens a real constraint, improves unit economics, strengthens distribution, or deepens customer value. We then want to buy those businesses at valuations that still demand discipline.
The point is not to find the loudest AI story. It is to find the durable economics beneath the headline.