1. The Compute Cycle Is Nearing Its End
Over the past two years, the hottest part of AI investing has been compute—GPUs, data centers, electricity. This is the classic "infrastructure phase," characterized by profits concentrated among a few suppliers.
But the excess returns from infrastructure typically last 2-4 years, after which supply catches up quickly and margins decline. The signals are already clear: prices are softening, and customers' bargaining power is returning.
2. The Platform Phase Is Arriving
The platform phase is defined by a few companies that package underlying capabilities into APIs/SDKs, allowing application developers to build products without training their own models. The moats here are network effects and data accumulation.
Winners in this phase are usually fewer but more durable than those in the infrastructure phase. By analogy to the last internet cycle: infrastructure winners were Cisco and Intel; platform winners were Google, Amazon, and Facebook—a huge difference.
3. The Application Layer Is About to Explode
The application layer is the most imaginative and also the most fragmented. Vertical scenarios, niche demographics, and new interaction forms will all emerge here.
The biggest challenge in investing in applications is that the winners are often companies that don't exist yet. This forces investors to either spread bets or wait until the picture clears—both strategies are valid, but you need to know which one you are.
