The AI Application Layer Is Becoming Investable Again
From Agent Traffic to AI ROI: Four Software Themes Investors Should Watch
For the past two years, the equity market has treated AI mostly as an infrastructure story. The obvious winners were GPUs, data centers, networking, power, cloud capacity, and companies closest to model training and inference. That made sense. Before AI could transform workflows, someone had to build the railroads.
But the next phase may be less about the railroads and more about traffic control.
The key signal from the latest software earnings cycle is not that every company now has an AI product. That is table stakes. The more important signal is that AI is creating new operating problems: internet traffic is becoming machine-led, enterprise identities are multiplying, token costs are harder to justify, and customers are asking whether AI projects are producing measurable outcomes.
The first shift is the move from a human-led web to an agent-led web. In the old internet, a human searching for a product might visit five websites. In an agentic internet, an AI assistant can visit thousands of pages, call APIs, compare databases, evaluate vendors, and return one synthesized recommendation. The user still wants the same thing — a camera, a hotel, a software subscription — but the path to that outcome becomes far more computationally intensive.
This matters because AI agents do not only create demand. They redistribute economics.
A thousand extra web visits may not create a thousand times more consumer value. The consumer still buys one camera. But the process now consumes more cloud resources, more model tokens, more database calls, more security checks, and more identity verification. Some companies will absorb that cost. Others will monetize the new control points.
That gives us the first theme: the rise of the agent-facing internet.
Cloudflare is one of the clearest application-layer names here. The company sits near the edge of internet traffic, where it can help identify, filter, secure, and govern machine-driven requests. More traffic does not automatically flow through its income statement dollar-for-dollar. But as bot traffic, agent access, API calls, and authentication demands rise, the need for more sophisticated security and traffic-management products should increase.
The second-order question is how businesses get discovered by agents. The old web was optimized for human search. The agentic web may require a new version of SEO — generative engine optimization, or making businesses legible to AI agents. Google and Meta are natural candidates to watch because they already monetize the bridge between consumer intent and commercial supply.
The second theme is enterprise agent governance.
If companies eventually operate with millions of software agents alongside human
