The money pouring into AI infrastructure is about to hit a wall — and it's made of concrete, copper, and power lines.
The Summary
- Parnassus Investments CIO Todd Ahlsten warns the AI buildout is entering a "riskier phase" as infrastructure spending collides with physical constraints
- Traditional seat-based software companies like Salesforce, Workday, and ServiceNow face economic pressure as AI agents replace human users
- Real opportunities are shifting beyond chips and data centers to the physical infrastructure that makes the agent economy possible
The Signal
The AI infrastructure boom is transitioning from a software story to a hardware problem. For two years, investors poured money into chips and cloud capacity. Now the constraint isn't compute, it's the unglamorous stuff: power grids, cooling systems, real estate for data centers, and the supply chains that feed them. Ahlsten points to opportunities in this physical layer as the next phase of AI investment.
This matters because it signals a shift in where the money flows and where the bottlenecks appear. The easy part was writing checks to Nvidia. The hard part is building out the electrical infrastructure to power inference at scale, securing enough water for cooling, and navigating local permitting for facilities that look like industrial plants.
"The AI boom is colliding with the limits of the physical world."
The more interesting signal is what Ahlsten says about traditional software companies. Salesforce, Workday, ServiceNow — these companies built empires on per-seat pricing. Every employee needed a license. But AI agents don't need seats. They don't need logins or dashboards or user training. They need API access and compute credits.
When one AI agent can handle the work of ten customer service reps or five sales development reps, the economics of enterprise software break. The revenue model collapses. You can't charge per agent the way you charged per human because the whole point of the agent is to do more with less. Software companies are going to have to rebuild their pricing models from scratch, and not all of them will survive the transition.
Key dynamics reshaping enterprise software:
- AI agents eliminate the "seat" as the unit of value in enterprise SaaS
- Companies will pay for outcomes or API calls, not user licenses
- Software vendors lose pricing power as automation replaces headcount
This is the Web4 moment playing out in real time. The companies that built Web2's read-write infrastructure are watching their business models get rewritten by agents that don't fit the old frames. Meanwhile, the capital requirements for building out Web4 infrastructure — the physical layer that makes agent economies possible — are massive and risky. Not every bet pays off when you're dealing with concrete and copper instead of code.
The Implication
If you're an investor, watch the companies building physical AI infrastructure: power management, cooling tech, data center real estate, fiber networks. These are the picks and shovels plays for the next phase. If you're working at a traditional SaaS company, start asking hard questions about how your product monetizes in a world where agents, not humans, are the primary users.
The AI buildout isn't slowing down. It's just getting harder, more expensive, and more capital-intensive. The winners in this phase won't be the ones with the best demos. They'll be the ones who can solve logistics, power, and physics.