Point of View
Artificial Intelligence · Technology Strategy
The Value Is Migrating Inside the AI Stack Faster Than Strategy Is Tracking It
AI as a technology will commoditize, the same way connectivity, cloud infrastructure, and every prior platform technology did. The durable value goes to companies applying it to specific, complex industry problems -- not the ones building the infrastructure underneath it.
Feb 2024
Points of View is Vaxa's opinion column. Each article states a position on a question where reasonable people disagree — and makes the argument for that position directly.
VAXA POINT OF VIEW
AI as a technology will commoditize. The durable value will be captured by companies that apply it to specific, complex problems in specific industries — not by the companies building the underlying infrastructure. This has happened before. It is happening again.
Pattern
The internet commoditized connectivity. In the early years, the value was in the infrastructure — the routers, the cable networks, the ISPs. The companies that built and owned the pipes were assumed to be the structural winners. Within a decade, connectivity became abundant and cheap, and the value migrated almost entirely to applications built on top of it. Amazon, Google, Salesforce — none of them own the infrastructure. They own the applications that run on it.
Cloud computing commoditized enterprise computing infrastructure. In the early years, the value was assumed to be in the data centers, the servers, the storage. The hyperscalers built that infrastructure and captured enormous value doing so. But the durable value migrated to the software layer — the SaaS applications, the data platforms, the workflow tools that companies actually pay for because they cannot run their businesses without them. The infrastructure became a utility. The applications became essential.
AI is following the same path. The current phase — enormous capital flowing into models, chips, data centers, and the infrastructure layer — resembles the early internet and early cloud phases closely. The capital is real. The infrastructure is being built. And the value will migrate, on the same timeline and for the same reasons it always has.
Commoditizes
The internet commoditized connectivity. In the early years, the value was in the infrastructure — the routers, the cable networks, the ISPs. The companies that built and owned the pipes were assumed to be the structural winners. Within a decade, connectivity became abundant and cheap, and the value migrated almost entirely to applications built on top of it. Amazon, Google, Salesforce — none of them own the infrastructure. They own the applications that run on it.
Cloud computing commoditized enterprise computing infrastructure. In the early years, the value was assumed to be in the data centers, the servers, the storage. The hyperscalers built that infrastructure and captured enormous value doing so. But the durable value migrated to the software layer — the SaaS applications, the data platforms, the workflow tools that companies actually pay for because they cannot run their businesses without them. The infrastructure became a utility. The applications became essential.
AI is following the same path. The current phase — enormous capital flowing into models, chips, data centers, and the infrastructure layer — resembles the early internet and early cloud phases closely. The capital is real. The infrastructure is being built. And the value will migrate, on the same timeline and for the same reasons it always has.
Infrastructure enables. Applications capture. This is not a prediction about AI — it is a description of how platform technology value chains work. AI is not different enough to break the pattern.
Compounds
The companies that will capture durable value from AI are those that combine two things that are genuinely difficult to replicate: deep domain expertise in a specific industry or problem type, and the ability to deploy AI effectively within that domain. Neither alone is sufficient. AI without domain expertise produces generic capability that is rapidly commoditized. Domain expertise without AI produces capability that is outpaced by competitors who are using AI to do the same thing faster and cheaper.
The systems integrator model is the clearest illustration of this. A company like Palantir is not primarily an AI company in the sense that the infrastructure providers are. It is a systems integrator whose value is the ability to deploy AI-powered analytics into specific, complex, high-stakes operational environments — defense, intelligence, industrial operations, healthcare — where the data is messy, the problems are domain-specific, and the deployment requires deep understanding of how the organization actually works. AI makes that capability more powerful. It does not replace it.
The thesis that AI-assisted software development will erode the value of companies like this misunderstands what the value actually is. AI making software development easier reduces the cost of commodity software development. It does not reduce the value of domain expertise, operational deployment capability, and the hard-won knowledge of how AI actually performs in specific real-world contexts. If anything, it accelerates the value of those who have it — because the barrier to building generic AI capability falls, and the remaining differentiator is the domain knowledge that AI cannot generate on its own.
Implication
For most organizations, the AI strategy question being asked — which foundation model to use, which cloud provider to run on, how much compute to procure — is the wrong question. These are infrastructure decisions. They matter for cost and capability, but they do not determine competitive advantage. Infrastructure access is becoming a commodity. The organizations that treat AI infrastructure decisions as their primary AI strategy are optimizing the wrong variable.
The right question is which specific problems in your industry AI can solve better than anything available before — and whether you are building the domain expertise, the data assets, and the deployment capability to capture that value. That is an application question, not an infrastructure question. It requires understanding your industry deeply enough to identify where AI creates genuine competitive advantage, not where it creates impressive demonstrations.
The companies that will look back at this period as a strategic inflection point are not the ones that made the best infrastructure decisions. They are the ones that identified the highest-value applications in their specific domains, built the expertise to deploy AI in those contexts effectively, and captured the compounding advantage that comes from being first in a domain where the learning curves are steep and the switching costs are high.
The Question to Ask
Is your AI strategy built around infrastructure access or application depth? The former is available to everyone with a budget. The latter requires something that cannot be procured — domain expertise, operational knowledge, and the ability to deploy AI where the problems are genuinely hard. That is where the durable value will be captured. The question is whether you are building it.