Daniel A. Keller, CEO and Cofounder of InFlux Technologies Limited and Flux.gettyEvery generation of enterprise technology arrives with a liberation narrative. The PC freed us from the mainframe. The cloud freed us from on-premise chaos. And now, AI agents are freeing us from bloated SaaS subscriptions.​But here is the question nobody is asking loudly enough: freed into what, exactly?A Trip To The PastLet’s start with the fundamentals. Computing has always been a pendulum. It swings between centralized and distributed, between locked and open and then, almost inevitably, swings back.​The mainframe gave enormous power to whoever owned the iron. The client-server revolution of the 1980s looked like the answer: distribute processing, break the monopoly of centralized IT. And it worked, at least briefly. But then it became unmanageable, and the cloud arrived as the fix. Your application, someone else's servers. Plus, it had elastic capacity, and you get to "pay as you go."However, we handed over the gate-keys to three companies. As of Q1 2026, AWS, Microsoft Azure and Google Cloud collectively command more than 60% of global cloud infrastructure spending. The rest of the market is, in the words of Synergy Research Group, "stuck in the low single digits." More like we traded IBM's dominance for Amazon's, Microsoft's and Google's, calling it progress.Arrival Of The SaasPocalypseIn roughly 48 hours in February 2026, over $285 billion in SaaS market capitalization evaporated. The reason? The emergence of AI agents demonstrated they could perform the knowledge work that an entire generation of per-seat software had been built to support. By mid-April, the total damage exceeded $2 trillion. The market had decided the per-seat model was over.​The narrative being written is that AI agents are the new liberators, delivering SaaS flexibility and freedom from subscriptions. But the infrastructure those agents run on? The compute powering workflows that replace a thousand enterprise tools? That is flowing right back to the same three landlords.We traded one master for another.Repeating The Same Mistake?Every distributed technology eventually centralizes because centralization is, in the short term, easier, to procure, to support, to explain to a board. When AWS arrived, it offered something genuinely revolutionary: infrastructure without the infrastructure team. Enterprises accepted the trade-off, sovereignty for convenience, without fully accounting for what they were giving up.​A decade later, 84% of enterprises cite managing cloud spend as their top cloud challenge. Vendor lock-in is no longer theoretical; it is a line item. Egress fees, proprietary APIs and the quiet leverage that accumulates when a provider becomes load-bearing are the hidden costs of the convenience bargain. The cloud did not liberate enterprises from dependency. It made the dependency more comfortable.​Now we are about to make the same trade with AI infrastructure. Microsoft, Meta and others have committed over $300 billion in capital expenditure to AI infrastructure in 2026 alone. They are not investing at that scale out of generosity. They are building the next generation of lock-in before the market has had time to ask whether there is another way.​What A Different Future Looks LikeI am an optimist about this moment. Not because the problem will solve itself, but because for the first time, the tools to build something different are mature enough to use.​Decentralized physical infrastructure networks (DePIN) are coordinating real-world GPU clusters, storage and bandwidth across distributed operators, and the sector is projected to grow from $50 billion in market cap today to $3.5 trillion by 2028. This mirrors early cloud adoption curves, but with fundamentally different ownership economics.​The significance is not just on the financial front. It expands to the architecture. As AI systems become more economically vital, the centralized infrastructure they rely on introduces a growing risk of single-point failure and control. When the infrastructure powering autonomous agents is owned by the same handful of companies those agents interact with, you have not built a new economy. You have built a more automated version of the old one.​Decentralized cloud offers more than cost efficiency, though 45% to 60% savings on GPU compute versus hyperscaler pricing is meaningful. It offers something more durable: infrastructure that no single entity can reprice, restrict or withdraw. For AI agents acting as the operating layer of enterprise, initiating transactions, managing workflows and coordinating across systems, the ownership of the underlying infrastructure becomes more of a governance question.The Choice In Front Of UsThe SaaSpocalypse has created a rare moment of openness. Enterprise CIOs are actively reallocating budgets, renegotiating contracts and reconsidering assumptions held for two decades. The old stack is being dismantled. What replaces it is not yet decided, but that window is closing.The hyperscalers are moving fast, with enormous advantages: existing relationships, mature tooling, deep capital reserves and operational consistency at global scale. For many enterprises, centralized AI infrastructure will remain the most practical choice in the near term, especially where reliability, compliance guarantees, ecosystem integration and enterprise support are nonnegotiable.At the same time, decentralized infrastructure is no longer theoretical. The DePIN ecosystem has expanded rapidly, demonstrating that globally distributed compute networks can operate at meaningful scale and increasingly competitive economics. But the model is still early in several critical areas. Standardization remains uneven. Developer experience is fragmented. Enterprise procurement and compliance pathways are still maturing. Questions around governance, accountability, service guarantees and long-term operational coordination have not been fully resolved.This is not a contest between good and bad systems. It is a question of trade-offs.Centralized infrastructure offers cohesion, predictability and speed of execution. Decentralized infrastructure offers resilience, distribution of ownership and potentially lower structural dependence over time. Both models are evolving under the pressure of AI's exponential demand for compute.ConclusionThe question for technology leaders is not simply which architecture is philosophically preferable, but which one proves capable of sustaining the next era of software: economically, operationally and politically.Ultimately, time and the performance of these systems under real-world pressure will decide which model endures.Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?