Data Centers are Not the Future, but They are Still in Our Future

When you type something into ChatGPT, the response doesn't come from thin air. It comes from a data center, a massive facility packed with specialized computing hardware, purpose-built to run AI. These buildings cost billions of dollars, take years to permit and construct, consume enormous amounts of power, and require cooling systems most communities aren't equipped to support. The biggest technology companies in the world are racing to build more of them, and they still can’t build them fast enough.

This model was manageable when all AI did for people was occasionally answer questions. It’s become an entirely different problem now that AI has started doing work continuously on our behalf.

I’m talking about agents. An AI agent isn't a chatbot that responds to questions. It's a system that takes actions. It can research a prospect, write a follow-up email, update a calendar, create a task, and hand the result off to another agent handling the next step, all running in the background without someone at a keyboard managing each step. Building one used to take months of software development, now a capable agent can be deployed in days, and they are multiplying fast. Also, agents use a lot of compute, not just to do the task they’re assigned, they use it to intake the assignment, figure out how to do the task, make tool calls, create and display outputs, all of it.

Software scales overnight, physical infrastructure doesn't. When millions of businesses shift from occasionally asking AI a question to running agents around the clock, the demand for compute doesn't go up incrementally, it goes up by an order of magnitude. The data centers under construction right now were designed for the AI use case of yesterday. And they won’t be enough for the way it’s going to be used tomorrow.

Something has to give. I believe we’re going to move towards decentralization. On-premise servers for businesses that need control and reliability. Home machines that host personal models, trained on your actual habits and interests, running locally without the cloud. Sure, they’ll need internet access for a lot of what they do, but they’ll live where they’re used.

The frontier model race will keep going, but it’ll narrow. Scientific research, large enterprise, and government will still be customers, but for everyone else the interesting development is efficiency. Smaller, sharper models built for specific purposes, designed to run on hardware that already exists in homes and offices. Better for power consumption, better for cost, and better for not needing the kind of liquid cooling infrastructure most people don't have.

Gaming, creative work, home automation, office productivity. The models coming for those spaces won't need a data center, they'll need a corner of your office. We’re already seeing a shift towards this in the AI space. Apple Intelligence and Microsoft Phi are examples already in operation. Other companies and labs like Standford’s Open Jarvis, Meta’s Llama and DeepSeek’s next big thing are on the way. Hardware manufacturers like Nvidia and Intel have announced they’re designing chips with this in mind. 

I believe we’re months away from this trend becoming obvious and less than a year away from it being the new normal.

Have questions about what this means for your business? I'd love to talk it through. Book a time on my Calendly, or reach me directly at sam@bright-pivot.com.