Wall Street loves a good panic. Whenever tech stocks wobble or a few founders suggest pumping the brakes on frontier model training, the narrative shifts overnight. Suddenly, everyone acts like multi-billion-dollar data centers are going to turn into expensive, empty warehouses.
Digital Realty CEO Andrew Power recently pushed back against that exact panic. His argument is simple. Even if the frantic pace of training massive new artificial intelligence models slows down, the underlying demand for real estate, power, and physical infrastructure isn't going anywhere.
Let's look at why the doom-and-gloom forecasts miss the mark, and why physical data center assets remain an entirely different beast compared to software stocks.
The Shift From Training to Inference
Most people hear "artificial intelligence" and picture a lab training a massive model from scratch. That training phase takes an astronomical amount of compute power. It is also cyclical. Companies invest billions, test limits, and occasionally pause to evaluate efficiency gains.
However, running those models day-to-day—known as inference—is an entirely different engine.
When enterprises actually integrate these tools into daily operations, customer service, and internal data analysis, the compute requirements shift from a speculative sprint to a permanent utility. Right now, enterprise adoption is moving from strategy to execution. Companies aren't just experimenting anymore; they are wiring these capabilities into their core stacks.
If frontier model training hits a temporary bottleneck or a regulatory speed bump, the transition toward running real-time workloads for everyday business users accelerates. That requires local, secure physical infrastructure.
Supply Constraints Are Real
You cannot spin up a modern enterprise-grade data center over a weekend. Power availability, liquid cooling integration, local zoning, and fiber connectivity create massive barriers to entry.
Major hubs like Northern Virginia, Dallas, Chicago, Tokyo, and Frankfurt have faced tight supply for years. Digital Realty's multi-billion-dollar construction backlog and high pre-lease rates show that customers are locking down capacity years in advance.
When land, power, and permits are scarce, a slight cool-down in venture-backed model training doesn't suddenly flood the market with cheap, high-density space. Tenants are competing for scarce physical locations that cannot easily be substituted. If one project stalls due to local permitting hurdles or power constraints, the remaining inventory gets even tighter.
Traditional Digital Transformation Never Stopped
It is easy to forget that the rest of the enterprise technology world kept spinning while AI grabbed every headline. Cloud migration, streaming media, financial transactions, global supply chain tracking, and healthcare data security require continuous digital expansion.
Digital Realty's business model relies on a diverse ecosystem. Traditional cloud workloads, hybrid IT deployments, and standard enterprise colocation provide a massive safety net. Even if AI-specific growth moderated from its hyper-aggressive peak, the baseline expansion of global data consumption guarantees steady occupancy rates.
What Investors and Builders Should Watch
If you are trying to read the real estate market without getting blinded by tech hype, ignore the daily stock ticker fluctuations and watch these three indicators instead:
- Power availability and utility lead times: The real bottleneck isn't silicon; it is electrical grid capacity. Track how fast utility providers are approving substation connections.
- Pre-leasing velocity: Look at long-term commitments for space under construction. If major tenants keep signing multi-year leases before ground is even broken, demand remains locked in.
- Sovereign AI mandates: Governments worldwide want their own domestic compute infrastructure. That regulatory push creates a completely separate layer of demand that is insulated from commercial market swings.
The data center sector isn't immune to economic gravity, but treating it like a pure software bubble ignores the concrete, steel, and power keeping the modern economy online.
