Why Time to Power Has Become One of the Most Critical Success Metrics for Neoclouds and AI Factories

This Industry Viewpoint was authored by Mike Tapp, Head of Finance at LiquidStack

Not a week goes by without news of yet another super-massive data center being planned, proposed, or built. The headlines are filled with eye-watering figures around size, cost, capacity, and aggressive timelines. But while big numbers are a good way to attract eyeballs, one of the figures that matters most to today’s data center operators is rarely reflected in today’s headlines.

With the arrival of AI and the token economy, data centers, which have traditionally been seen primarily as cost centers, are now increasingly treated as revenue generators, especially for AI workloads. And many operators’ priorities have been transformed in kind. When looking at neoclouds and AI factories, the ability to bring high-density compute capacity online quickly is now becoming one of the key determinants of market leadership.

That figure, otherwise known as “time to power” (TTP), is changing the way the data center industry thinks about development, and how the players within it compete. To better understand why that is happening, we have to look at the metric itself and what it means in today’s evolving data economy.

Understanding Time to Power: The Race to Revenue

In simplest terms, time to power encompasses the full timeline from the decision to build a data center to bringing new compute capacity online. This includes everything from site acquisition and permitting to grid interconnection, electrical infrastructure deployment, cooling systems, and final commissioning. And each one of these stages is rife with opportunities for roadblocks, setbacks, and delays. To navigate these challenges effectively, each stage often requires specialized expertise.

Historically, data center development timelines could easily stretch three to five years from concept to operation. In the AI era, however, that timeline is increasingly incompatible with the pace of technological and economic development.

The world’s appetite for AI compute has become enormous. NVIDIA’s latest GPUs can draw hundreds of watts each, and modern AI clusters often contain tens of thousands of GPUs. Entire campuses are now being designed to support hundreds of megawatts of capacity, with some projects targeting gigawatt-scale deployments.

With that level of power demand, a new bottleneck is introduced: an increasingly constrained energy grid. In many regions, utilities are struggling to keep pace with the surge in demand from AI infrastructure. In many of these areas, interconnection queues for large power loads can stretch several years as utilities scramble to assess transmission capacity, upgrade substations, and coordinate generation resources.

For infrastructure developers, this means the timeline to energize a facility is increasingly dictated not by construction schedules, but by power availability.

Understanding Neoclouds’ Need for Speed

The rise of neocloud providers has made this challenge even more pronounced. Unlike traditional hyperscalers that built their infrastructure over decades, neoclouds are purpose-built to quickly deliver AI compute capacity. Many specialize in offering GPU infrastructure as a service for startups, enterprises, and AI research organizations that cannot access hyperscale infrastructure.

Similarly, the concept of the “AI factory”—a data center optimized specifically for handling large-scale AI workloads—is quickly gaining traction as a reliable revenue generator. These facilities are designed around GPU clusters and high-density power infrastructure rather than traditional enterprise workloads.

In both cases, speed is a primary competitive advantage. When GPUs can cost tens of thousands of dollars per unit, depending on configuration, and demand for compute continues to surge, every month of delay represents a lost revenue opportunity. Operators that can deploy new capacity faster are likely to capture workloads that might otherwise go elsewhere. As a result, time to power is emerging as one of the most critical competitive advantages for a large swath of modern data center operators.

The Infrastructure Bottlenecks Slowing Deployment

Bringing AI infrastructure online often involves several additional bottlenecks that revolve around infrastructure, construction, and architecting.

Electrical infrastructure has become one of the biggest of those challenges. Key components such as transformers, switchgear, and high-voltage equipment now frequently face extended procurement timelines due to global demand. In some cases, lead times for major electrical components can extend to 18-24 months or more in constrained markets.

Construction timelines also remain unpredictable. Permitting processes vary widely by region, and labor shortages in some markets can slow large-scale builds.

But perhaps the most significant shift is happening inside the data center itself. AI clusters are dramatically increasing rack power density. Most traditional enterprise data centers were typically designed for racks drawing 5–10 kilowatts of power. Modern hyperscale facilities have pushed that number closer to 20–30 kW per rack, on average.

Now, AI workloads are taking things to another level entirely. High-performance GPU racks now commonly exceed 80–120 kW, and next-generation designs may go even higher. These densities are pushing conventional air cooling approaches toward their practical limits in some high-density environments, often requiring hybrid or alternative cooling strategies.

Cooling as a Strategic Enabler

As rack densities increase, cooling infrastructure becomes a critical factor in how quickly new compute can be deployed. Air cooling architectures remain highly effective for many workloads, but at higher densities they can require larger mechanical systems, more advanced airflow management, and increasingly complex facility designs as total heat loads rise. In some cases, this can introduce practical limits on how densely compute infrastructure can be deployed within a given environment.

Liquid cooling technologies are increasingly proving themselves to be a formidable solution to these challenges. By transferring heat far more efficiently than air, liquid cooling can enable much higher rack densities while simplifying thermal management inside the data center. In most cases, liquid-based approaches can also support greater compute density within the same power footprint, often while reducing mechanical infrastructure requirements.

As one might imagine, this has major implications for deployment speed. Higher-density cooling solutions make it possible for operators to maximize the value of available power capacity and deploy AI infrastructure more rapidly within existing facilities. In an environment where grid access is constrained, the ability to better utilize available power capacity becomes invaluable. As a result, we’re seeing liquid cooling rapidly gain traction as the AI race ramps up.

How Infrastructure Strategy Is Changing to Speed Up Time-to-Power

As time to power becomes one of today’s defining constraints, data center development strategies are changing in response. Developers are increasingly prioritizing power-first site selection, targeting regions where utilities can deliver capacity quickly rather than simply focusing on traditional connectivity hubs.

Modular and prefabricated infrastructure approaches are also gaining popularity as a way to accelerate deployment timelines and reduce construction complexity.

High-density architectures are another focus for operators looking to generate more compute more quickly. In simplest terms, by designing facilities to support higher rack power levels from the outset, operators can maximize compute output without requiring additional grid capacity. Taken together, these shifts reflect a broader reality: the AI infrastructure race is increasingly about how quickly capacity can be transformed into productivity, not just how much capacity can ultimately be built.

Time-to-Power: The Defining Constraint of the AI Era

The rapid rise of AI is transforming nearly every aspect of the digital infrastructure landscape. But while advances in GPUs and model architectures often capture the headlines, the real constraints are increasingly physical. Electric power, grid capacity, and infrastructure deployment timelines are becoming the gating factors for AI expansion. In this environment, time to power is emerging as one of the most important metrics for neoclouds and AI factories alike.

The organizations that can secure power, deploy infrastructure efficiently, and bring new capacity online quickly will be best positioned to serve the exploding demand for AI compute. Because in the new AI economy, it’s about more than just how many GPUs you can buy, it’s also about how quickly you can turn them on.

If you haven't already, please take our Reader Survey! Just 3 questions to help us better understand who is reading Telecom Ramblings so we can serve you better!