AI's energy problem has a solution that we’ve been ignoring

Artificial intelligence is often framed as an energy crisis in the making. As demand for AI explodes, hyperscalers and developers are racing to build massive new data centres to power the next generation of intelligence.

The prevailing assumption is that more AI requires more infrastructure, and more infrastructure requires more electricity.

But this narrative misses a critical reality. AI’s energy problem isn’t just about how much power we can deliver, it’s about how efficiently that power is used when it arrives. Imagine the GPUs and AI workloads as the digital brain and the electric grid that supplies energy as the circulatory system. The challenge is that the circulatory system simply can’t deliver enough oxygen (energy) to the brain fast enough.

According to the IEA, energy demand from data centres is expected to double or more within the next five to seven years. But building new transmission lines, substations, or generation capacity can take up to 10 years or longer. Some analysts predict 40% of AI data centres could face power constraints by 2027 because utilities simply won’t be able to expand capacity quickly enough.

Moreso, even when electricity reaches a facility, it isn’t always reaching the workloads that need it the most. Building more data centres can only go so far. To meet AI demand in the near term, operators must learn to work more intelligently within their power constraints. That’s where the overlooked concept of orchestrating grey space (mechanical and cooling systems) becomes critical.

Grey space: The overlooked bottleneck inside the data centre

Most conversations about AI infrastructure focus on the white space of the data centre - the racks of GPUs and servers performing the computational work. Between the incoming electrical supply and those computing systems lies another layer: the grey space.

Grey space is the infrastructure that converts raw electricity into usable computing power, including power distribution equipment, cooling systems, airflow management and other supporting systems. In effect, this layer determines whether megawatts entering the building actually translate into usable AI compute.

Using the same anatomy analogy, you can think of the grey space as playing a major metabolic role. Metabolism isn’t defined by how much food you eat but how efficiently your body converts that energy into function.

The same principle applies to data centres. Grey space plays the metabolic role of the facility. If power distribution, cooling and airflow are poorly coordinated with IT workloads, available power can become stranded. GPUs sit idle, compute performance drops and energy is wasted.

This dynamic is becoming increasingly visible as older data centres struggle to support modern AI hardware. Thousands of these facilities contain significant electrical capacity that cannot be fully utilised because the infrastructure connecting power delivery and computing workloads isn’t optimised for AI’s intensity.

In other words, vast amounts of usable capacity already exist but remain inaccessible because the white and grey spaces of the data centre tend to operate independently. Bridging them together is one of the most immediate opportunities to expand AI capacity without building new facilities.

Why orchestration beats expansion - at least for now

The timeline required to secure energy for a new data centre, commonly known to operators as “time to power,” has become the dominant growth constraint. Developers can plan new campuses but they cannot accelerate the physics and policy processes governing electrical infrastructure.

Thousands of legacy infrastructure sitting idle across North America hold an immediate solution. With the right retrofits and orchestration technologies, the existing data centres can support far more AI workload capacity than they do today and much faster than greenfield projects can be permitted, constructed and connected to the grid.

Grey and white space orchestration acts as a force multiplier within these environments. Instead of treating infrastructure and compute as separate systems, intelligent orchestration solutions coordinate them in real time, aligning power distribution, cooling performance and workload placement to maximise efficiency.

The world will still eventually need more data centres built and more energy sources to support AI growth long-term. For now, with time as the industry’s biggest constraint, retrofitting and intelligently managing existing infrastructure to increase usable capacity without adding a single watt of new electricity is a more effective approach than building new facilities from the ground up.

The next phase of AI infrastructure

AI demand is accelerating faster than the infrastructure can support. Long interconnection queues, constrained utility capacity and rising construction timelines are limiting near-term growth. Simply adding more power will not solve the problem.

Without better coordination inside the data centre, additional megawatts can translate into inefficiency or underutilised GPUs. Without a healthy 'metabolic' or grey space layer, power enters the building but does not always translate into productive compute.

Intelligent orchestration allows data centres to become adaptive, dynamic and far more efficient. By coordinating workloads in real time, data centres can adapt power distribution, cooling performance and workload placement to maximise usable capacity. The result is maximising “Compute to Megawatts” - unlocking AI performance without new physical resources.

When workloads can be dynamically scheduled and shifted, AI infrastructure becomes more responsive to grid conditions and renewable energy availability. Instead of behaving as static power consumers, data centres can transform into flexible and stabilising participants of the power system.

Operators, developers and investors must shift their mindset and prioritise optimising existing infrastructure with the same urgency as building new capacity. That means investing in orchestration technologies, retrofitting legacy facilities and treating gray space as a strategic asset instead of an afterthought.

The next phase of AI infrastructure will not be defined solely by how many megawatts the industry builds, but rather how much capacity it can unlock from the systems already in place. Ultimately, more power without smarter coordination doesn’t just increase capability, it increases waste.

About the author: Wannie Park is founder and chief executive of PADO AI.