Power Shortages Are Now the Biggest Roadblock for AI Data Centers

The biggest obstacle to building AI data centers is no longer space or chips. It is electricity. NVIDIA says companies around the world are now building on its DSX AI Factory Platform, which aims to plan, simulate and run the power grid and the data center as one system, from the substation to the rack.

What changed

NVIDIA calls power the biggest bottleneck in infrastructure today. It says more than $300 billion in power equipment is backlogged. It also says over 200 gigawatts of projects are waiting in U.S. interconnection queues, which are the lines of projects waiting for permission to connect to the grid.

One cause, according to NVIDIA, is siloed systems. The grid, the building and the computers inside have long been designed separately. That creates delays, risk and stranded capacity, meaning power that exists but cannot be used.

Who is involved

Several energy and engineering firms are adding their products to the platform. These are NVIDIA's own descriptions of the work.

• Hitachi Energy: its battery storage system qualified as NVIDIA DSX Ready.
• Schneider Electric, with AVEVA and ETAP: power and cooling designs plus a digital twin, a virtual copy used to test a data center.
• Siemens Energy: a reference design for power generation, transmission and distribution.
• GE Vernova: grid-to-facility and rack power expertise to support DSX systems.

The numbers worth noting

Phaidra says its self-learning software, which manages cooling and power use, delivers about 10% more computing within the same power supply. Emerald AI built a tool to help utilities tap underused grid capacity. It is modeled to unlock up to 100 gigawatts on the existing U.S. grid. Both are claims from the companies' own work, not independent tests.

Why it matters to you

AI services such as chatbots and writing tools run in power-hungry data centers. If power limits how fast new ones open, that could affect how quickly new AI tools arrive and what they cost your business. Getting more output from the same electricity is one way to ease that pressure.

What is not known yet

The material does not say when any of this will reach real data centers at scale. It gives no costs, and it does not say how much the platform will shorten connection waits. Much of it is reference designs and modeled results, which are plans and estimates rather than finished projects. The announcement ties into GTC Berlin, where NVIDIA and energy firms are presenting.