The Electricity Problem at the Heart of Modern AI
Every AI data centre runs on a fixed power budget. The company that squeezes the most useful work out of each watt wins. Here is what that means in plain English.
Key points
- NVIDIA's GB300 NVL72 system delivers up to 25 times more useful AI work per watt than the previous Hopper generation, according to analysis by SemiAnalysis InferenceX.
- Software improvements alone boosted performance per watt on one model by up to 5 times within a single month in 2025.
- NVIDIA claims its power-management software lets operators run up to 40 percent more chips within the same electricity budget.
- Companies including Anthropic, OpenAI, and Perplexity currently run production workloads on Blackwell NVL72 hardware, according to NVIDIA.
- The figures in this article come from an NVIDIA promotional post, not an independent peer-reviewed study.
AI is, at its core, an electricity problem.
Every time a chatbot answers your question or an AI agent, software that carries out multi-step tasks on its own, books your travel or drafts your legal document, a data centre somewhere burns power to do it. That data centre has a fixed electricity budget. The company that generates the most useful output from that budget makes the most money. The company that does not, cannot scale.
NVIDIA describes this as "performance per watt," meaning how much AI output you get for each unit of electricity consumed. The higher that number, the more tokens (the small text chunks AI systems process) a facility can produce before hitting its power ceiling.
The numbers NVIDIA published are striking. Its GB300 NVL72 system, a rack-scale server platform built around its latest Blackwell chips, reportedly delivers up to 25 times more performance per watt than its older Hopper generation on DeepSeek V4 Pro, one of the current open AI models. On GLM5.1, another frontier model, the gain is up to 20 times. On Kimi K2.6, a model built for longer automated tasks, it reaches up to 10 times.
Those figures come from SemiAnalysis InferenceX, a third-party benchmarking group, cited in NVIDIA's own blog post. They are not from a peer-reviewed study or an independent audit.
Does any of this affect ordinary people?
Yes, indirectly but meaningfully. The speed, cost, and availability of AI tools ordinary people use every day all depend on how efficiently data centres run. Perplexity, the AI search company, says it runs hundreds of millions of queries daily on Blackwell hardware. If the underlying infrastructure becomes more efficient, the services built on top of it can get faster and cheaper.
NVIDIA also highlights a less obvious drain: cooling. In a typical AI facility, only around 60 percent of electricity drawn from the grid actually reaches the chips and does useful work. The rest is lost to heat management and other inefficiencies. NVIDIA's DSX MaxLPS software, which shifts power between chips in real time and supports liquid cooling, aims to recover some of that waste.
For the companies running these systems, Anthropic and OpenAI among them according to NVIDIA, this is about economics as much as engineering. More output per watt means lower cost per query, which feeds directly into profit margins.
Software matters as much as hardware here. NVIDIA says that on DeepSeek V4, updates to its inference software stack improved performance per watt by up to 5 times within a single month. That is a large gain without replacing a single chip.
All of this is foundation-laying for NVIDIA's next platform, called Vera Rubin, which the company says will push rack-scale efficiency further still.
The core tension is real even if the marketing is selective: electricity is finite, AI appetite is not, and every lab building the next frontier model has to reckon with that gap.
