Data centers use as little as 60% of power consumed for actual AI work, Nvidia claims — failures and restarts 'consume' a whopping 10% of every watt used
Nvidia claims AI data centers use only 70% of power for AI work, with 30% lost to inefficiencies like cooling, power distribution, and failures. Vertiv, collaborating with Nvidia, aims to build more sustainable data centers. Power shortages, not chip shortages, are now the main bottleneck for AI data center capacity, according to Nvidia's Rod Evans.
How this was made
The 30-second read
Why it matters
The statement introduces a new operational metric that may influence cost models for AI infrastructure investments.
Market read
Provides fresh insight into AI data‑center efficiency, a factor that could affect valuation of hardware and infrastructure firms.
What to watch
Regional grid capacity, renewable energy sourcing, and emerging cooling technologies could mitigate the cited inefficiencies.
Background
Nvidia and Vertiv discussed AI data‑center power usage at an EMEA event, noting that only 70% of consumed electricity is used for AI processing.
Ticker impact
Nvidia disclosed that only 70% of power in AI data centers is used for AI work, highlighting significant inefficiencies.
potential modest pressure as investors reassess power‑cost assumptions for Nvidia‑powered data centers
No immediate financial metric was released; the news is a qualitative operational insight that could influence valuation models.
Vertiv is cited as Nvidia's partner in co‑designing more efficient AI data centers.
limited impact; any upside depends on future contract announcements
The article only mentions Vertiv in a collaborative context without new contract terms or financial data.
Market effects
Highlights power‑efficiency challenges for the broader AI data‑center and semiconductor supply chain.
May affect European and EMEA data‑center operators that rely on Nvidia GPUs.
Relevant to global AI infrastructure investors as power costs become a key factor.
Counterpoint
The inefficiency figures could be overstated; future hardware improvements may offset power losses.
Key entities
- CompanyNvidia
AI hardware and GPU manufacturer.
- CompanyVertiv
Data‑center infrastructure provider.
