Jensen Huang: Telecom Base Stations Will Evolve into Distributed AI Networks
nashnova research
Nvidia CEO Jensen Huang told the Berlin GTC conference that telecom base stations will become distributed AI compute nodes, and disclosed that a 1-gigawatt data center now costs roughly $50 billion — a price tag that puts AI infrastructure on par with major energy projects.
How do cell towers become an AI network?
Huang said on October 21 that every telecom carrier will eventually offer AI compute.
This means → base stations stop being mere signal relays and become the AI compute nodes closest to the user.
In plain terms = today a cell tower handles your calls and data; tomorrow it also runs AI tasks on the spot, instead of routing everything back to a distant mega data center.
What is the "five-layer cake"?
Huang framed the AI economy as a five-layer cake: energy, compute, AI factories, models, and applications.
Only one company logo appeared on the compute layer — Nvidia's. This means → Huang publicly declared that in compute, Nvidia is the sole player.
Companies on the other layers include Siemens Energy and Schneider Electric (energy), Nscale and Firebird (AI factory builders), and Mistral, Legora, and DeepL (models and applications).
What does a top-tier data center actually cost?
Huang disclosed that building a 1-gigawatt data center — roughly the output of a large nuclear plant — now runs about $50 billion.
That covers land, power, construction, compute hardware, storage, and cooling. In plain terms = everything from breaking ground to switching on, bundled into one figure.
This reflects AI infrastructure entering a capital-heavy, long-cycle phase where only a handful of hyperscalers or sovereign funds can foot the bill alone.
市场有风险，内容仅供研究参考，不构成投资建议。