Nvidia's Jensen Huang says all telecom operators will offer AI computing power
In short
Nvidia CEO Jensen Huang said at the GTC conference in Berlin on 21 October that all telecom operators will offer AI computing power in future and that base stations could become a distributed AI network. He also said building a 1-gigawatt data centre currently costs about $50 billion.
Read the full story 1 min read
Nvidia chief executive Jensen Huang said at the GTC conference in Berlin on 21 October that all telecom operators will offer AI computing power in future, and that telecom base stations could develop into a distributed AI network, according to CLS, Jiemian, 36Kr and IT Home. [ 1 , 2 , 3 , 4 ]
Huang compared the AI economy to a five-layer cake made up of energy, computing power, AI factories, models and applications. In his presentation, the computing layer showed the logo of only one company, Nvidia. Other layers included energy companies such as Siemens Energy and Schneider Electric, AI factory builders Nscale and Firebird, European AI model company Mistral, legal AI company Legora and translation tool DeepL. IT Home described the layers as energy, chips, infrastructure, models and applications, and said Huang stressed that AI is not only software but underlying industrial infrastructure like electricity and the internet, with the layers closely linked and pulling upward. [ 1 , 3 , 4 ]
Huang also said building a 1-gigawatt data centre currently costs about $50 billion, covering land, power, construction, computing equipment, storage and cooling. [ 1 , 3 ]
In a separate item, 36Kr, citing Jiemian, summarised Huang's keynote as saying AI will change the telecom industry, improve spectrum efficiency and turn base stations into distributed AI infrastructure. It reported him saying about 100,000 sites could form a distributed data centre network of about 100 gigawatts, and that after nearly eight years of development Nvidia and Nokia had completed a first commercial phone call on T-Mobile's network. Only this summary carried those two points. [ 5 ]
Jiemian included the report in its AI morning roundup on 22 October. [ 6 ]
Why it matters
The remarks, as reported by Chinese business and tech outlets, set out Nvidia's view of how telecom networks and the wider AI industry could develop. They are statements by the company's chief executive, and the documents give no independent assessment.
Key facts
- Huang spoke at the GTC conference in Berlin on 21 October. [ 1 , 2 , 3 , 4 ]
- He said all telecom operators will offer AI computing power in future, and that telecom base stations could develop into a distributed AI network. [ 1 , 2 , 3 , 4 ]
- He likened the AI economy to a five-layer cake of energy, computing power, AI factories, models and applications. [ 1 , 2 , 3 ]
- He said building a 1-gigawatt data centre currently costs about $50 billion, covering land, power, construction, computing equipment, storage and cooling. [ 1 , 2 , 3 ]
- 36Kr, citing Jiemian, reported Huang saying about 100,000 sites could form a distributed data centre network of about 100 gigawatts. [ 5 ]
- 36Kr, citing Jiemian, reported Huang saying that after nearly eight years of development Nvidia and Nokia completed a first commercial phone call on T-Mobile's network. [ 5 ]
Confirmed by several sources
Still unclear
- Whether Nvidia and Nokia completed a first commercial call on T-Mobile's network after nearly eight years of development. Only one summary, which credits Jiemian, reports it; no article text is available.
- The claim that about 100,000 sites could form a roughly 100-gigawatt distributed data centre network. Only one summary reports it, and no article text gives detail on how the figure was derived.
- How the five layers are defined. Most outlets list energy, computing power, AI factories, models and applications; IT Home lists energy, chips, infrastructure, models and applications.
- Huang presented the AI economy as a five-layer cake; in his slide the computing layer showed only Nvidia's logo, while other layers included Siemens Energy, Schneider Electric, Nscale, Firebird, Mistral, Legora and DeepL. Reported by a single source so far
- Huang said a 1-gigawatt data centre currently costs about $50 billion to build. Reported by a single source so far
What local media are saying
Timeline, local time
- CLS publishes its report on Huang's remarks. [ 1 ]
- 36Kr publishes the report, citing Jiemian. [ 2 ]
- Jiemian publishes its report. [ 3 ]
- IT Home publishes its report, including the five-layer cake and the companies named. [ 4 ]
- 36Kr publishes a second item citing Jiemian on the base station network and the Nokia call. [ 5 ]
- Jiemian includes the report in its morning AI roundup of 22 October. [ 6 ]