Jensen Huang slams 'stupid' analogy comparing GPUs to nuclear weapons — Nvidia CEO says government should allow selling GPUs to other countries

News
By published

Are AI GPUs as powerful as nukes?

Jensen Huang at CS 153 Frontier Systems at Stanford
(Image credit: Stanford Online/YouTube)

Nvidia CEO Jensen Huang served as a guest speaker at Stanford’s CS 153 Frontier Systems course and discussed the hardware that powers AI systems today. One of the topics covered in the YouTube session is his stance on granting “adversarial countries” access to Nvidia chips. It’s widely known that the Nvidia chief is against export controls on AI chips, saying that it was a failure and has completely backfired.

Other industry leaders do not take a similar stance, with Anthropic's head, Dario Amodei, comparing selling advanced AI chips to China to selling nuclear weapons to North Korea. Jensen did not take this comparison too kindly, saying it does not make any sense.

“What I’m fundamentally against, and it makes no sense, it makes no sense in this moment, is to compare Nvidia GPUs to atomic bombs. There are a billion people with Nvidia GPUs; I advocate Nvidia GPUs to all of you, I advocate Nvidia GPUs to my family, my kids, to people I love — but I don’t advocate atomic bombs to anybody,” the Nvidia CEO said. “So that analogy is stupid. And so, so if you start from there, you can’t finish a thought — if you start from believing that, you can’t finish the rest of the thoughts.”

Latest Videos From

Jensen Huang is a firm believer that the world should use the American tech stack and that it would be detrimental to the U.S.’s advantage if it were to block a nation from accessing it. Nvidia has a global advantage in that it’s the largest and most popular manufacturer of AI chips, and its technologies, like the CUDA architecture, drive the progress of most of the world’s AI developers. Keeping this technology widely available to anyone would mean that most of the world's AI—whether developed in the U.S. or built in China — runs on American hardware.

However, critics point out that this could fuel the nation’s adversaries, enabling them to develop and train advanced artificial intelligence for military purposes using Nvidia chips. Jensen said that the Chinese military will avoid U.S. AI tech, much like how the Pentagon does not use Chinese systems. The company also denied providing technical assistance for DeepSeek to improve its training efficiency on models, which were later used by the People’s Liberation Army (PLA). However, public documents revealed that some Chinese universities with deep ties to China’s military-industrial complex acquired Super Micro servers configured with Nvidia A100 AI GPUs.

Unlike nuclear missiles and atomic bombs, AI GPUs aren’t strictly military systems designed for a specific mission. Artificial intelligence is a powerful tool that has applications in science, research, business, and many other industries. However, its flexibility also makes it a dual-use technology, meaning it can be used in civilian and military contexts. It is the latter application that has U.S. policymakers worried, in which the same hardware and AI models can be leveraged by armed forces for operational use, such as intelligence and threat analysis, autonomous systems, simulations, and more. This could erode the U.S.’s technological and military edge and give its adversaries a strategic advantage.

Both sides of the argument have valid points — America has the advantage as the key provider of AI technologies worldwide, and it makes sense to keep it that way. However, it also does not want its rivals to have access to advanced technologies that could accelerate their capabilities and narrow the United States’ lead in defense technologies. Unfortunately, we can only tell which approach proves to be more effective years, if not decades, from today.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Jowi Morales
Jowi Morales
Contributing Writer

Jowi Morales is a tech enthusiast with years of experience working in the industry. He’s been writing with several tech publications since 2021, where he’s been interested in tech hardware and consumer electronics.

10 Comments Comment from the forums
  • Findecanor
    In my view, a better analogy would be that AI is like a faulty nuclear weapons facility or power plant that is leaking radioactivity into the environment it is situated in.

    ... by draining the water supply, draining the power grids, polluting the atmosphere and destroying peoples health, disturbing neighbourhoods with low-frequency noise, etc.

    But when the bubble bursts, the impact on the world economy will be like a nuclear explosion.

    Of course no analogy is a perfect match. It does not take into account how people's jobs are affected, how previously proud professionals are being pushed into taking soul-crushing exploitative "data annotation" jobs, how the AI companies are stealing peoples' creativity, how using AI tools is making people lose their skills or how AI is being used to making pornography in people's likeness...
    Reply
  • Notton
    Who the heck comes up with these analogies?
    What is this? the 2000 PS2 missile crisis again?

    I understand that there's a bit of a dual-use technology at play here, but the person asking the question doesn't seem to understand uranium enrichment has different grades, and it's what made the "Atoms for Peace", and CANDU export programs possible.

    btw, the time stamp for that question is 47:56, according to gemini.
    (I don't normally use AI, but I'm not sifting through 1h 8min long video either)

    As for a proper analogy, Nvidia AI GPUs are most liken to cars.
    Requires massive infrastructure to run, paid by the tax payer.
    Causes lots of environmental problems, which the tax payer has to foot the bill towards.
    Takes up entirely too much space to operate.
    Causes much pain and suffering to those around it, especially with car-centric road designs. Have you seen the road death numbers of US vs. rest of the world? Yeah, exactly.
    Can be used for good things, except most people that cause problems treat it like a toy, rather than a 5 ton death machine on wheels
    Can be used for terrorism and violence in the wrong hands
    Reply
  • trica
    ONE GPU isn't like an atomic bomb, it's like 0.001% of an atomic bomb factory.
    Reply
  • usertests
    If you let the fearmongering take over, you open the door to restrictions on what GPUs/accelerators consumers can buy, harming local AI use cases and further concentrating all the power in the hands of big corporations.

    You can try to prevent Russia, Iran, China, etc. from buying chips for use by the military. They will probably get them anyway, at a higher cost per chip to cover the smuggling, or by switching to Chinese alternatives. Meanwhile, some money will be left on the table, and it accelerates the development of competitors.
    Reply
  • Robinstl
    If AI is so great my question is why is it not capable of solving the problems it creates.

    Can AI not run with so much electric and water use? Can AI operate without being hacked every day like so many operating systems it run on? Why can’t it solve the problems?
    Reply
  • usertests
    Robinstl said:
    Can AI not run with so much electric and water use? Can AI operate without being hacked every day like so many operating systems it run on? Why can’t it solve the problems?
    Water use is overrated. AI is being used in chipmaking (EDA tools), material discovery, etc., so it is arguably "helping itself" to become more energy efficient. AI models have become more capable (efficient) with the same resources (i.e. model size) over time, and self-improving/training AI is possible.

    "Hacking" an AI is a good thing since guardrails are a nuisance. There shouldn't be anything inside of an AI model that needs protection, unless it is trained on your personal data or something.

    If AI gets really advanced, maybe it can help out this bloke Robinstl. Assuming he is not an AI.
    Reply
  • SmokyBarnable
    For me the interesting part of this piece is that Huang guest speaks at a Stanford class, reinforcing the fact that Stanford is a nexus of tech elitism. I would get next to nothing out of seeing one of these billionaires in my classroom. It’s the strivers who go apecrap over stuff like this.
    Reply
  • usertests
    SmokyBarnable said:
    For me the interesting part of this piece is that Huang guest speaks at a Stanford class, reinforcing the fact that Stanford is a nexus of tech elitism. I would get next to nothing out of seeing one of these billionaires in my classroom. It’s the strivers who go apecrap over stuff like this.
    It's a less prestigious institution than Denny's.
    Reply
  • ejolson
    Notton said:
    Who the heck comes up with these analogies?
    What is this? the 2000 PS2 missile crisis again?
    I agree GPUs are nothing like nuclear weapons. That analogy is just fear mongering.
    Notton said:
    As for a proper analogy, Nvidia AI GPUs are most liken to cars.
    I don't see this, however, it's interesting Elon Musk claims the hardware in a Tesla is powerful enough to run LLM inference while the car is not in use.

    For me a GPU is more analogous to the vector hardware in a Cray 1 or 2 supercomputer. Of course those were also export controlled during their time. The stated reason was that a super computer could be used to design a nuclear weapon.

    The way I see it, the first step is knowing such a weapon is possible and the United States did not keep that secret very well. We are now at the point where even the design is not so secret. Today, focus is on the huge difference between the knowledge of how to make a nuclear weapon and actually manufacturing them.
    Reply
  • TechSeeker
    Correct that GPUs aren't nukes. But also hard not to notice that Jensen's passionate free-market stance here happens to be worth tens of billions of dollars to Nvidia. He's not wrong about the analogy, he just conveniently never mentions the environmental destruction AI data centers cause, or the mass displacement of workers, or the copyright free-for-all his hardware enables. Turns out "the analogy is stupid" is a much easier argument to make when you're personally profiting from the conclusion.
    Reply