"Terrifying..."
4 trillion dollars disagrees
Yes, trillion.
Nvidia (NVDA), the company you might have only just heard of, is the most valuable company in the world. Their market cap mis 4 trillion dollars.
Unless you played World of Warcraft with an augmented Alienware desktop, you might not have heard of them. Well, unless you are a US stock trader or you work for DARPA, in AI, or have another reason, like crypto mining, the fastest computer chips on the earth might not be of interest to you.
The thing is, like Intel, we are all about to discover who Nvidia are and what they want.
We understand Apple. They design consumer products, and we watch Severance on them. We pay them for products and services and protection and books and music and media (tv shows and films and movies).
What will we pay Nvidia for?
Strangely we have already paid them. So, to speak. Through DARPA (yes, they are real) we have funded research with our tax dollars.
Blackrock’s Reider has commented that he is “…most impressed by the influence of automation and robotics…”.
That influence, in the US economy alone? 15 million robots at a cost of 50.000 dollars each. Annually. That is 750 billion dollars.
Combine that with the added ‘agentic’ AI and potential super-intelligence based robotics (potentially powered by Quantum computers remotely) we are staring a technological revolution fueled by Nvidia’s Thor computers and their chips. Ultimately, technologists believe a hybrid approach where robots can access their own internal computers for ‘day to day’ requirements and then access quantum computers remotely for the more difficult or demanding requirements they might meet.
If a large language model has 175 billion parameters, then N = 175*10 to the 9th power.
For an LLM to achieve anything near what we require for robotics is 30 billion quadrillion computations per minute.
That is 350*10 to the 15th power.
To run 15 million computers (for the robots Huang suggests) we run at a capacity of 30 times 10 to the 9th power FLOPS per watt.
Nvidia are suggesting a future where the US power grid can handle our current day-to-day demand, i.e. 400 to 500 gigawatts. The robotic future run by AI requires another 700 gigawatts (current estimates).
That is 3x our current capacity utilization.
The strange part: no one knows anything. It is all theory.
The power though. That is what we will pay. On top of everything else.
We still have to build the infrastructure required, too.
We do not yet have the resources, either. People capable of doing what we have not yet defined. All understanding that what we can do in 10 years is n ot enough to achieve what is being forecast in the Superintelligence market.
the definition of which is still unclear.