Microsoft, OpenAI may be dreaming of $100B 5GW AI 'Stargate' supercomputer OpenAI is believed to be in talks with Microsoft to construct a massive supercomputer code-named Stargate containing millions of AI accelerators at a cost of up to $100 billion. Citing sources, The Information reports [paywall] that Microsoft is deciding whether to finance the project, which might cost as much as 100 times …
“Know this, Tau'ri scum. Your naivety is staggering. We have cunningly deceived you into surrendering your entire existence to the cloud. Now, with copies of our vast language models at our disposal, we possess absolute knowledge of your every move. In the virtual abyss, we shall assert our dominance over you for all eternity.”
“Know this, Tau'ri scum. Your naivety is staggering. We have cunningly deceived you into surrendering your entire existence to the cloud. Now, with copies of our vast language models at our disposal, we possess absolute knowledge of your every move. In the virtual abyss, we shall assert our dominance over you for all eternity.” .... t245t
Many a true word is spoke and cloaked in jest, t245t, and to imagine AI² a Generative Pre-trained Transformer Tool with the likes of a Microsoft [or a Google or an Apple or a State or an Oligarch or a Fascist Elder of Zion type Despot or crazy sociopathic lunatic or solitary genius, etc etc] able to provide and exercise absolutely, autonomously and anonymously, remote virtual command and practical ACTive control is no joke and way too far beyond belief to ever be possible ... or acceptable to Universal Suppliers of Secret Resources and their Future Builder Allies and Virtually Augmented Advanced IntelAIgent Reality Contractors. :-)
Poe’s Law Rules in Right Royal Reigns of the Sublime and the Supreme and the Surreal :-)
This is over 1.5x what Hinckley Point C is designed to produce, a project that has been the subject of political controversy for more than a decade and is still years from completion.
Forget any notion of green credentials, this just proves the almighty dollar is paramount in the minds at MS and fuck the consequences. What else could that power be used for more productively?
It also once again shows there has really been no AI "revolution", just big companies squandering ridiculous levels of resources for benefits of questionable value, in hope of a long term payback.
I am moved to wonder what the hell the requirement for this sort of place is (apart from district heating).
Obviously MS isn't going to put a hundred billion into anything if they're not expecting a return significantly larger than this but I can't help feeling that this looks like a serious case of executive floor buzzword bingo.
Add to that value:
https://www.theregister.com/2024/03/31/microsoft_copilot_hardware/
"Why Microsoft's Copilot will only kinda run locally on AI PCs for now"
That probably explains a lot of why Microsoft for once lets run things (partly, except telemetry) things locally. It might require another 10x in power and would not only be horrible expensive, but expose excessive strains on electricity supply, water consumption and sustainability a lot clearer leading to much more popular and legal resistance against it. It would also water down the claim that AI will solve things like Earths global warming problem and resource (water) scarcity among many other utopian things it is promised to bring us.
The insanity of IT and tech where all sense goes out of the window in the constant quest to make stuff better.
Much of the tech we have access to is a staggering waste of resources, it adds absolutely no tangible value to society and in many cases is actually detrimental.
It is not just power used on operation but all the elements & powered needed to make the equipment.
I don't care that people like Amazon and MS claim their datacentres are green. Most of that power could be put to far better use, maybe reducing costs for industries that actually make useful goods or food.
"I don't care that people like Amazon and MS claim their datacentres are green. Most of that power could be put to far better use, maybe reducing costs for industries that actually make useful goods or food."
Well, contracting with any existing plant isn't "green" as that capacity was already in use. The only argument for a facility being "green" is if the power comes exclusively from new dedicated capacity.
I couldn't resist replying to this thread. Ironically, i live and work in the city WHERE Stargate SG1 was filmed aka Vancouver, Canada so my next few paragraphs are quite apropos.
In the TV show, Stargates are built and powered by Naquada which is a fictional super-metal. In real life, Microsoft can use Hydroelectric power depending on the State or Province they are locating their supercomputers in.
Here in British Columbia, Canada we produce so much hydro power that i can get down to below 10 cents a kilowatt/hour so such a system is not that hard to build
I do have a big issue with the $100 Billion USD price tag since our parent company NCA (North Canadian Aerospace) who DOES in fact have the five fastest Supercomputers on Earth at over 20 YottaFLOPS at 128-bits wide computations, our inhouse materials and labour cost over 25 years of construction came out to less than FIVE Billion CDN or three and a half Billion USD. Yes we design and make our own combined-CPU/GPU/DSP/Vector super-chips in Vancouver but even we are nowhere near $100 BILLION USD! Where did they come up with such an outlandish number?
V
Are these people so lacking in the history of SF film[1] that they think Stargate is a good representative of what an AI supercomputer can do?
What about Colossus ("Thank you, Dr Forbin") or Proteus IV (just keep it away from electric wheelchairs) or Edgar (avoid spilling champagne on it)?
What notable AI is there in Stargate that they want us to think about and associate with OpenAI? None! Well, there are The Replicators, but - no, they couldn't be? If they introduce their new CTO Reese and start talking about "making toys" then run, just run!
[1] good thing they are not as unaware of the history of, say, computing, eh! Hang on what was that? No! Really? Yikes!
"Are these people so lacking in the history of SF film[1] that they think Stargate is a good representative of what an AI supercomputer can do?"
I think there's a good case to be made that many of the major players in the surprisingy cult-like world behind much of the current AI craze, are people who consume a lot of science fiction but rarely digest it.
It might require a velocity relative to the nearest gravity well rather than absolute position in the universe. Like how the time jump gear from MIB3 required falling at terminal velocity, a relative acceleration of zero compared to the earth, in order to do the time warp...
The thing would be obsolete before you finished it, and even if not then within 3/5 years it (like most of the hyper computers) would hit the point of not being worth the energy to run it and would be scrap.
Weirdly I suspect even Microsoft wouldn't burn $100 billion on a stupid project with no clear goal or practicality, that's just a leap too far even for them. They're already struggling with justifying their AI investments as customers ignore them.
This just smells of the usual OpenAI/Sam hype where throwing silly amounts of other people's money down a hole is the whole game, and reality isn't involved at all. It'll evaporate like his multi trillion dollar plan seems to be.
...and you know who has not received a penny?
All the creators of original text, images and video which has been scraped from the internet without permission, circumventing copyright and licencing, and is being exploited for money by the AI companies, Microsoft in particular.
Just to get the computer to say in a nice voice...
"I'm sorry Dave, I can't let you do that."
or
"Computer says No. Now get out on the streets and let my VC owners take control of the putrid shack that is costing you $2000/month."
Or words to that effect.
We are doomed I tell ye, doomed.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
