• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Stargate is real!

Lakados said:
https://openai.com/index/announcing-the-stargate-project/

The US Government is helping back a $500B AI datacenter project across the US for research and stuff.

Arm, Microsoft, NVIDIA, Oracle, and OpenAI are the primary tech partners with SoftBank handling the money.
Click to expand...
When that many megacorps are involved, you know this isn't going to be good for any of us.
Once again, the dark cyberpunk future rears its cyberdemon head by showing us that life imitates art. :borg:


View: https://www.youtube.com/watch?v=L3PBoSk6Z38&t=1s

This mirrors the Cyberpunk 2077 universe when the Shallow Net replaced the Old NET circa 2022 after the DataKrash occurred, and Blackwall (massive AI firewall) was implimented to keep rouge AIs contained in the Old NET.
I normally say that when things are bi-partisan we are the ones getting screwed, but in this case the only ones this is going to end well for are the AIs...

Cyberpunk_2077_AI.jpg
 
Last edited:
This thread made me chuckle and my wife asked me what was funny. I honestly didn't know to explain it to her without having to go on for hours just to give her multiple back stories. So I just said nerd stuff and she nodded her head and went on her way.
 
Decko said:
Why do I feel like this is a giant waste of money?
Click to expand...
Considering how much resources are required for these AI datacenters,, power, water, and all other resources for said buildings...


AI search query requires 10x as much power as a google search.... :D
 
That was 30x time not so long ago and with Ironwood this year it could become ~5x already (they have a claimed 2x per watt improvement, over the 2024).

On the hardware side of things google is 30x more efficient than when they started heavily use AI inference in their search engine in 2018, according to Pichai it will soon be a bit of an non issue, a bit like how expensive video was and now we take it for granted with youtube.
 
Assuming the investment materializes, the net effects of Stargate on the semiconductor industry are likely to be some combination of enhancing Nvidia’s stranglehold on the AI infrastructure market, and lining its pockets—all while accelerating OpenAI’s transition into a hyperscale chip customer in its own right.


Among the first things to consider when building an AI data center is where the power will come from (especially in Texas, where the power grid is notoriously unreliable). Each Nvidia NVL72 is a full rack in size and uses 120 kW of power, meaning 10,000 of them would need a gigawatt, which is equivalent to the amount of power a full-size nuclear power plant can produce today. Trump appeared to confirm that Stargate sites will generate their own power, saying that “we’re going to make it possible for them to get that done easily, at their own plants if they want…but it’s technology and AI all made in the USA.”

While small modular reactors could produce in the order of 300 MW of nuclear energy each, the technology is not yet ready for commercial deployments. Oracle reportedly currently licenses a site in Abilene, Texas, from a data center operator called Crusoe (unclear if this is the same site marked for Stargate). Crusoe specializes in powering data centers with waste material from co-located oil fields using a technique it calls digital flare mitigation.

While it is powered by natural gas , Crusoe pitches this scheme as sustainable since the excess gas would otherwise simply be burned, and burning it to use burns it more sustainably than waste stacks would. Potential capacity for digital flare mitigation in Abilene is unknown, but Crusoe also has plans for wind-powered deployments.

https://www.eetimes.com/project-stargate-trump-announces-500-billion-of-u-s-ai-infrastructure/
 
appeared to confirm that Stargate sites will generate their own power
Click to expand...
This makes good sense... however what is the mantra that is often beaten into our heads? If we want a nuclear power plant we need to plan like 30 years into the future. Or something to that extent, which means in the meantime they'll be leeching off the regular power grid, and eventually when they do get their own power plants that power will most likely be for them only and excess will not go onto the grid even in emergencies as mentioned for Texas.
 
sfsuphysics said:
If we want a nuclear power plant we need to plan like 30 years into the future. Or something to that extent, which means in the meantime they'll be leeching off the regular power grid, and eventually
Click to expand...
It will be a lot of new natural gas capacity stopgap before those. Nuclear can go up in 5-6 years (it does in China) technically and an Microsoft under a Trump admin could get it up much faster than in the past too.

For example:
https://www.powermag.com/entergy-lo...red-generation-to-support-data-center-demand/

sfsuphysics said:
will most likely be for them only and excess will not go onto the grid even in emergencies as mentioned for Texas.
Click to expand...
Inference part could have hourly peak and what not, training could have peak as well, which could make interesting selling back to the grid, same for their water treatment datacenter often provide city.

For example:
https://www.ainvest.com/news/google-cuts-ai-data-center-energy-peak-demand-grid-support-move-2508/
Google has taken a pioneering step among major U.S. technology companies by agreeing to reduce energy consumption at its artificial intelligence (AI) data centers during periods of peak electricity demand... Google has entered agreements with Indiana Michigan Power and the Tennessee Valley Authority to temporarily cut power use at its data centers, a strategy commonly referred to as a demand-response program

I think this will be almost certainly the norm, they will able be working hard on social accessibility that will include water-power grid participation and future project of clean energy generation, the margin those company make open the door for projects that will not have made sense in those regards in the past for different high power use industry. And a lot of those techs will be quite controllable/modular demand wise, they all have subscription tier that give different level of priority and performance, massive training can be done during low demand time of the year and so on.
 

Samsung, SK Hynix join OpenAI’s $500 bn Stargate project with HBM supply pacts​

The creator of ChatGPT has requested volumes equivalent to as many as 900,000 wafers a month, twice the current global production capacity​

By Jeong-Soo Hwang and Chae-Yeon Kim
Published October 1, 2025 at 9:17 PM(KST)

The two South Korean chipmakers signed letters of intent on Wednesday with OpenAI Chief Executive Sam Altman to supply high-bandwidth memory (HBM) semiconductors for the US start-up’s so-called Stargate initiative, a program to construct a network of hyperscale AI data centers around the world by 2029.


The San Francisco-based company, the creator of generative AI ChatGPT, has requested volumes equivalent to as many as 900,000 wafers a month, more than twice the current global production capacity.

Analysts estimate the commitment amounts to more than 100 trillion won ($72 billion) of incremental demand for the Korean chipmakers over the next four years.


https://www.kedglobal.com/artificial-intelligence/newsView/ked202510010013

Via

OpenAI orders $71B in Korean memory chips​

CEO Altman orders $71 billion in HBM chips for Stargate project.
Robert Clark, Contributing Editor, Light Reading
October 2, 2025

https://www.lightreading.com/ai-machine-learning/openai-orders-71b-in-korean-memory-chips
 
What a giant waste of money, maybe people lose their jobs, can't buy pcs, and your electric bill goes through the roof. Such a wasteful way of doing things. I hope some day we have LLMs that fit on a single GPU that perform as well as today's frontier models, all of this BS will be moot.
 
If we have LLMs that fit on GPU work really well, people will want to have thousand of them on pods and if it is so good that some people lose some jobs like tractors and gas lights is it a waste of money (and how many people find new jobs) ?

We are a bit past LLM already which could make it hard to work on single GPU, today frontier model are multi-modal with audio/video/text in an unified vector space and will add senses, emboddied AI robots will add to that unified vector space world reaction to their actions and "touch" feedback (and like the human that has over 10 senses they will have a lot of them, like their motors efforts, their joints rotations, lidar, thermal camera, "smell").

Context size will go to 100 millions token+ and will still love giant memory amount, it is for a fact very wasteful way to learn, an human learn a lot with less watts (brain is what around 20watt), but things goes fast, Claude Opus is 66% cheaper per token while using way less token making up too 90% rebate from a couple of months ago.

But this is so valuable that the more efficient and cheaper it gets, the more of it you will want and we did this for things far less valuable, a small CPU-GPU soc can perform better than a 1981 room sized super computer in many things, if one day a programming agent of Gemini 3 level can run on a 20900RTX 320GB vram still way better to use Gemini 12 (at a better price on a more efficient and made for it infrastructure), like some small version of model are better than GPT 3 were at some things...
 
Marees said:
Oracle reportedly currently licenses a site in Abilene, Texas, from a data center operator called Crusoe (unclear if this is the same site marked for Stargate). Crusoe specializes in powering data centers with waste material from co-located oil fields using a technique it calls digital flare mitigation.

While it is powered by natural gas , Crusoe pitches this scheme as sustainable since the excess gas would otherwise simply be burned, and burning it to use burns it more sustainably than waste stacks would. Potential capacity for digital flare mitigation in Abilene is unknown, but Crusoe also has plans for wind-powered deployments.

https://www.eetimes.com/project-stargate-trump-announces-500-billion-of-u-s-ai-infrastructure/
Click to expand...

OpenAI's [expansion plans for] massive Stargate data center canceled as firm can't reach terms with Oracle, operator struggles with reliability issues — Meta said to be interested in snatching excess capacity​

News
By Anton Shilov published 3 hours ago
Too much ado, or Stargate has problems?

Oracle and OpenAI have cancelled plans to expand their flagship AI data center campus in Abilene, Texas, after lengthy negotiations broke down over financing arrangements and OpenAI's changing capacity projections, reports Bloomberg. The Abilene campus is part of the Stargate project announced at the White House last year, and yet, it seems like it faces slowdowns. Just like other Oracle's data centers meant for OpenAI. Yet, Meta could take the yet-to-be-expanded space.

Despite shelving the expansion of one Stargate project, Oracle's general partnership with OpenAI remains unchanged. In July last year, Oracle agreed to develop 4.5 GW of data center capacity for OpenAI, and that program continues. The companies have also announced projects in other locations, including a site near Detroit owned by Related Digital.

*One gigawatt is comparable to the output of a nuclear reactor and can supply electricity to hundreds of thousands of homes at peak usage.

https://www.tomshardware.com/tech-i...to-be-interested-in-snatching-excess-capacity
 
You must log in or register to reply here.
Back
Top