Wednesday, January 22nd 2025
US Prepares for Stargate Project: 500 Billion Dollars of AI Infrastructure Buildout
On Tuesday, the newly inaugurated United States president, Donald Trump, announced a massive AI infrastructure expansion in the US called Stargate Project. Stargate is an idea that brings private investments across the US land, with up to 500 billion US dollars committed to the project over the next four years. This is single-handedly one of the most significant infrastructure projects ever planned, and this time it is all about AI and data centers. The initial phase involves deploying 100 billion US Dollars immediately, while the remaining 400 billion will be deployed periodically over the next four years. OpenAI and SoftBank are leading this project, with Softbank's CEO Masayoshi Son being the project's chairman. Major equity partners include SoftBank, OpenAI, Oracle, and MGX. Major technology partners who will supply the know-how, planning, software, and hardware are Arm, Microsoft, NVIDIA, Oracle, and OpenAI.
Leading the entire operation will be up to OpenAI, who is gaining operational lead in the project, while Softbank oversees financial planning. Interestingly, the buildout has already begun. OpenAI is currently exploring a few sites in Abilene, Texas, which includes ten 500,000 sq. ft. data centers with 20 planned for the future. Interestingly, the infrastructure expansion will most likely be present in every US state that can provide ample land and power capacity. OpenAI is looking for partners to help with the massive data centers' power, land, and construction. The most significant impact of this project will be on the power grid, which will require additional buildout and implementation of small nuclear reactors running locally nearby to satisfy the power draw from hundreds of thousands and even millions of GPUs. OpenAI is praising NVIDIA for its almost decade-long partnership, meaning that most GPUs will likely be NVIDIA-sourced.
For AI, this means that the rapid acceleration of AI infrastructure availability will allow new technologies to be deployed broadly and the GPU capacity to expand to accommodate additional model training, inference, and, as of recently, scaling test-time computing. The current top-performing models from OpenAI, o1, and o3 use inference time scaling, where models spend a significant amount of time "reasoning"—creating a chain of thought before providing an answer, thus improving their intelligence for every situation. This is also called test-time scaling, which requires massive computational capacity, which the new Stargate Project plans to deliver.
Sources:
OpenAI, CBS News
Leading the entire operation will be up to OpenAI, who is gaining operational lead in the project, while Softbank oversees financial planning. Interestingly, the buildout has already begun. OpenAI is currently exploring a few sites in Abilene, Texas, which includes ten 500,000 sq. ft. data centers with 20 planned for the future. Interestingly, the infrastructure expansion will most likely be present in every US state that can provide ample land and power capacity. OpenAI is looking for partners to help with the massive data centers' power, land, and construction. The most significant impact of this project will be on the power grid, which will require additional buildout and implementation of small nuclear reactors running locally nearby to satisfy the power draw from hundreds of thousands and even millions of GPUs. OpenAI is praising NVIDIA for its almost decade-long partnership, meaning that most GPUs will likely be NVIDIA-sourced.
26 Comments on US Prepares for Stargate Project: 500 Billion Dollars of AI Infrastructure Buildout
i do think it’s a giant waste of money and resources For a nation with an ancient crumbling infrastructure.
Besmirching one of my all-time favourite Sci-fi IP's by association with this shit.
the exact same thing is true for ai, it will stay around and become an industry that is worth its capitalisation but that is still years away.
its far To expensive to run in its current form to be commercially viable and the QE stimmy money of the past decade and a half is gone to fuel the venture capital that’s being blown on it.
just Like real estate it needs to go pop.
Even if the AI bubble would suddenly collapse, I imagine it would take a long time for instance for Nvidia to start looking at gamers as a income possibility. I think they'd sooner try to make home cryptomining great again than that.
there is little money left on the table for nvidia mining wise after etherium moved to proof of stake and many coins followed.
But that doesn't mean that in desperation, with the certain amount of backing of GPU mineable coin, sudden large investment from the unknown entities, sudden enthusiasm in the crypto circles (paid influencers suddenly talking just about that) it couldn't come back with etherium fork, or even some other new coin.
Wrap it somehow in "AI" buzzwords, market it as perfect for mining on AI acceleration hardware that would be lying around after AI collapse - in fact it could start even without AI collapse - just with a bit of downturn.
When a bubble bursts it's companies like nvidia/supermicro that stop receiving the massive multi billion dollar orders and developers who see projects dry up.
Companies and universities won't be unplugging their servers and trying to get rid of them.
They will just use them and try to make the most of it and won't expand on them until they somehow manage to find a viable product and thats when things propperly take off
What a waste of money and resources.
Why not investing that money to space instead? That would be MUCH MORE BETTER investment than this "AI".
Dot com bubble collapsed at the height of investment, and at the rising revenue from customers - but the investors suddenly noticed they were overly optimistic and that it would take too long to return their investment and generate profit, so they pulled out, crashing the bubble.
"When space agency officials discuss why people should care about human exploration, they often say it's for the benefit of humanity."
www.scientificamerican.com/article/why-well-never-live-in-space/
The gilded age indeed...
just like there are still tulips in the Netherlands.
investments will just stall,
What I'm saying is all speculative but this massive investment will shape the age of machine learning
nvda being a nearly 4 trillion usd company…
- A small task appeared on my list, that is, as simple as possible Http Server in C
- I was asked to complete it as fast as possible
- I've responded that have No any experience with Http Servers but I would take a look at it right away
- In my Chromium browser a default Search Engine is Bing
- Here is simple log of what happened next:
...
<ScaLibBDP prompt> Simple Http Server in C.
Copilot generated codes for Http Server in C ( ~100 code lines / For Linux OS )
<ScaLibBDP> Compiled the code after solving some ptoblems with header files on Windows OS
<ScaLibBDP prompt> Does Not work on Windows. Error code 10093.
Copilot explained that the Error code 10093 is related to the Windows Sockets API (Winsock) and typically indicates that the application has not called WSAStartup or WSAStartup failed. This function initializes the use of the Winsock DLL by a process.
<ScaLibBDP prompt> Update the code of Http server to resolve error 10093 on Windows.
Copilot re-generated codes for Http Server in C ( ~100 code lines / For Windows OS )
Summary is as follows: I've spent about ten more minutes on finalizing the re-generated codes and a prototype is working now. In total, about 30 minutes...
Here is a screenshot:
It really improves productivity especially when you have Zero knowledge of something but the Manual Work is Still Needed! Period.
Now we wait for the quid pro quo...