You have to keep in mind Microsoft is planning on spending almost 100B in datacenter capex this year and they're not alone. This is basically OpenAI matching the major cloud provider's spending.
This could also be (at least partly) a reaction to Microsoft threatening to pull OpenAI's cloud credits last year. OpenAI wants to maintain independence and with compute accounting for 25–50% of their expenses (currently) [2], this strategy may actually be prudent.
Depends on your definition of profitability, They are not recovering R&D and training costs, but they (and MS) are recouping inference costs from user subscription and API revenue with a healthy operating margin.
Today they will not survive if they stop investing in R&D, but they do have to slow down at some point. It looks like they and other big players are betting on a moat they hope to build with the $100B DCs and ASICs that open weight models or others cannot compete with.
This will be either because training will be too expensive (few entities have the budget for $10B+ on training and no need to monetize it) and even those kind of models where available may be impossible to run inference with off the shelf GPUs, i.e. these models can only run on ASICS, which only large players will have access to[1].
In this scenario corporations will have to pay them the money for the best models, when that happens OpenAI can slow down R&D and become profitable with capex considered.
[1] This is natural progression in a compute bottle-necked sector, we saw a similar evolution from CPU to ASICS and GPU in the crypto few years ago. It is slightly distorted comparison due to the switch from PoW to PoS and intentional design for GPU for some coins, even then you needed DC scale operations in a cheap power location to be profitable.
They will have an endless wave of commoditization chasing behind them. NVIDIA will continue to market chips to anyone who will buy... Well anyone who is allowed to buy, considering the recent export restrictions. On that note, if OpenAI is in bed with the US government with this to some degree, I would expect tariffs, expert restrictions, and all of that to continue to conveniently align with their business objectives.
If the frontier models generate huge revenue from big government and intelligence and corporate contracts, then I can see a dynamo kicking off with the business model. The missing link is probably that there need to be continual breakthroughs that massively increase the power of AI rather than it tapering off with diminishing returns for bigger training/inference capital outlay. Obviously, openAI is leveraging against that view as well.
Maybe the most important part is that all of these huge names are involved in the project to some degree. Well, they're all cross-linked in the entire AI enterprise, really, like OpenAI Microsoft, so once all the players give preference to each other, it sort of creates a moat in and of itself, unless foreign sovereign wealth funds start spinning up massive stargate initiatives as well.
We'll see. Europe has been behind the ball in tech developments like this historically, and China, although this might be a bit of a stretch to claim, does seem to be held back by their need for control and censorship when it comes to what these models can do. They want them to be focused tools that help society, but the American companies want much more, and they want power in their own hands and power in their user's hands. So much like the first round where American big tech took over the world, maybe it's prime to happen again as the AI industry continues to scale.
Why would China censoring Tiananmen Square/whatever out of their LLMs be anymore harmful to the training process when the US controlled LLMs also censor certain topics, eg "how do I make meth?" or "how do I make a nuclear bomb?".
Because China censors very common words and phrases such as "harmonized", "shameless", "lifelong", "river crabbed", "me too". This is because Chinese citizens uses puns and common phrases initially to get around censors.
They want their LLMs explicitly approved to align with the values of the regime. Not necessarily a bad thing, or at least that avenue wasn't my point. It does get in the way of going fast and breaking things though, and on the other side there is an outright accelerationist pseudo-cult.
Ignoring the moral dimension for a second, I do wonder if it is harder to implement a rather cohesive, but far-reaching censorship in the chinese style, or the more outrage-driven type of "censorship" required of American companies. In the West we have the left pre-occupied with -isms and -phobias, and the right with blasphemy and perceived attacks on their politics.
With the hard shift to the right and Trump coming into office, especially the last bit will be interesting. There is a pretty substantial tension between factual reporting and not offending right-wing ideology: Should a model consider "both sides" about topics with with clear and broad scientific consensus if it might offend Trumpists? (Two examples that come to mind was the recent "The Nazis were actually left wing" and "There are only two genders".)
> they (and MS) are recouping inference costs from user subscription and API revenue with a healthy operating margin.
I tried to Google for more information. I tried this search: <<is openai inference profitable?>>
I didn't find any reliable sources about OpenAI. All sources that I could find state this is not true -- inference costs are far higher than subscription fees.
I hate to ask this on HN... but, can you provide a source? Or tell us how do you know?
Not necessarily. DeepSeek will probably only threaten the API usage of OpenAI, which could also be banned in the US if it's too sucessful. API usage is not a main revenue for OpenAI (it is for Anthropic last time I checked). The main competitor for R1 is o1, which isn't gnerally available yet.
Not quite. In 2 years their revenue has ~20x from 200M ARR to 3.7B ARR. The inference costs I believe pay for themselves (in fact are quite profitable). So what they're putting on their investor's credit cards are the costs of employees & model training. Given it's projected to be a multi-trillion dollar industry and they're seen as a market leader, investors are more than happy to throw in interest free cash flow now in exchange for variable future interest in the form of stocks.
That's not quite the same thing at all as your credit card's revenue stream as you have a ~18%+ monthly interest rate on that revenue stream. If you recall AMZN (& all startups really) have this mode early in their business where they're over-spending on R&D to grow more quickly than their free cash flow otherwise allows to stay ahead of competition and dominate the market. Indeed if investors agree and your business is actually strong, this is a strong play because you're leveraging some future value into today's growth.
Platform economics "works" in theory only upto a point. Its super inefficient if you zoom out and look not at system level but ecosystem level. It hasn't lasted long enough to hit failure cases. Just wait a few years.
As to openai, given deepseek and the fact lot of use cases dont even need real time inference its not obvious this story will end well.
I also can't see it ending well for OpenAI. This seems like it's going to be a commodity market with a race to the bottom on pricing. I read that NVIDIA has a roughly 1000% (10x) profit margin on H100's, which means that someone like Google making their own TPUs has a massive cost advantage.
Moore's law seems to be against them too... hardware getting more powerful, small models getting more powerful... Not at all obvious that companies will need to rely on cloud models vs running locally (licencing models from whoever wants that market). Also, a lot of corporate use probably isn't that time critical, and can afford to run slower and cheaper.
Of course the US government could choose to wreck free-market economics by mandating powerful models to be run in "secure" cloud environments, but unless other countries did same that might put US at competitive price disadvantage.
Meanwhile, Azure has failed to keep up with the last 2-3 generations of both Intel and AMD server processors. They’re available only in “preview” or in a very limited number of regions.
I wonder if this is a sign of the global economic downturn pausing cloud migrations or AI sucking the oxygen out of the room.
Lots of back door deals. Just expect more government things put in TX just like the Army built that place in Austin, when we have plenty of dead bases that could be reused
My kneejerk response was to point to the incoming administration, but the fact Stargate has been in the works for more than a year now says to me it's because of tax credits.
~$125B per year would be 2-3% of all domestic investment. It's similar in scale to the GDP of a small middle income country.
If the electric grid — particularly the interconnection queue — is already the bottleneck to data center deployment, is something on this scale even close to possible? If it's a rationalized policy framework (big if!), I would guess there's some major permitting reform announcement coming soon.
They say this will include hundreds of thousands of jobs. I have little doubt that dedicated power generation and storage is included in their plans.
Also I have no doubt that the timing is deliberate and that this is not happening without government endorsement. If I had to guess the US military also is involved in this and sees this initiative as important for national security.
Is there really any government involvement here? I only see Softbank, Oracle, and OpenAI pledging to invest $500B (over some timescale), but no real support on the government end outside of moral support. This isn't some infrastructure investment package like the IRA, it's just a unilateral promise by a few companies to invest in data centers (which I'm sure they are doing anyway).
It’s light on details, but from The Guardian’s reporting:
> The president indicated he would use emergency declarations to expedite the project’s development, particularly regarding energy infrastructure.
> “We have to get this stuff built,” Trump said. “They have to produce a lot of electricity and we’ll make it possible for them to get that production done very easily at their own plants.
On the one hand the number is a political thumb-suck which sounds good. It's not based in any kind of actual reality.
Yes, the data center itself will create some permanent jobs (I have no real feel for this, but guessing less than 1000).
There'll be some work for construction folk of course. But again seems like a small number.
I presume though they're counting jobs related to the existence of a data center. As in, if I make use of it do I count that as a "job"?
What if we create a new post to leverage AI generally? Kinda like the way we have a marketing post, and a chunk of the daily work there is Adwords.
Once we start gustimamating the jobs created by the existence of an AI data center, we're in full speculation mode. Any number really can be justified.
Of course ultimately the number is meaningless. It won't create that many "local jobs" - indeed most of those jobs, to the degree they exist at all, will likely be outside the US.
So you don't need to wait for a post-mortem. The number is sucked out of thin air with no basis in reality for the point of making a good political sound bite.
Just as there is an AWS for the public, with something similar but only for Federal use, so it could be possible that there is AI cloud services available to the public and then a separate cloud service for Federal use. I am sure that military intelligence agencies etc. would like to buy such a service.
Gas turbines can be spun up really quickly through either portable systems (like xAI did for their cluster) [1] or actual builds [2] in an emergency. The biggest limitation is permits.
With a state like Texas and a Federal Government thats onboard these permits would be a much smaller issue. The press conference makes this seem more like, "drill baby drill" (drilling natural gas) and directly talking about them spinning up their own power plants.
> It's similar in scale to the GDP of a small middle income country
I’ve been advocating for a data centre analogue to the Heavy Press Programme for some years [1].
This isn’t quite it. But when I mapped out costs, $1tn over 10 years was very doable. (A lot of it would go to power generation and data transmission infrastructure.)
One-time capital costs that unlock a range of possibilities also tend to be good bets.
The Flood Control Act [0], TVA, Heavy Press, etc.
They all created generally useful infrastructure, that would be used for a variety of purposes over the subsequent decades.
The federal government creating data center capacity, at scale, with electrical, water, and network hookups, feels very similar. Or semiconductor manufacture. Or recapitalizing US shipyards.
It might be AI today, something else tomorrow. But there will always be a something else.
Honestly, the biggest missed opportunity was supporting the Blount Island nuclear reactor mass production facility [1]. That was a perfect opportunity for government investment to smooth out market demand spikes. Mass deployed US nuclear in 1980 would have been a game changer.
They are trying. Microsoft wants to star the 3 Mile Island reactor. And other companies have been signing contracts for small modular reactors. SMRs are a perfect fit for modern data centers IF they can be made cheaply enough.
Wind, solar, and gas are all significantly cheaper in Texas, and can be brought online much quicker. Of course it wouldn't hurt to also build in some redundancy with nuclear, but I believe it when I see it, so far there's been lots of talk and little success in new reactors outside of China.
just as likely to be natural gas or a combination of gas and solar. I don't know what supply chain looks like for solar panels, but I know gas can be done quickly [1], which is how this money has to be spent if they want to reach their target of 125 billion a year.
The companies said they will develop land controlled by Wise Asset to provide on-site natural gas power plant solutions that can be quickly deployed to meet demand in the ERCOT.
The two firms are currently working to develop more than 3,000 acres in the Dallas-Fort Worth region of Texas, with availability as soon as 2027
There have been literally 0 production SMR deployments to date so there’s no possibility they’re basing any of their plans on the availability of them.
Hasn't the US decided to prefer nuclear and fossil fuels (most expensive generation methods) over renewables (least expensive generation methods)?[1][2]
I doubt the US choice of energy generation is ideological as much a practicality. China absolutely dominates renewables with 80% of solar PV modules manufactured in China and 95% of wafers manufactured in China.[3] China installed a world record 277GW of new solar PV generation in 2024 which was a 45% year-on-year increase.[4] By contract, the US only installed ~1/10th this capacity in 2024 with only 14GW of solar PV generation installed in the first half of 2024.[5]
Notably it is significantly more than the revenue of either of AWS or Azure. It is very comparable to the sum of both, but consolidated into the continental US instead distributed globally.
Small or modular reactors in the US are more than 10 years away, probably more like 15-20. These are facts and not made-up political or pipe-dreaming techno-snobes.
> Small or modular reactors in the US are more than 10 years away, probably more like 15-20
Could be 5 to 10 with $20+ bn/year in scale and research spend.
Trump is screwing over his China hawks. The anti-China and pro-nuclear lobbies have significant overlap; this could be how Trump keeps e.g. Peter Thiel from going thermonuclear on him.
I work in the sector and it's impossible to build a full-sized reactor in less than 10 years, and the usual over-run is 5 years. That's the time for tried and tested designs. The tech isn't there yet, and there are no working analogs in the US to use as an approved guide. The Department of Energy does not allow "off-the-cuff" designs for reactors. I think there is only two SMRs that have been built, one by the Russians and the other by China. I'm not sure they are fully functioning, or at least working as expected. I know there are going to be more small gas gens built in the near future and that SMRs in the US are way off.
Guessing SMRs are a ways off, any thoughts on the container-sized microreactors that would stand in for large diesel gens? My impression is that they’re still in the design phase, and the supply chain for the 20% U-235 HALEU fuel is in its infancy, but this is just based on some cursory research. I like the prospect of mass manufacturing and servicing those in a centralized location versus the challenges of building, staffing, and maintaining a series of one-off megaprojects, though.
> it's impossible to build a full-sized reactor in less than 10 years, and the usual over-run is 5 years
I'm curious why that is. If we know how to build it, it shouldn't take that long. It's not like we need to move a massive amount of earth or pour a humongous amount of concrete or anything like that, which would actually take time. Then why does it take 15 years to build a reactor with a design that is already tried and tested and approved?
When you're the biggest fossil fuel producer in the world, it's vital that you stay laser-focused on regulating nuclear power to death in every imaginable detail while you ignore the vast problems with unchecked carbon emissions and gaslight anyone who points them out.
If you didn't intend your comment to be a snarky one-liner, that didn't come across to me, and I'm pretty sure that would also be the case for many others.
Intent is a funny thing—people usually assume that good intent is sufficient because it's obvious to themselves, but the rest of us don't have access to that state, so has to be encoded somehow in your actual comment in order to get communicated. I sometimes put it this way: the burden is on the commenter to disambiguate. https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
I take your point at least halfway though, because it wasn't the worst violation of the guidelines. (Usually I say "this is not a borderline case" but this time it was!) I'm sensitive to regional flamewar because it's tedious and, unlike national flamewar or religious flamewar, it tends to sneak up on people (i.e. we don't realize we're doing it).
So you are sorry and take it back? Should probably delete your comments rather than striking them out, as the guidelines say.
I live, work, and posted this from Texas, BTW...
Also it takes up more than one line on my screen. So, not a "one-liner" either. If you think it is, please follow the rules consistently and enforce them by deleting all comments on the site containing one sentence or even paragraph. My comment was a pretty long sentence (136 chars) and wouldn't come close to fitting in the 50 characters of a Git "one-liner".
Otherwise, people will just assume all the comments are filtered through your unpredictable and unfairly biased eye. And like I said (and you didn't answer), this kind of thing is no longer in fashion, right?
None of this is "borderline". I did nothing wrong and you publicly shamed me. Think before you start flamewars on HN. Bad mod.
How much capacity does solar and wind add compared to nuclear, per square foot of land used? Also I thought the new administration was placing a ban on new renewable installations.
The ban is on offshore wind and for government loans for renewables. Won't really affect Texas much, it's Massachusetts that'll have to deal with more expensive energy.
Is the new rail you’re talking about the brightline?
It pretty much exclusively goes to and from tourist centers and is far too expensive ($40-$60 per seat each way) to deter most residents from just driving to Orlando. I wouldn’t really call it infrastructure (like the tri rail is.)
Not to mention that it’s the deadliest train in the US. People here barely follow traffic laws, but when you have it passing through major foot traffic areas every hour like on antlantic avenue in Delray Beach, people are going to get hit.
Grid is fine, snow is melting, everything is business as usual. CenterPoint had 99.9% deliverability for the past 24 hours, and ERCOT has 14,781 MW in reserve power available (https://www.ercot.com/gridmktinfo/dashboards/gridconditions). Source: I live in Houston.
I know this was tongue in cheek, but c'mon, we can respect each other, right? :)
No, he didn't. Both links have zero mentions of solar or wind, and very specifically define their terms thus:
"a) The term “energy” or “energy resources” means crude oil, natural gas, lease condensates, natural gas liquids, refined petroleum products, uranium, coal, biofuels, geothermal heat, the kinetic movement of flowing water, and critical minerals, as defined by 30 U.S.C. 1606 (a)(3)."
This is Trump accepting bribes from the legacy and fossil fuel industries to keep those "nasty" new clean energy sources from competing with them.
I'm confused and a bit disturbed; honestly having a very difficult time internalizing and processing this information. This announcement is making me wonder if I'm poorly calibrated on the current progress of AI development and the potential path forward. Is the key idea here that current AI development has figured out enough to brute force a path towards AGI? Or I guess the alternative is that they expect to figure it out in the next 4 years...
I don't know how to make sense of this level of investment. I feel that I lack the proper conceptual framework to make sense of the purchasing power of half a trillion USD in this context.
Let me avoid the use of the word AGI here because the term is a little too loaded for me these days.
1) reasoning capabilities in latest models are rapidly approaching superhuman levels and continue to scale with compute.
2) intelligence at a certain level is easier to achieve algorithmically when the hardware improves. There's also a larger path to intelligence and often simpler mechanisms
3) most current generation reasoning AI models leverage test time compute and RL in training--both of which can make use of more compute readily. For example RL on coding against compilers proofs against verifiers.
All of this points to compute now being basically the only bottleneck to massively superhuman AIs in domains like math and coding--rest no comment (idk what superhuman is in a domain with no objective evals)
It is superhuman in a very specific domain. I didn't use AGI because its definitions are one of two flavors.
One, capable of replacing some large proportion of global gdp (this definition has a lot of obstructions: organizational, bureaucratic, robotic)...
two, difficult to find problems in which average human can solve but model cannot. The problem with this definition is that the distinct nature of intelligence of AI and the broadness of tasks is such that this metric is probably only achievable after AI is already in reality massively superhuman intelligence in aggregate. Compare this with Go AIs which were massively superhuman and often still failing to count ladders correctly--which was also fixed by more scaling.
All in all I avoid the term AGI because for me AGI is comparing average intelligence on broad tasks rel humans and I'm already not sure if it's achieved by current models whereas superhuman research math is clearly not achieved because humans are still making all of progress of new results.
> All of this points to compute now being basically the only bottleneck to massively superhuman AIs
This is true for brute force algorithms as well and has been known for decades. With infinite compute, you can achieve wonders. But the problem lies in diminishing returns[1][2], and it seems things do not scale linearly, at least for transformers.
> Is the key idea here that current AI development has figured out enough to brute force a path towards AGI?
My sense anecdotally from within the space is yes people are feeling like we most likely have a "straight shot" to AGI now. Progress has been insane over the last few years but there's been this lurking worry around signs that the pre-training scaling paradigm has diminishing returns.
What recent outputs like o1, o3, DeepSeek-R1 are showing is that that's fine, we now have a new paradigm around test-time compute. For various reasons people think this is going to be more scalable and not run into the kind of data issues you'd get with a pre-training paradigm.
You can definitely debate on whether that's true or not but this is the first time I've been really seeing people think we've cracked "it", and the rest is scaling, better training etc.
Largest GPU cluster at the moment is X.ai's 100K H100's which is ~$2.5B worth of GPUs. So, something 10x bigger (1M GPUs) is $25B, and add $10B for 1GW nuclear reactor.
This sort of $100-500B budget doesn't sound like training cluster money, more like anticipating massive industry uptake and multiple datacenters running inference (with all of corporate America's data sitting in the cloud).
It seems you'd need to figure periodic updates into the operating cost of a large cluster, as well as replacing failed GPUs - they only last a few years if run continuously.
I've read that some datacenters run mixed generation GPUs - just updating some at a time, but not sure if they all do that.
It'd be interesting to read something about how updates are typically managed/scheduled.
> current AI development has figured out enough to brute force a path towards AGI? Or I guess the alternative is that they expect to figure it out in the next 4 years...
Or they think the odds are high enough that the gamble makes sense. Even if they think it's a 20% chance, their competitors are investing at this scale, their only real options are keep up or drop out.
"There are maybe a few hundred people in the world who viscerally understand what's coming. Most are at DeepMind / OpenAI / Anthropic / X but some are on the outside. You have to be able to forecast the aggregate effect of rapid algorithmic improvement, aggressive investment in building RL environments for iterative self-improvement, and many tens of billions already committed to building data centers. Either we're all wrong, or everything is about to change." - Vedant Misra, Deepmind Researcher.
Maybe your calibration isn't poor. Maybe they really are all wrong but there's a tendency here to these these people behind the scenes are all charlatans, fueling hype without equal substance hoping to make a quick buck before it all comes crashing down, but i don't think that's true at all. I think these people really genuinely believe they're going to get there. And if you genuinely think that, them this kind of investment isn't so crazy.
this is an announcement not a cut check. Who knows how much they'll actually spend, plenty of projects never get started let alone massive inter-company endeavors.
I think the only way you get to that kind of budget is by assuming that the models are like 5 or 10 times larger than most LLMs, and that you want to be able to do a lot of training runs simultaneously and quickly, AND build the power stations into the facilities at the same time. Maybe they are video or multimodal models that have text and image generation grounded in a ton of video data which eats a lot of VRAM.
It's a typical Trump-style announcement -- IT'S GONNA BE HUUUGE!! -- without any real substance or solid commitments
Remember Trump's BIG WIN of Foxconn investing $10B to build a factory in Wisconsin, creating 13000 jobs?
That was in 2017. 7 years later, it's employing about 1000 people if that. Not really clear what, if anything, is being made at the partially-built factory. [0]
> Is the key idea here that current AI development has figured out enough to brute force a path towards AGI? Or I guess the alternative is that they expect to figure it out in the next 4 years...
Can't answer that question, but, if the only thing to change in the next four years was that generation got cheaper and cheaper, we haven't even begun to understand the transformative power of what we have available today. I think we've felt like 5-10% of the effects that integrating today's technology can bring, especially if generation costs come down to maybe 1% of what they currently are, and latency of the big models becomes close to instantaneous.
This announcement is from the same office as the guy that xeeted:
“My NEW Official Trump Meme is HERE! It's time to celebrate everything we stand for: WINNING! Join my very special Trump Community. GET YOUR $TRUMP NOW.”
Your calibration is probably fine, stargate is not a means to achieve AGI, it’s a means to start construction on a few million square feet of datacenters thereby “reindustrializing America”
FWIW Altman sees it as a way to deploy AGI. He's increasingly comfortable with the idea they have achieved AGI and are moving toward Artificial Super Intelligence (ASI).
Do you think Sam Altman ever sits in front of a terminal trying to figure out just the right prompt incantation to get an answer that, unless you already know the answer, has to be verified? Serious question. I personally doubt he is using openai products day to day. Seems like all of this is very premature. But, if there are gains to be made from a 7T parameter model, or if there is huge adoption, maybe it will be worth it. I'm sure there will be use for increased compute in general, but that's a lot of capex to recover.
twitter hype is out of control again.
we are not gonna deploy AGI next month, nor have we built it.
we have some very cool stuff for you but pls chill and cut your expectations 100x!
I realize he wrote a fairly goofy blog a few weeks ago, but this tweet is unambiguous: they have not achieved AGI.
Where are they getting the $500B? Softbank's market cap is 84b and their entire vision fund is only $100b, Oracle only has $11b cash on hand, OpenAI's only raised $17b total...
That's their total fund and I doubt they are going all in with it in the US. Still, to reach $500bn, they need $125bn every single year. I think they just put down the numbers they want to "see" invested and now they'll be looking for backers. I don't think this is going anywhere really.
This would be a large outlay even for UAE, who would be giving it to a direct competitor in the space: UAE is one of the few countries outside of the US who are in any way serious about AI.
there doesn't appear to be any timeline announced here. the article says the "initial investment" is expected to be $100bn, but even that doesn't mean $100bn this year.
if this is part of softbank's existing plan to invest $100bn in ai over the next four years, then all that's being announced here is that Sama and Larry Ellison wanted to stand on a stage beside trump and remind people about it.
Softbank is being granted a block of TRUMP MEMES, the price of which will skyrocket when they are included in the bucket of crypto assets purchased as part of the crypto reserve.
>> Where are they getting the $500B? Softbank's market cap is 84b and their entire vision fund is only $100b, Oracle only has $11b cash on hand, OpenAI's only raised $17b total...
1. The outlays can be over many years.
2. They can raise debt. People will happily invest at modest yields.
Oracle's cash on hand is presumably irrelevant- I think they are on the receiving end of the money, in return for servers. No wonder Larry Ellison was so fawning.
Is this is a good investment by Softbank? Who knows.. they did invest in Uber, but also have many bad investments.
The moon program was $318 billion in 2023 dollars, this one is $500 billion. So that's why the tech barons who were present at the inauguration were high as a kite yesterday, they just got the financing for a real moon shot!
> Other partners in the project include Microsoft, investor MGX and the chipmakers Arm and NVIDIA, according to separate statements by Oracle and OpenAI.
It appears this basically locks out Google, Amazon and Meta. Why are we declaring OpenAI as the winner? This is like declaring Netscape the winner before the dust settled. Having the govt involved in this manner can’t be a good thing.
Since the CEOs of Google, Amazon and Meta were seated at the front row of the inauguration, IN FRONT OF the incoming cabinet, I'm pretty confident their techno -power-barrel will come via other channels.
Interestingly, there seems to be no actual government involvement aside from the announcement taking place at the White House. It all seems to be private money.
Government enforcing or laxing/fast tracking regulations and permits can kill or propel even a 100B project, and thus can be thought as having its own value on the scale of the given project’s monetary investment, especially in the case of a will/favor/whim-based government instead of a hard rules based deep state one.
Isn't that a state and local-level thing, though? I can't imagine that there is much federal permitting in building a data center, unless it is powered by a nuclear reactor.
I generally agree that government sponsorship of this could be bad for competition. But Google in particular doesn't necessarily need outside investment to compete with this. They're vertically integrated in AI datacenters and they don't have to pay Nvidia.
They don't have to spend $500B to compete. Their costs should be much lower.
That said, I don't think they have the courage to invest even the lower amount that it would take to compete with this. But it's not clear if it's truly necessary either, as DeepSeek is proving that you don't need a billion to get to the frontier. For all we know we might all be running AGI locally on our gaming PCs in a few years' time. I'm glad I'm not the one writing the checks here.
This seems to be getting lost in the noise in the stampede for infrastructure funding
Deepseek v3 at $5.5M on compute and now r1 a few weeks later hitting o1 benchmark scores with a fraction of the engineers etc. involved ... and open source
We know model prep/training compute has potentially peaked for now ... with some smaller models starting to perform very well as inference improves by the week
Unless some new RL concept is going to require vastly more compute for a run at AGI soon ... it's possible the capacity being built based on an extrapolation of 2024 numbers will exceed the 2025 actuals
Also, can see many enterprises wanting to run on-prem -- at least initially
They’re a big company. You could tell a story that they’re less efficient than OpenAI and Nvidia and therefore need more than $500b to compete! Who knows?
Probably not popular opinion - but I actually think Google is winning this now. Deep research is the most useful AI product I have used (Claud is significantly more useful than openAI)
I am not sure if OpenAI will be the winner despite this investment. Currently, I see various DeepSeek AI models as offering much more bang for the buck at a vastly cheaper cost for small tasks, but not yet for large context tasks.
How involved is the government at all? I’m still having a hard time seeing how Trump or anyone in the government is involved except to do the announcement. These are private companies coming together to do a deal.
The actual press release makes it clearer that this isn't a lockout of any kind and there's no direct government involvement. Softbank and some of other banks persuaded by Softbank are ponying up $500B for OpenAI to invest in AI. Trump is hyping this up from the sidelines because "OpenAI says this will be good for America". It's basically just another day in the world of press-releases and political pundits commenting on press-releases.
I hear this joked about sometimes or used as a metaphor, but in the literal sense of the phrase, are we in a cold war right now? These types of dollars feel "defense-y", if that makes sense. Especially with the big focus on energy, whatever that ends up meaning. Defense as a motivation can get a lot done very fast so it will be interesting to watch, though it raises the hair on my arms
Right, but they've been doing that for a while, to everyone. The US is much quieter about it, right? But you can twist this move and see how the gov would not want to display that level of investment within itself as it could be interpreted as a sign of aggression. but it makes sense to me that they'd have no issue working through corporations to achieve the same ends but now able to deny direct involvement
I don't think this administration is worried too much about showing aggression. If anything they are embracing it. Today was the first full day, and they have already threatened the sovereignty of at least four nations.
You know those booths at events where money is blown around and the person inside needs to grab as much as they can before the timer runs out? This is that machine for technologists until the bubble ends. The fallout in 2-3 years is the problem of whomever invested or is holding bags when (if?) the bubble pops.
That was literally my question. Is this basically just for more datacenters, NVidia chips, and electricity with a sprinkling of engineers to run it all? If so, then that $500bn should NOT be invested in today's tech, but instead in making more powerful and power efficient chips, IMO.
Nvidia and TSMC are already working on more powerful and efficient chips, but the physical limits to scaling mean lots more power is going to be used in each new generation of chips. They might improve by offering specific features such as FP4, but Moore's law is still dead.
$500bn of usefully deployed engineering, mostly software, seems like it would put AMD far ahead of Nvidia. Actually usefully deploying large amounts of money is not so easy, though, and this would still go through TSMC.
I'll make a wild guess that they will be building data centers and maybe robotic labs. They are starting with 100B of committed by mostly Softbank, but probably not transacted yet, money.
> building new AI infrastructure for OpenAI in the United States
The carrot is probably something like - we will build enough compute to make a supper intelligence that will solve all the problems, ???, profit.
If we look at the processing requirements in nature, I think that the main trend in AI going forward is going to be doing more with less, not doing less with more, as the current scaling is going.
Thermodynamic neural networks may also basically turn everything on its ear, especially if we figure out how to scale them like NAND flash.
If anything, I would estimate that this is a space-race type effort to “win” the AI “wars”. In the short term, it might work. In the long term, it’s probably going to result in a massive glut in accelerated data center capacity.
The trend of technology is towards doing better than natural processes, not doing it 100000x less efficiently. I don’t think AI will be an exception.
If we look at what is -theoretically- possible using thermodynamic wells, with current model architectures, for instance, we could (theoretically) make a network that applies 1t parameters in something like 1cm2. It would use about 20watts, back of the napkin, and be able to generate a few thousand T/S.
Operational thermodynamic wells have already been demonstrated en silica. There are scaling challenges, cooling requirements, etc but AFAIK no theoretical roadblocks to scaling.
Obviously, the theoretical doesn’t translate to results, but it does correlate strongly with the trend.
So the real question is, what can we build that can only be done if there are hundreds of millions of NVIDIA GPUs sitting around idle in ten years? Or alternatively, if those systems are depreciated and available on secondary markets?
Reasonably speaking, there is no way they can know how they plan to invest $500 billion dollars. The current generation of large language models basically use all human text thats ever been created for the parameters... not really sure where you go after than using the same tech.
That's not really true - the current generation, as in "of the last three months", uses reinforcement learning to synthesize new training data for themselves: https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero
Right but that's kind of the point: there's no way forward which could benefit from "moar data". In fact it's weird we need so much data now - i.e. my son in learning to talk hardly needs to have read the complete works of Shakespeare.
If it's possible to produce intelligence from just ingesting text, then current tech companies have all the data they need from their initial scrapes of the internet. They don't need more. That's different to keeping models up to date on current affairs.
> Notably, it is the first open research to validate that reasoning capabilities of LLMs can be incentivized purely through RL, without the need for SFT.
It seems to me you could generate a lot of fresh information from running every youtube video, every hour of TV on archive.org, every movie on the pirate bay -- do scene by scene image captioning + high quality whisper transcriptions (not whatever junk auto-transcription YouTube has applied), and use that to produce screenplays of everything anyone has ever seen.
I'm not sure why I've never heard of this being done, it would be a good use of GPUs in between training runs.
The fact that OpenAI can just scrape all of Youtube and Google isn't even taking legal action or attempting to stop it is wild to me. Is Google just asleep?
what are they going to use to sue - DMCA? OpenAI (and others) are scraping everything imaginable (MS is scraping private Github repos…) - don’t think anyone in the current government will be regulating any of this anytime soon
Such a biased source of data-that gets them all the LaTeX source for my homeworks, but not my professor's grading of the homework, and not the invaluable words I get from my professor at office hours. No wonder the LLMs have bizarre blindnesses in different directions.
> a lot of fresh information from running every youtube video
EVERY youtube video?? Even the 9/11 truther videos? Sandy Hook conspiracy videos? Flat earth? Even the blatantly racist? This would be some bad training data without some pruning.
The best videos would be those where you accidentally start recording and you get 2 hours of naturalistic conversation between real people in reality. Not sure how often they are uploaded to YouTube.
Part of the reason that kids need less material is that the aren't just listening, they are also able to do experiments to see what works and what doesn't.
"create hundreds of thousands of American jobs"... Given the current educational system in the US, this should be fun to watch. Oh yeah, Musk and his H-1B Visa thing. Now it's making sense.
How does this work out in the long term? Operating a data center does not require that many blue-collar workers.
I'm imagining a future where the US builds a Tower of Babel from thousands of data centers just to keep people employed and occupied. Maybe also add in some paperclip factories¹?
This is what the 2024 Nobel prize winners in economics call "creative destruction" to repeat from their book Why Nations Fail. They really did not have a lot of sympathy for those they lumped in with Luddites who were collateral damage to progress.
After they build the Multivac or Deep Thought, or whatever it is they’re trying to do, then what happens? It makes all the stockholders a lot of money?
The way I think about this project, along with all of Trump's plans, is that he wants to maximize the US's economic output to ensure we are competitive with China in the future.
Yes, it would make money for stockholders. But it's much more than that: it's an empire-scale psychological game for leverage in the future.
> he wants to maximize the US's economic output to ensure we are competitive with China in the future.
LOL
Under Trump policies, China will win "in the future" on energy and protein production alone.
Once we've speedrunned our petro supply and exhausted our agricultural inputs with unfathomably inefficient protein production, China can sit back and watch us crumble under our own starvation.
No conflict necessary under these policies, just patience! They're playing the game on a scale of centuries, we can't even stay focused on a single problem or opportunity for a few weeks.
> Once we've speedrunned our petro supply and exhausted our agricultural inputs with unfathomably inefficient protein production, China can sit back and watch us crumble under our own starvation.
China is the largest importer of crude oil in the world. China imports 59% of its oil consumptions, and 80% of food products. Meanwhile, US is fully self sufficient on both food and oil.
> They're playing the game on a scale of centuries
Is that why they are completely broke, having built enough ghost buildings that house entire population of France - 65 million vacant units? Is that why they are now isolated in geopolitics, having allied with Russia and pissed off all their neighbors and Europe?
China's oil reserve only lasts 80 days. In case of any conflict that disrupts oil import, China would be shutting down very quickly. Since you brought up crumble and starvation.
America can subject itself to domestic and international turmoil by invading as many allies as it wants. China's winning strategy is still to keep innovating on energy and protein at scale and wait for starvation while they build their soft power empire and America becomes a pariah state. They're in no rush at all.
Our military and political focus will be keeping neighbors out on one side and trying to seize land on the other side while China goes and builds infrastructure for the entire developing world that they'll exploit for centuries.
Is this a serious suggestion? America can just keep invading people ad infinitum instead of... applying slight thumb pressure on the market's scales to develop more efficient protein sources and more renewable fuel sources before we are staring at the last raw economic input we have?
China is dead broke and will shrink to 600M in population before 2100. State owned enterprises are eating up all the private enterprises. Meanwhile, Chinese rich leaves China by tens of thousands per year, and capital outflow increases every year.
Unfortunately one of those things that authoritarianism has a lot more methods to solve than other systems, which really underscores the importance of beating them in the long term.
Their current very advanced method, is to send village elders to couples and single guys and berate them on why they are not having sex or having kids (hint: no jobs and no money)
Things can always change, but today China is significantly more dependent on petrochemicals than the US. I'm not sure what you're referring to with regards to agriculture, both the US and China have strong food industries that produce plenty of foods containing protein.
In 2023 China had more net new solar capacity than the US has in total, and it will only climb from there. In order to do this, they're flexing muscles in R&D and mass production that the US has actually started to flex, and now will face extreme headwinds and decreased capital investment.
Regarding agriculture: America's agricultural powerhouse, California's Central Valley, is rapidly depleting its water supplies. The midwest is depleting its topsoil at double the rate that USDA considers sustainable.
None of this is irreversible or irrecoverable, but it very clearly requires some countervailing push on market forces. Market forces do not naturally operate on these types of time scales and repeatedly externalize costs to neighbors or future generations.
It sounds like those countervailing pushes are ongoing? The Nature article mentions how California passed regulatory reforms in 2014 to address the Central Valley water problem. The Smithsonian article describes how no-till practices to avoid topsoil depletion have been implemented by a majority of farmers in four major crops.
Uhhh I’m going to describe a specific case, but you can extrapolate this to just about every single sustainability initiative out there.
No-till farming has been significantly supported by the USDA’s programs like EQIP
During his first term, Trump pushed for a $325MM cut to EQIP. That's 20-25% of their funding and would have required cutting hundreds if not thousands of employees.
Even BEFORE these cuts (and whatever he does this time around), USDA already has to reject almost 75% of eligible EQIP applicants
Regarding CA’s water: Trump already signed an EO requiring more water be diverted from the San Joaquin Delta into the desert Central Valley to subsidize water-intensive crops. This water, by the way, is mostly sold to mega-corps at rates 98% below what nearby American consumers pay via their municipal water supplies, effectively eliminating the blaring sirens that say “don’t grow shit in the desert.”
Now copy-paste to every other mechanism by which we can increase our nation’s climate security and ta-da, you’ve discovered one of the major problems with Trumpism. It turns out politics do matter!
I think that coming down from 5T to 0.5T means that TSMC cannot be reproduced locally, but everything else is on the table. At least TSMC has a serious roadmap for its Arizona fab facility, so that too is domestically captured, although not its latest gen fab.
How have they already selected who gets this money? Usually the government announces a program and tries to be fair when allocating funds. Here they are just bankrolling an existing project. Interesting
> The new entity, Stargate, will start building out data centers and the electricity generation needed for the further development of the fast-evolving AI in Texas, according to the White House.
Wouldn't a more northern state be a better location given the average temperatures of the environment? I've heard Texas is hot!
To get out from under OpenAI’s considerable obligation to Microsoft.
That is why there is the awkward “we’ll continue to consume Azure” sentence in there. Will be interesting to see if it works or if MS starts revving up their lawyers.
Because tech CEOs have decided to go all-in on fascism as they see it's a way to make money. Bow to Trump, get on his good side, reap the benefits of government corruption.
It's why TikTok thanked Trump in their boot-licking message of "thanks, trump" after he was the one who started the TikTok ban.
A harder question is: why wouldn't billionaires like Trump and his oligarchic kleptocracy?
"SoftBank, OpenAI, Oracle, and MGX" seems like quite the lineup. Two groups who are good at frivolously throwing away investment money because they have so much capital to deploy, there really isn't anything reasonable to do with it, a tech "has-been" and OpenAI. You become who you surround yourself with I guess.
I'm sure they're getting tax credits for investment (none of the articles I can find actually detail the US gov involvement) but the project is mostly just a few multinationals setting up a datacenter where their customers are.
It seems early for this sort of move. This is also a huge spin on the whole thing that could throw a lot of people off.
Is there any planned future partnerships? Stargate implies something about movies and astronomy. Movies in particular have a lot of military influence, but not always.
So, what's the play? Help mankind or go after mankind?
If one is expecting to have an AGI breakthrough in the next few years, this is exactly the prepositioning move one would make to be able to maximally capitalize on that breakthrough.
From my perspective humanity has all breakthroughs in intelligence it needs.
The breaking of The Enigma gave humans machines that can spread knowledge to more humans. It already happened a long time ago, and all of it was cause for much trouble, but we endured the hardest part (to know when to stop), and humans live in a good world now. Full of problems, but way better than it was before.
I think the web is enough. LLMs are good enough.
This move to try to draw water from stone (artificial intelligence in sillicon chips) seems to be overkill. How can we be sure it's not a siphon that will make us dumber? Before you just dismiss me or counter my arguments, consider what is happening everywhere.
Maybe I'm wrong, or not seeing something. You know, like I believed in aliens for a long time. This move to artificial intelligence causes shock and awe in a similar way. However, while I do believe aliens do not exist, I am not sure if artificial intelligence is a real strawman. It could be the case that is not made of straw, and if it is more than that, we might have a problem.
I am specially concerned because unlike other polemic topics, this one could lead to something not human that fully understands those previous polemic topics. Humans through their generations forget and mythologize those fantasies. We don't know what non-humans could do with that information.
I am thinking about those issues for a long time. Almost a decade, even before LLMs running on silicon existed. If it wanted, non-human artificial intelligence could wipe the floor with humans just by playing to their favorite myths. Humans do it in a small scale. If machines learn it, we're in for an unknown hostile reality.
It could, for example, perceive time different from us (also a play on myths), and do all sorts of tricks with our minds.
LLMs and the current generation of artificial intelligence are boolean first, it's what they run. Only true or false bits and gates. Humans can understand the meaning of trulse though, we are very non boolean.
So, yeah, I am worried about booleaning people on a massive scale.
Not sure how they knew to buy them or why but they have them. Mostly seem to be lending them out. Think mostly OpenAI. Or was it MS. One of the big dogs
Still, the worst positioned cloud provider to tackle this job. Both for the project and for eventual users of whatever eldritch abomination that cames out of this.
Texas already is the leading state in new grid battery and grid solar installs for the last 3 years. Governor Abbott also did nuclear deregulation last year.
> building new AI infrastructure for OpenAI in the United States
That's nice, but if I were spending $500bn on datacenters I'd probably try to put a few in places that serve other users. Centralised compute can only get you so far in terms of serving users.
Some reports[0] paint this as something Trump announced and that the US Government is heavily involved with but the announcement only mentions private sector (and lead by Japan's Softbank at that). Is the US also putting in money? How much control of the venture is private vs public here?
You probably still need to train the initial models in data centers, with local host mostly being used to run train models. At most we’d augment trained models with local data storage on local host.
If compute continues to become cheaper, local training might be feasible in 20 years.
You definitely still need data centers to train the models that you’ll run locally. Also if we achieve AGI you can bet it won’t be available to run locally at first.
Why are corporations announcing business deals from the White House? There doesn’t seem to be any public ownership/benefit here, aside from potential job creation. Which could be significant. But the American public doesn’t seem to gain anything from this new company.
This isn't an overseas trip though. It's a private partnership announced by the sitting president in the Roosevelt room, literally across the hall from the oval office. I don't know how unprecedented that truly is, but it certainly feels unusual.
It will. The short-term sale is that it will create thousands of temporary jobs, and long-term reduce hundreds of thousands of jobs, while handing the savings to stock holdings and moving wealth to the stockholders.
Looks on pace to eliminate every human job over 10 years.
What is the hard limiting factor constraining software and robots from replacing any human job in that time span? Lots of limitations of current technology, but all seem likely to be solved within that timeframe.
>> Ingka says it has trained 8,500 call centre workers as interior design advisers since 2021, while Billie - launched the same year with a name inspired by IKEA's Billy bookcase range - has handled 47% of customers' queries to call centres over the past two years.
Do you expect all companies to retrain? Do you expect CEOs to be wrong? Do you expect AI to stay the same, get better, or get worse? I never made the claim that new jobs will NOT be made, that is yet to be seen, but jobs will be lost to AI.
>> “For a company like BT there is a huge opportunity to use AI to be more efficient,” he said. “There is a sort of 10,000 reduction from that sort of automated digitisation, we will be a huge beneficiary of AI. I believe generative AI is a huge leap forward; yes, we have to be careful, but it is a massive change.”
If the announced spending target is true, this will be a strategic project for the US exceeding Biden's stimulus acts in scale. I think it would be pretty normal in any country to have highest-level involvement for projects like this. For example, Tesla has a much smaller revenue than this and Chancellor Olaf Scholz was still present when they opened their Gigafactory near Berlin.
This is my question too, but I haven't seen a journalist ask it yet. My baseless theory: Trump has promised them some kind of antitrust protections in the form of legislation to be written & passed at a later date.
An announcement of a public AI infrastructure program joined by multiple companies could have been a monumental announcement. This one just looks like three big companies getting permission to make one big one.
Easier: Trump likely committed that the federal agencies wouldn't slow roll regulatory approval (for power, for EIS, etc.).
Ellison stated explicitly that this would be "impossible" without Trump.
Masa stated that this (new investment level?) wouldn't be happening had Trump not won, and that the new investment level was decided yesterday.
I know everyone wants to see something nefarious here, but simplest explanation is that the federal government for next four years is expected to be significantly less hostile to private investment, and - shocker - that yields increased private investment.
That is a better one. I don't know why three rich guys investing in a new company would result in a slowness that Trump could fix, though, and a promise to rush or sidestep regulatory approval still sounds nefarious.
It made me laugh when Sam said "I'm thrilled that we get to do this in the United States of America", I shouted at the TV 'Yeah you almost had to do it in Saudi Arabia' !!
I hope the Japanese government demands seismic isolation for Softbank, otherwise it will be the Japanese citizens who have to foot the bill when this hype hits the ground and shakes hard the Japanese economy :/
Softbank should not be allowed to invest more than ARM Holdings sold at a loss.
I put the word "some" in front of "crypto" for a reason.
There is some crypto that we know how to break with a sufficiently large quantum computer [0]. There is some we don't know how to do that to. I might be behind the state of the art here, but when I wasn't we specifically really only knew how to use it to break cryptography that Shor's algorithm breaks.
Nope. Any crypto you can break with a real, physical, non-imaginary quantum computer, you can break faster with classical. Get over it. Shor's don't run yet and probably never will.
You are misdirecting and you know it. I don't even need to discredit that paper. Other people have done it for me already.
This is like asking whether $500 billion to fund warp drives would yield better returns.
Money can't buy fundamental breakthroughs: money buys you parallel experimental volume - i.e. more people working from the same knowledge base, and presumably an increase in the chance that one of them does advance the field. But at any given time point, everyone is working from the same baseline (money also can improve this - by funding things you can ensure knowledge is distributed more evenly so everyone is working at the state of the art, rather then playing catch up in proprietary silos).
Leading state in new grid battery and grid solar installations for the last three years, and deregulated nuclear power last year. Abilene is near the Dallas Fort-Worth Metroplex area which has a massive 8M+ upper-income population highly skilled in hardware and electrical engineering (Texas Instruments, Raytheon, Toyota, etc). The entire area has massive tracts of open land that are affordably priced without building restrictions. Business regulations and tax environment at the state and city level are very laissez faire (no taxes on construction such as in the Seattle area or many parts of California).
I could see DFW being a good candidate for a prototype arcology project.
Like dwnw said, anything goes in Texas if you have money and there’s already a decent number of qualified tech workers. Corporate taxes are super low as well.
ChatGPT may be better than Google Search in content but at end of day, you have to make money and last report I saw, ChatGPT is burning through money at prestigious rate.
> Technology advancing more quickly year over year?
> That’s a crazy notion and I’ll be sure everyone knows.
The version I heard from an economist was something akin to a second industrial revolution, where the pace of technological development increases permanently. Imagine a transition from Moore's law-style doubling every year and a half, to doubling every week and a half. That wouldn't be a true "singularity" (nothing would be infinite), but it would be a radical change to our lives.
> The pace of technological development has always been permanently increasing.
Not in the same way though. The pace of technological development post-industrial-revolution increased a lot faster - technological development was exponential both before and after, but it went from exponential with a doubling time of maybe a century, to a Moore's law style regime where the doubling time is a couple of years. Arguably the development of agriculture was a similar phase change. So the point is to imagine another phase change on the same scale.
You know, I expected that they'd find or synthesize some naquadah to build an actual stargate and maybe even defeat the Goa'uld. The exciting stuff, not AI.
It was rumoured in early 2024 that "Stargate" was planned to require 5GW data centre capacity[1][2] which in early 2024 was the entire data centre capacity Microsoft had already built[3]. Data centre capacity costs between USD$9-15m/MW[6] so 5GW of new data centre capacity would cost USD$45b-$75b but let's pick a more median cost of USD12m/MW[6] to arrive at USD$60b for 5GW of new data centre capacity.
This 5GW data centre capacity very roughly equates to 350000x NVIDIA DGX B200 (with 14.3kW maximum power consumption[4] and USD$500k price tag[5]) which if NVIDIA were selected would result in a very approximate total procurement of USD$175b from NVIDIA.
On top of the empty data centres and DGX B200's and in the remaining (potential) USD$265b we have to add:
* Networking equipment / fibre network builds between data centres.
* Engineering / software development / research and development across 4 years to design, build and be able to use the newly built infrastructure. This was estimated in mid 2024 to cost OpenAI US$1.5b/yr for retaining 1500 employees, or USD$1m/yr/employee[7]. Obviously this is a fraction of the total workforce needed to design and build out all the additional infrastructure that Microsoft, Oracle, etc would have to deliver.
* Electricity supply costs for current/initial operation. As an aside, these costs seemingly not be competitive with other global competitors if the USA decides to avoid the cheapest method of generation (renewables) and instead prefer the more expensive generation methods (nuclear, fossil fuels). It is however worth noting that China currently has ~80% of solar PV module manufacturing capacity and ~95% of wafer manufacturing capacity.[10]
* Costs for obtaining training data.
* Obsolescence management (4 years is a long time after which equipment will likely need to be completely replaced due to obsolescence).
* Any other current and ongoing costs of Microsoft, Oracle and OpenAI that they'll likely roll into the total announced amount to make it sound more impressive. As an example this could include R&D and sustainment costs in corporate ICT infrastructure and shared services such as authentication and security monitoring systems.
The question we can then turn to is whether this rate of spend can actually be achieved in 4 years?
Microsoft is planning to spend USD$80bn building data centres in 2025[7] with 1.5GW of new capacity to be added in the first six months of 2025[3]. This USD$80bn planned spend is for more than "Stargate" and would include all their other business units that require data centres to be built, so the total required spend of USD$45b-$75b to add 5GW data centre capacity is unlikely to be achieved quickly by Microsoft alone, hence the apparent reason for Oracle's involvement. However, Oracle are only planning a US$10b capital expenditure in 2025 equating to ~0.8GW capacity expansion[9]. The data centre builds will be schedule critical for the "Stargate" project because equipment can't be installed and turned on and large models trained (a lengthy activity) until data centres exist. And data centre builds are heavily dependent on electricity generation and transmission expansion which is slow to expand.
There's a good amount of irony in the results that AI have achieved, particularly if we reach AGI - they have improved individual worker efficiency by removing other workers from the system. Naming it Stargate implies a reckoning with the actual series itself - an accomplishment by humanity. Instead, what this pushes, is accomplishing the removal of humans from humanity. I like cool shiny tech, but I like useful tech that really helps humans more. Work on 3D-printing sustainable food, or something actually useful like that. Jenson doesn't need another 1B gallons of water under his belt.
> Instead, what this pushes, is accomplishing the removal of humans from humanity.
If you buy the marketing, yeah. But we aren't really seeing that in the tech sector. We haven't seen it succeed in the entertainment sector... it's still fighting for relevance in the medical and defense industries too. The number and quality of jobs that AI replaced is probably still quite low, and it will probably remain that way even after Stargate.
AI is DOA. LLMs have no successor, and the transformer architecture hit it's bathtub curve years ago.
> Jenson doesn't need another 1B gallons of water under his belt.
Jensen gets what he wants because he works with the industry. It's funny to see people object to CUDA and Nvidia's dominance but then refuse to suggest an alternative. An open standard managed by an independent and unbiased third-party? We tried that, OEMs abandoned it. NPU hardware tailor-made for specific inference tasks? Too slow, too niche, too often ends up as wasted silicon. Alternative manufacturer-specific SDKs integrated with one high-level library? ONNX tried that and died in obscurity.
Nvidia got where they are today by doing exactly what AMD and Apple couldn't figure out. People give Jensen their water because it's wasted in anyone else's hands.
Uh, they invented multilatent attention and since the method for creating o1 was never published, they’re the only documented example of producing a model of comparable quality. They also demonstrated massive gains to the performance of smaller models through distillation of this model/these methods, so no, not really. I know this is the internet, but we should try to not just say things.
> The buildout is currently underway, starting in Texas, and we are evaluating potential sites across the country for more campuses as we finalize definitive agreements.
For those interested, it looks like Albany, NY (upstate NY) is very likely one of the next growth sites.
This could potentially trigger an AI arms race between the US and China. The standard has been set, lets see what China responds with. Either way, it will accelerate the arrival of ASI, which in my opinion is probably a good thing.
It will be similar to the space race between Soviet Union and US. And just like Soviet Union going broke and collapsing, China too will go even more broke and collapse.
Texas is the leading state in new grid batteries and grid solar for three years now. Also Governor Abbott deregulated nuclear last year. Sure there will be some new natural gas too, which is the least scary fossil fuel. They call it the "all of the above" approach to energy.
You'd really think that arguably the leader in generative AI could come up with a unique project name instead of ripping off something extant and irrelevant.
But then again that's their entire business, so I shouldn't be too surprised.
Personally I wish they invested in optical photonic computing, taking it out of the research labs. It can be so much more energy efficient and faster to run than GPUs and TPUs.
$500 billion is a lot of money even by US government standards. It's about the size of all the new spending in the 2021 bipartisan infrastructure bill.
The political will is trying to balance a large existing debt at increasing interest rates, a significant primary deficit even in a good economy, rising military threats from China, a strong Republican desire for tax cuts, extremely popular entitlement programs that no one wants to touch, and an aging population with a declining birthrate
Modern monetary systems function through two main channels: government spending and bank lending. Every dollar in circulation originates from one of these sources - either government fiscal operations (deficit spending) or bank credit creation through loans. This means all money is fundamentally based on debt, though "debt" has very different implications for a currency-issuing government versus private borrowers.
Government debt operates fundamentally differently from household debt since the government controls its own currency. As former Fed Chairman Alan Greenspan noted to Congress, the U.S. can always meet any obligation denominated in dollars since it can create them. The real constraints aren't financial but economic - inflation risk and the efficient allocation of real resources.
The key question then becomes one of political priorities and public understanding. If public opposition to beneficial government spending stems from misunderstanding how modern monetary systems work, then better education about these mechanisms could help advance important policy goals. The focus should be on managing real economic constraints rather than imaginary financial ones.
Yes, people hate inflation, because inflation creates a demand for more money! Inflation means there is not enough money for people. So why did prices go up, is it just because of fiscal spending?
The relationship between inflation and monetary policy is more complex than often portrayed. While recent inflation has created financial strain for many Americans, its root causes extend beyond simple money supply issues.
Recent data shows that corporate profit margins reached historic highs during the inflationary period of 2021-2022. For example, in Q2 2022, corporate profits as a percentage of GDP hit 15.5%, the highest level since the 1950s. This surge in corporate profits coincided with the aftermath of Trump's 2017 Tax Cuts and Jobs Act, which reduced the corporate tax rate from 35% to 21%. This tax reduction increased after-tax profits and may have given companies more flexibility to pursue aggressive pricing strategies.
Multiple factors contributed to inflation:
Supply chain disruptions created genuine scarcity in many sectors, particularly semiconductors, shipping, and raw materials
Demand surged as economies reopened post-pandemic
Many companies used these market conditions to implement price increases that exceeded their cost increases
The corporate tax environment created incentives for profit maximization over price stability
For instance, many large retailers reported both higher prices and expanded profit margins during this period. The Federal Reserve Bank of Kansas City found that roughly 40% of inflation in 2021 could be attributed to expanded profit margins rather than increased costs.
This pattern suggests that market concentration, pricing power, and tax policy played significant roles in inflation, alongside traditional monetary and supply-chain factors. Policy solutions should therefore address market structure, tax policy, and monetary policy to effectively manage inflation.
> This project will ... also provide a strategic capability to protect the national security of America and its allies.
> All of us look forward to continuing to build and develop ... AGI for the benefit of all of humanity.
Erm, so which one is it? It is amply demonstrable from events post WW2 that US+allies are quite far from benefiting all of humanity & in fact, in some cases, it assists an allied minority at an extreme cost to a condemned majority, for no discernable humanitarian reasons save for some perceived notion of "shared values".
God forbid anyone would invest $500,000,000,000 to create jobs. No no no. 500 billion to destroy them for "more efficiency" so the owner class can get richer.
No amount of money invested in infrastructure is going to solve the "garbage in, garbage out" problem with AI, and it looks like the AI companies have already stolen the vast majority of content that is possible to steal. So this is basically a massive gamble that some innovation is going to make AI do something better than faultily regurgitate its training data. I'm not seeing a corresponding investment which actually attempts to solve the "garbage in, garbage out" problem.
A fraction of this money invested in building homes would end the homelessness problem in the U.S.
I guess the one silver lining here is that when the likely collapse happens, we'll have more clean energy infrastructure to use for more useful things.
I'm also curious how a global leader in multimodal generative AI chose this particular image. Did they prompt a generator for a super messy impressionist painting of red construction cranes with visible brush strokes, distorted to the point of barely being able to discern what the image represents?
Considering Stargate's introduction and plan seems to be a super messy concept of impressions of ideas and very lacking in details, the picture makes a lot of sense. Let AI evangelists see the future in the fuzz; let AI pessimists see failure in the abstract; let investors see $$$ in their pockets.
For me it's watching a gay man grovel at the feet of one of the most anti-LGBT politicians, a day after Trump signed multiple executive orders that dehumanized Altman and the LGBT community. Every token thinks they're special until they're spent.
Trump was the first president to come into office supporting gay marriage. Trump only has a problem with the "t" part of the community and only in bathrooms and sports, not in general.
I watched the announcement live, I could have sworn that the softbank guy said "initial investment of 100 MILLION, we hope to EARN 500 BILLION by the end of your (Trumps) term"
Gave me a real "this is just smoke and mirrors hiding the fact that the white house is now a glory hole for Trump to enjoy" feel.
The Silicon-Valley bubble universe continues to introduce entropy that it feeds off of itself... Naming this Stargate when some of the largest effects AI has had is removing humans from processes to make other, fewer humans more efficient is emblematic of this hollow naming ethos - continuing to use the portal to shunt more and more humans out of the process that is humanity, with fairly reckless abandon. Who is Ra, and who is sending the nuke where, in this naming scheme? You decide.
> This project will not only support the re-industrialization of the United States but also provide a strategic capability to protect the national security of America and its allies.
> The initial equity funders in Stargate are SoftBank, OpenAI, Oracle, and MGX. SoftBank and OpenAI are the lead partners for Stargate, with SoftBank having financial responsibility and OpenAI having operational responsibility. Masayoshi Son will be the chairman.
I'm sorry, has SoftBank suddenly become an American company? I feel like I'm taking crazy pills reading this.
Japan companies were a threat just a couple weeks ago.
There is credible evidence that leads me to believe that (1) Nippon Steel Corporation, a corporation organized under the laws of Japan . . . might take action that threatens to impair the national security of the United States;
Sometimes the person writing the copy is writing it because they talk good, not because they are the biggest proponent of the idea.
Give a clever, articulate person a task to write about something they don't believe in and they will include the subtlest of barbs, weak praise, or both.
I thought this meant it was $500 billion in government money.
Some of these companies do have huge cash reserves they don't know what to do with so if it is $500 billion of private money, I am not going to complain.
I will believe it when I see it though and that this isn't a 100 billion in private money with a 400 billion dollar free US government put option for the "private" investors if things don't go perfect.
Texas has a .... unique energy market (literally! They don't connect to the national grid so they can avoid US Government regulations- that way it's not interstate commerce). Because of that spot prices fluctuate very wildly up and down, depending on the weather, demand, and their large quantity of renewables (Texas is good for solar and wind energy). When the weather is good for renewables they have very cheap electricity (lots of production and can't sell to anyone outside the state), when the weather is bad they can have incredibly expensive electricity (less production, can't buy from anyone outside the state). Larger markets, able to pull from larger pools of producers and consumers, just fluctuate less.
I know some bitcoin miners liked to be in Texas and basically worked as energy speculators: when electricity was cheap they would mine bitcoin, when it was expensive they shut down their plant- sometimes they even got paid by producers to shut-down their plant! I would bet that you could do a lot of that with AI training as well, given good checkpointing.
You wouldn't want to do inference there (which needs to be responsive and doesn't like 'oh this plant is going to shut down in one minute because a storm just came up') but for training it should be fine?
No state income tax, fewer regulations (zoning, environmental regulations) than other parts of the country, relatively cheap power, large existing industrial base. For skilled labor that last bit is important. Also one of the cheapest states wrt minimum wage (same as federal, nothing added), which is important for unskilled labor.
Depending on the part of the state, relatively low costs of living which is helpful if you don't like paying people much. Large areas that are relatively undeveloped or underdeveloped which can mean cheaper land.
Unfortunately that figure wouldn’t get everyone healthcare in the US. I agree though, it could be deployed for better use but someone needs to think of those poor desperate shareholders.
AGI is the magic bullet to all of humanity's problems for some people. There is no explanation of how AGI will accomplish such things, just a belief that it will.
It'll do this one, certainly. Maybe not the most cost-effective jobs program, but I expect there will be many construction and data centre jobs in the next few years.
Health care can’t be solved by just throwing a shitload of money at it. I’m skeptical about food production as well.
Now quality transportation (trains and metros that work), cheap & accessible housing, and cheap energy might be a good idea because IIUC those are just super large price tags
