We will soon read that US is investing 50b into AI infrastructure by means of Anthropic commitments. Both balance sheets line up, no money exchanges hands.
NVIDIA, Intel, AMD, and OpenAI are all already engaging in this type of behavior.
Several parties are writing the checks. It's a complex structure of future VC rounds, debt financing, and most importantly, massive cloud credit commitments from their investors like Amazon and Google
"$50 billion investment" is classic "press release capital," not actual cash in a bank account. Anthropic isn't sitting on a mountain of cash. This is most likely a multi-year "commitment," funded by future investment rounds, debt financing, and possibly large pre-payments for cloud capacity from their investors, and it's a signal to competitors and the government about their ambitions, not a real $50B expenditure happening today
> We still need to fix the problem with powering these Datacenters....
Not really. We need to insulate consumers from the market that is solving and will solve that problem. That's a financial engineering and policy problem. America is good at the first. We're bad at the second. That implies state and local initiatives should take the lead.
My proposal: one market for essential residential consumption, defined as the median household consumption per region [1]. (If you don't use your allocation, you should earn a rebate.) Above that, market price. Same for preferred commercial uses, e.g. retail and local government.
There should be a tenure element to power access. The reductio is a well-funded adversary should not be able to buy all the power in a region just because they are able to pay more.
Our utilities are generally regulated, and have some mandate to provide power to residents. If a datacenter creates a sudden dislocation in the demand for power that causes massive unplanned CapEx by the utility, what is the argument in favor of longtime residents having to pay for that CapEx?
Data centers notoriously do not bring much in the way of either jobs or tax base.
Meta's planned gigawatt-scale DC in Louisiana is projected to create ~500 jobs. If energy prices for the state increase by only 20%, that does not feel like something the people of the region would obviously want to do if given the choice.
I don't understand why the energy rates for residents should go up at all? Why isn't the data center paying for it? Even my personal bill gets more expensive per kwh the more I use, shouldn't a data center using exponentially more power just pay a higher per kwh rate, honestly to the point that they are making it cheaper for normal customers??
Georgia, where data center construction has been booming, has seen rate increases significantly higher than this. IIRC there have been 6 rate increases in the last 2 years, with a 24% increase in 2024 alone.
I don't believe any of this was part of the forecast of 10 years ago, so I would not trust a long-term forecast issued now.
I actually think we should have more power plants, and we should generate a lot more power.
At the same time, it is not obvious that utility systems in communities are equipped to respond well to actors who are 1) not local 2) high growth 3) demand rapid deployment 4) are willing to spend flagrantly to move fast.
Put another way: should OpenAI be allowed to pay (say) 10x the going rate[1] for electricity in a region in exchange for being given the right to consume 90% of the electricity generated in that region? What principles should govern this balance and the acceptable price changes? How do those principles extend to determining who should cover CapEx for expansion? Does Texas need the 57 permanent jobs in Abilene that Stargate will generate so badly that millions of Texans should subsidize the richest startup ever?
1 - this is a hypothetical, but OpenAI's Stargate project in Texas is projected to cost multiples of the total assets of any of the large multi-state power generators in the country. None of the utilities are setup to engage with a buyer like that.
If the existing power supply is sufficient to supply the A/C and lights for an area, but a new buyer comes in and offers 10% more $ to purchase all the power currently being generated, there exist also the options of 1) make the new customer pay a lot more to deliver power quickly or 2) simply deny the connection rights (presumably much easier to do when talking about gigawatt-scale requests).
This announcement isn't just for the construction companies; it's for the power companies.. Anthropic is basically saying we're going to build a factory the size of a nuclear plant here. If you start building it now, you'll have a guaranteed buyer in 5 years
Super large DC's yes. However, there is tons of what I call stranded power. Only need 1-2MW? Well, that's $30-100M worth of compute capex, but it really isn't hard to find power / space for that.
Yes they will, they will prefer natural gas of those two. The main reason of emission reductions as we transition off of coal.
I really hope we go all in on nuclear though, with some natural gas and get rid of the windmills. Solar, hydro, geothermal can stay where it makes sense.
Unfortunately our skies and land are littered with windmills made out of unrecyclable polymers that are terrible for raptors.
Queue someone quoting how many cats kill birds like competing with the top predator is a good thing, or ignoring the fact that we put these in raptors' wind streams and cats don't hunt those large birds (which are usually endangered)
Aah a True Bird Lover. Wants to protect birds from windmills. Doesn't care how many bird habitats oil drilling destroys. Hasn't seen any pictures of oil coated birds from the Exxon Valdez or the Deepwater Horizon.
> made out of unrecyclable polymers
Does this mean you support banning plastic straws too? All plastics are essentially unrecyclable.
> Our kids deserve nuclear.
By all means. If it can be done as cheaply as wind and solar.
Ideally we could get rid of everything for nuclear, but that's not today's reality. The truth is natural gas is needed to transition off of coal and has helped reduce emissions.
That being said, I would like to reduce those drilling rigs. However I don't see their existence justification for more damage by putting up windmills.
It's classic whataboutism. Both things can be harmful.
Below is an exercise to help point it out.
> Like our lands and oceans are not littered ... already
Without the "what about" subject, this statement looks bad doesn't it?
That reasoning is no different than the people who see trash on the ground and throw more because it's already trashed, so why not.
> it's really the only way forward. seems like a win/win.
There is another way forward, which is not building these data centers, forcing AI companies to use power more efficiently, and use the excess energy production capacity towards the energy transition in order to avoid the worst consequences of climate change.
It's not going to happen, at least not right now, but it's clearly what we ought to do. ChatGPT can wait.
> What I'm saying is that using gigawatts of power for "AI" in this day and age is madness
Why? American datacentres--of all types--use about 250 TWh per year, with another 500 TWh additional capacity expected by 2030 [1]. American paper manufacturing used about that much energy in 2018 [2].
If I read the data right (1) the US currently produces roughly 4,000 TWh of electricity every year. 500 TWh is a significant portion of that! The US will need a lot of additional capacity for things like electric cars and heat pumps. Most of the effort should be going towards that, not huge data centers attending to unproven demand (how many people will pay the real price for ChatGPT once the VC subsidy ends remains to be proven).
The sources for some of the future data centers will be local and not necessarily influence the US grid. Consider also that cement production uses about 3000 TWh per year worldwide, and aluminium smelting uses about 1000 TWh per year.
A lot of that cement production still uses fossil fuels.
In my mind, all the electricity production capacity we can build needs to go to the electrification of the existing economy, not new stuff and especially not the current brand of AI.
The US electricity grid is a complex problem at the interface of state and local governments and cannot scale quickly. There are no major concerns with scaling up the local electricity production however, other than the installation time (and safety, if nuclear). Reliability is easier to solve locally than for a huge grid.
> OK then I guess we should disconnect every large industrial electricity customer from the grid.
No, what we should do is put every "large load" electricity customer (including but not limited to these data centers) into their own rate-payer class like they did in Maryland and Oregon instead of lumping them in with everyone else.
Once they are in their own rate-payer class then their rates can be adjusted to pay for the costs of the increased infrastructure that is only needed because they exist (take away data center build-outs and electricity usage is largely flat or falling pretty much everywhere in the US).
I hope the data center developers are paying you to lobby for their ongoing corporate welfare? Because that's what you're basically doing here.
If they pay the same rates as everyone else then that's not corporate welfare. If the rates are artificially low thus causing shortages then we have a different problem.
> If they pay the same rates as everyone else then that's not corporate welfare.
It is, because the electricity companies don't have magic electricity generating machines that can scale infinitely.
To satisfy the new demand which only exists because the data center was built, they spend a lot of money on new infrastructure. They then raise everyone's prices by an equal percentage to support this new infrastructure even though the infrastructure was not needed until the data center was built.
Not charging the data center developers for that extra build out and expecting everyone to absorb the costs for new infrastructure that never would have been built if the data center wasn't built is absolutely corporate welfare.
I think you may be missing the whole national security part of the AI race. This isn't just about asking a computer what recipe you can cook tonight with the items in your fridge. In many ways it is similar to the race to build a nuclear bomb. We may individually not like that, but we might individually be best served to live in the nation that got there first.
We can see in Ukraine that AI plays a very very small role in the war. Production of small drones and their control across jammed areas is the current problem space.
And I bet that a majority of those LLM usage in Ukraine is via local focused models, good for picking shapes on a grainy images and not much else, but fast and cheap. There is literally zero use for the gigantic general LLMs which can produce human-like output and routinely generate fake numbers, in the army setting.
That's the story the proponents of the AI bubble would have you believe, because they are sucking in all available funding to their enrichment, or because they've been huffing their own hype gas for so long that they have no brain cells of their own left.
It is, however, complete nonsense, and the next few years of failed promises on AGI will eventually bring people to their senses, if a market crash and sustained economic depression doesn't do that first. It would be funny if it wasn't going to cause suffering for millions of people, whether we succeed at AGI or not.
I _like_ AI, I find LLMs and many other aspects of useful, and I am optimistic for the long term prospects of AI. But the rush to try and get to AGI is completely out of control at this point, and the fallout from when the bubble pops will set AI, and our societies, back a long time.
"Allowing" doesn't necessarily translate into "doing". Many people are seeing higher energy prices which are at least in part or wholly due to data center loads on local power grids. In Ohio, for example, we just skated by with a ruling from our Public Utilities Commission which effectively required data centers to pay for their impact on local grids [1].
Additionally, while these data centers do provide some jobs, where states are giving them grants, loans, infrastructure improvement, or otherwise they are ultimately extractive developments (like parking lots) where the wealth flows out from states like Ohio and flows in to states where the CEOs and HQ sit (California, New York, etc.).
I can tell you that people in Ohio across the political spectrum are not happy. We are losing good farm land, utilizing water, and our power costs are going up for negligible benefits at best. But hey now our state representatives can say "Meta is coming to central Ohio". Meanwhile costs are going up and we still have to ship produce in from other countries and states.
If our representatives and governors office thought about this all for about 2 seconds they would require any data center development to include 2x the number of corporate jobs over a certain income threshold or else not approve the development. If the developers balk, then fine it's not like we want them anyway.
The Trump Administration (and for that matter probably any admin) isn't doing jack shit.
Oh nice, that basically solves the issue. I've been hearing horror stories of datacenters overloading existing grids and raising prices for the average person, If datacenters generate their own power that basically solves the issue.
Out here in AZ, solar combined with battery would be perfect for datacenters.
> the public is expected to deal with higher prices after every data center build?
No. But the states that let companies put trillions of dollars of datacentre and power hardware in their borders will probably reap benefits from it for decades to come regardless of how AI pans out.
The most cost effective way to run a datacenter at some definition of "pedal to the metal", 24/7. This is not appropriate for solar, which is why these companies are looking into power sources that are most cost effective when they run pedal to the metal, 24/7, like nuclear.
With all that extra cheap energy, we can start making more steel again. Desalinate water. There are a lot of things that can be done with vast amounts of underutilized cheap energy.
but it’s private money? who gives a fuck about the $/job created? if anything, it’s a good thing that Anthropic can afford to do it because they can so efficiently use capital. at least, so far…
I think GP is criticising how ‘wasteful’ that is compared to better uses of the money in $/employee. It's really not creating that many jobs for the amount of cash we're talking about.
Anthropic just borrowed 133 billion dollars from different investors in September, and 20 billion in total over 2025. Just where would they get such amount of actual money? Ask Softbank for a donation? Those guys do like to pour money into a fire pit.
Will this pan out? We don't know, no one knows. But this isn't "a scam" there is a plausible future where a large percentage of white collar (or dare I say it, blue collar) work will have an assistant and that assistant requires a considerable subscription (200/mo? 1000/mo?).
Interesting to see all of the leading labs in the West make this bet.
12k a year out of your paycheck for an advanced Clippy "assistant"? Sorry, this is not plausible. Oh and by blue collar work do you mean done by walking talking robots? I bet you think we'll be flying cars to work w/in 5 years too huh. Oh yeah and when is your chatbot going to solve physics and cure cancer again? You ppl have lost your minds.
Actually yeah 12k a year better be really really good, because that can get you a lot of quality human. At 12k per year and $100/hour, that gets you 120 hours of time which means you get ~20 minutes per day, on average. Or if you get down to $33 an hour it's an hour a day.
I don't think it requires robots. Although that's possible too.
I think that HoloLens has a reasonable demonstration of how to assist blue collar work about 10 years ago (AFAIK it flopped). I would bet a dollar that similar technology augmented with LLMs could be useful to blue collar work.
Well clearly Microsoft don't see how given that they canceled the project this year. Do you really think we need some video game HUD in our vision at all times? Come back to the real world it's nice here.
Growth curves mean nothing if you're selling $0.90 dollars. You have to show a growth curve when price > cost. It's not even clear that value > cost.
I absolutely love Anthropic; but I am worried about the fiscal wall they will hit that will ratchet up my opex as they will need to steeply raise prices.
You have to include the carrying cost per customer as well which is mostly labour. Most of SaaS undercounts the payroll attached to a subscription which is why it is so hard to get to positive net margins and maintain lifetime value.
I am sceptical an LLM foundation model company can get away with low human services either directly on its own payroll or by giving up margin to a channel of implementation partners. Thats because the go to market requires organizational change on the customer sites. That is a lot of human surface area.
Lets put that quote in its full context, because its designed to sound much more impressive than it actually is.
> Anthropic serves more than 300,000 business customers, and our number of large accounts—customers that each represent over $100,000 in run-rate revenue—has grown nearly sevenfold in the past year.
Let me deconstruct that:
> Anthropic serves more than 300,000 business customers
Hard fact. No qualification on spend or activity, are they on trails or fully paid with contracts and minimum spend
> and our number of large accounts—customers that each represent over $100,000 in run-rate revenue
run-rate revenue is an extrapolation. (https://www.fool.com/terms/r/run-rate/) That could be buisnesses that trail anthropic for a month, spend 24K and think "fuck thats expensive" and stops spending. average that over 2 months, then times by 12, boom 100k account.
> has grown nearly sevenfold in the past year.
no starting base....
Its unconvincing, because its smoke and mirrors. Give me the numbers of paying customers, over time with revenue. Then show the opex/capex.
Less impulse, more "oh we expected that we'd get more return".
We have a project at the moment thats all based around sharepoint. They have ingested many tens of thousands of documents, and are expecting that MS copilot studio will be able to a) RAG and B) produce meaningful answers with a 4 line prompt.
Thats my point right? you get monthly billing, look at the curve and go "oops thats not good"
And then cut spending. The point is, run-rate revenue, which is extrapolated rather than _actual_ can easily mask this change, depending on how its calculated.
They buy hardware, replace it as the years go on, and continue doing business.
The investment isn't just in raw compute - they have to build buildings, pay staff, and other things. For the hardware and software - they just keep pace as all the other computing companies have to.
https://news.ycombinator.com/item?id=45896707