Nscale, Anthropic In Talks For New $45 Billion AI Cloud Deal (reuters.com) 22
An anonymous reader quotes a report from Reuters: Nscale is in talks to sell AI computing power to Claude maker Anthropic in a new deal that could be worth about $45 billion, a person familiar with the matter told Reuters on Monday. The data center developer is seeking another major deal with the AI lab for a second facility at its Monarch data center campus in West Virginia, the person said. In August, Anthropic struck a deal to spend $45 billion to rent AI cloud computing power from Nscale's West Virginia data center campus. A second Monarch building would deliver 460 megawatts of computing power using Nvidia's advanced Vera Rubin AI chips, according to the source. [...] According to the Nscale prospectus, Microsoft and Anthropic are expected to be the company's major customers in the future. Further reading: Microsoft Abandons UK AI Supercomputer As Grid Struggles to Keep Up
the desperation to launch the IPO (Score:3, Interesting)
Re: (Score:3)
The IPO is dead unless they can replace the revenue they just lost from MSFT. You'd think it would be a problem that Anthropic is a dead man walking whereas MSFT is the strongest tech company, financially speaking, right now. But investors are dumb enough to not care, so here we are.
This whole thing is beginning to look like another "greater fool" grift like cryptocurrency. But they NEED AI to REQUIRE massive compute, because then they control and observe every transaction with that model. You can't control the masses without central control of what they see. Think META.
Open weight models keep getting better at a rate just a few months behind OpenAI and Anthropic. And models created in hardware that like https://chatjimmy.ai/ [chatjimmy.ai] that you upgrade as you choose (like a CPU) look to be fa
Re: (Score:1)
You do realize that you've never had privacy.
As soon as you got internet and/or a cellphone, or an email, or joined Facebook or Twitter, you signed over every single little bit of data (take a day and carefully read the EULA and ToS... you might be surprised)
Re: (Score:2)
You do realize that you've never had privacy.
As soon as you got internet and/or a cellphone, or an email, or joined Facebook or Twitter, you signed over every single little bit of data (take a day and carefully read the EULA and ToS... you might be surprised)
There are many mitigations available to protect privacy. Yes, they require knowledge of how the Internet works. Eliminating pixel beacons and ads is becoming trivial.
No platform is going to help you avoid *their* scrutiny. I don't use any META or Twitter/X. And I do read EULAs.
Slashdot appears pretty light on the surveillance. Though their advertisers are about the same as on all websites with ads.
Re: (Score:3)
You do realize that you're sending data over the 'net, whether encrypted or not, it can be read by some Agency... if it can't be read, then umpteen millions could possibly be planning some attack, maybe, possibly, who knows.
Are you just going to assume some traffic heading to Israel is an attack plan, even though it's actually only a news article?
More compute is not needed (Score:2)
I don't understand that instead of spending a few million on improving the code and the algorithmd they are instead spending billions on infrastructure? Is this all a racket? It makes no sense. Even I can help them optimise so they don't need this (it's what I am doing in my job anyhow), but it seems like there's something else going on that we are not supposed to know...?
Re:More compute is not needed (Score:5, Interesting)
I don't understand that instead of spending a few million on improving the code and the algorithmd they are instead spending billions on infrastructure? Is this all a racket? It makes no sense. Even I can help them optimise so they don't need this (it's what I am doing in my job anyhow), but it seems like there's something else going on that we are not supposed to know...?
The optimistic (fantasy) view: They hope the hardware will be useful for something else when a sea-change optimization happens in the AI world.
The pessimistic (realist) view: Everybody involved from the tech bros to the investors are so busy sniffing their own and each other's farts that they can't be bothered to look at reality and are instead letting the vapor trail suck them up into the stratosphere with zero protections or parachutes, because even the few left with an eye toward reality have a firm belief that the government will throw the public under them to protect them when they inevitably fall from their lofty heights. What's really sad is? They have plenty of reasons to believe it. It's happened before when big ideas failed. It'll likely happen again. And the more of the economy they can suck into the scheme before the inevitable "correction" comes along, the more sure they can be that they'll get that sweet, sweet bailout relief at the expense of everybody else.
Re: (Score:2)
Live on debt and hope the products actually mature fast enough to become profitable before the company goes bankrupt. If they show anything positive, those who own the debt will let them keep going in the hope that what was spent already won't be wasted money.
Re: (Score:2)
The companies won't go bankrupt... ever seen Wall-E or I, Robot?
Re: (Score:2)
Live on debt and hope the products actually mature fast enough to become profitable before the company goes bankrupt. If they show anything positive, those who own the debt will let them keep going in the hope that what was spent already won't be wasted money.
Sounds just like the vast majority of startups in the 1999-2000 period.
There was not much of a bailout then, nor should there be now. Though now everyone has these overextended suppliers in their retirement portfolios. The dot-com crash eventually killed many hardware suppliers like Sun Microsystems. When you could buy their hardware for pennies on the dollar.
Re: (Score:2)
Re:More compute is not needed (Score:5, Interesting)
This is a hoarding situation on the part of AI datacenters.
At this point I would consider the current hardware availability situation a national emergency and regulate that some share of the hardware production be required to go to anything not having to do directly with an AI customer. Specifically RAM.
anyone turning this into a political discussion please just don't.
Re: (Score:2)
I'd like the RAM landscape to change too, but if the government is allocating it you can bet the price will go even higher.
And note that it would have to be the government of a place that makes memory. And could only apply to RAM that won't be manufactured for a year or two, as all the current and planned production runs have already been bought.
Re: (Score:2)
Long money is on what to do with that cheap compute when this AI dross bubble pops.
Hopefully someone does something useful with it.
Re: (Score:3)
Long money is on what to do with that cheap compute when this AI dross bubble pops. Hopefully someone does something useful with it.
The problem is, nobody can afford to do anything with it. Just keeping the power on is very expensive. It costs tens of thousands of dollars in power alone just to train a shitty model. None of these companies are anywhere near a profit, so they cannot afford anything, so Nvidia isn't profitable either. They're loaning out more than they're making, and there's no route for them to ever get it back. But they can declare bankruptcy and completely fuck all of our pensions, after they run off with the actual ca
Re: (Score:2)
The same could be said for computers. Why spend all the money on increasing processor or RAM speed or new video cards when you could just update the code behind the OS and software? We'd still be using 486 machines!
Re: (Score:2)
I don't understand that instead of spending a few million on improving the code and the algorithmd they are instead spending billions on infrastructure?
The same could be said for computers. Why spend all the money on increasing processor or RAM speed or new video cards when you could just update the code behind the OS and software? We'd still be using 486 machines!
Software improvements make really poor moats. Proprietary hardware with costs in excess of what any group of potential customers could do themselves that provides a permanent customer paying lock-in treadmill (like Oracle is designed) is the wet dream of all those pushing the current AI race.
Re: (Score:3)
The hardware manufacturers are major investors in the AI companies. The AI companies promised to buy more hardware as a condition of the investments. e.g. MS is an investor in the AI companies, and leases compute time to the AI companies, on hardware it buys from Nvidia, and operates in datacenters it lease from Nscale.
Around and 'round and 'round the money goes inflating the stock value at every pass.
birds of a feather (Score:2)
We went to the moon on my phone (Score:2)
Well a lot less than my phone. The urban myth was my Texas Instruments scientific calculator. Viva 69! When coders could code.
O Capitalism, you are an ignorant, greedy hoar, but someone will always love you.
Home improvement opportunities (Score:2)