Data Centers in Nvidia's Hometown Stand Empty Awaiting Power (yahoo.com) 40
Two of the world's biggest data center developers have projects in Nvidia's hometown that may sit empty for years because the local utility isn't ready to supply electricity. From a report: In Santa Clara, California, where the world's biggest supplier of artificial-intelligence chips is based, Digital Realty Trust applied in 2019 to build a data center. Roughly six years later, the development remains an empty shell awaiting full energization. Stack Infrastructure, which was acquired earlier this year by Blue Owl Capital, has a nearby 48-megawatt project that's also vacant, while the city-owned utility, Silicon Valley Power, struggles to upgrade its capacity.
The fate of the two facilities highlights a major challenge for the US tech sector and indeed the wider economy. While demand for data centers has never been greater, driven by the boom in cloud computing and AI, access to electricity is emerging as the biggest constraint. That's largely because of aging power infrastructure, a slow build-out of new transmission lines and a variety of regulatory and permitting hurdles. And the pressure on power systems is only going to increase. Electricity requirements from AI computing will likely more than double in the US alone by 2035, based on BloombergNEF projections. Nvidia's Jensen Huang and OpenAI's Sam Altman are among corporate leaders that have predicted trillions of dollars will pour into building new AI infrastructure.
The fate of the two facilities highlights a major challenge for the US tech sector and indeed the wider economy. While demand for data centers has never been greater, driven by the boom in cloud computing and AI, access to electricity is emerging as the biggest constraint. That's largely because of aging power infrastructure, a slow build-out of new transmission lines and a variety of regulatory and permitting hurdles. And the pressure on power systems is only going to increase. Electricity requirements from AI computing will likely more than double in the US alone by 2035, based on BloombergNEF projections. Nvidia's Jensen Huang and OpenAI's Sam Altman are among corporate leaders that have predicted trillions of dollars will pour into building new AI infrastructure.
Let's get some benefit from the AI bubble (Score:5, Insightful)
Require the datacentre builders or AI companies to pay for transmission grid upgrades to the whole grid to add the capacity they need.
Then when the bubble collapses, we will at least be left with a greatly upgraded electricity grid that we can use to transit cheap renewable power to consumers, replacing expensive gas and petroleum energy sources.
Re: (Score:3)
Thorium MSR have never been proven commercially. The test reactor at ORNL was shut down frequently, did not provide electricity and used U-233 as fuel
Re: (Score:2)
China disagrees.
[...]
Or is there something I'm missing?
What you're missing, as usual, is that IT DOESN'T EXIST YET but you're claiming that it does, as usual.
Re: (Score:2)
"two megawatt experimental reactor in Gansu province
I said commercially.
"What are we waiting for to build thorium reactors in the USA? That China beats us to it? I can expect some new excuse to come up then, some reason to claim the technology is out of our reach"
I don't know & I've never mentioned China.
I have heard a lot of people beginning with Kirk Sorenson almost 20 years ago talking about thorium reactors although i only vaguely recall the details.
There was something about shifting from MSR to LFT
Re: (Score:2)
"I've never mentioned China"
typo, should be - "i didn't mention China"
Re: (Score:2)
We need copr and consumer-level thorium reactors. centralized energy has become a choke point of control, Thorium molten salt reactors are intrinsically safe, they cannot melt down..
Not saying this or the rest of your statement is inherently wrong. Just saying the last ones who tried to make that whole "cannot melt down" claim, own a Chernobyl-sized exclusion zone today. Which they also tried to initially cover up and deny. Not that (cough, every) government is guilty of doing that from time to time.
Consumer-level? "Mostly harmless" went out the window when they made way for the intergalactic superhighway. You're gonna have to do better than that for the bank or the homeowners ins
Re: (Score:2)
Also water infrastructure! Datacenters take a lot of water for cooling, and our current water infrastructure is already strained.
Re: (Score:3)
I don't think the bubble will collapse (Score:2)
The main purpose of all these llms is to replace White collar jobs. Everything else is just leading up to that.
So even when the industry shakes out and your left with two maybe three big players they're still going to need all that electricity to replace those jobs.
So we're not going to get any cheap capacity and infrastructure out of them because they're going to be using it.
Financially a bubble might burst when the industry shakes out everybody but the big b
Re: (Score:2)
Like back when the deregulated comms sector over-built capacity.
regulatory and permitting hurdles (Score:1)
This is the problem
Armies of Vetocrats whose only job is to say no
And even bigger armies of NIMBYs and their lawyers
Meanwhile, Chinese companies can build stuff
And yeah, they aren't perfect. The term Tofu Dreg (exceptionally poor construction) comes to mind
There was once a time when it was easier to build stuff in the US
And it's way worse in the EU
Data centers are bad for communities (Score:4, Insightful)
These data centers provide little benefit to the communities where they locate (they only generate maybe a handful of jobs) but result in jacked up electricity prices for everyone. We should be making these companies pay for all the extra power infrastructure they need, not passing the bill on to the rest of us.
Re:Data centers are bad for communities (Score:4, Insightful)
This is exactly right. I don't have a problem with data centers per se. I think they're useful things. However, I do have a problem when they compete for residential utilities. If they can run 100% off-grid using some combination of solar, wind, hydro, geo, nuclear, then go for it. If they want to share existing public utilities, then they should be required to subsidize consumer rates so that they remain capped at pre-datacenter levels for as long as the datacenter is in operation. If the current residential rate is $0.20/kWh and the utility seeks to raise residential rates to $0.25/kWh, the datacenter operator's rates must rise to negate any impact to consumers. If the local grid requires updates, the datacenter must bear the bulk of the cost. This will not be popular with data center operators, but it does negate the rent-seeking impetus and creates a tangible local benefit from hosting data centers.
Re: (Score:3)
Or, here's an idea: don't build large facilities that can't be powered up in geographical areas that can't provide the resources you need to operate the large facility.
What critical need is served by building these things in Santa Clara rather than somewhere with a bit more infrastructure to handle it, other than proximity to HQ?
Re: (Score:2)
I get what you're saying but I don't think there are many places in America sitting on large amounts of surplus power.
Re: (Score:2)
What critical need is served by building these things in Santa Clara rather than somewhere with a bit more infrastructure to handle it, other than proximity to HQ?
It's very simple. Most of Silicon Valley is served by PG&E. Santa Clara City has its own electricity utility with rates far lower than offered by PG&E. They are willing to wait for a hookup because the rates are so favorable.
Meanwhile the city utility has proposed yet another rate increase at least partially due to "construction costs, as well as to fund new projects." New home construction in Santa Clara proceeds at a slow pace, so the utility expansion is largely to support the data centers, which
Very specific to SVP (Score:5, Interesting)
Silicon Valley Power has always operated as a nonprofit, low-cost, major consumer focused utility. They have minimal generation resources and import the vast majority of their power. Their peak demand was ~500MW last time I checked. Adding 10% to that number with a single connection is a huge ask; traditionally customers that wanted a connection like that would be co-generating and only using the utility for load leveling and backup of critical loads. A new 70MW combined cycle plant has a lead time these days of around 5-8 years.
Which ultimately gets into why there is an opportunity for small modular reactors IMO. The customer power increments are just too big for conventional generation, but full scale reactors end up in the wrong place where transmission capacity starts to constrain a project.
Re:Very specific to SVP (Score:5, Insightful)
You have a firm like NVDIA that currently has a market cap of $4.5 trillion. If it were a country and we compared GDP (this is an apples to dragonfruit comparison, I know ) NVDIA would be 4th on the list after USA, China, and Germany and ahead of Japan
United States 29.1T
China 18.7T
Germany 4.6T
Japan 4.0 T
Why can't...why shouldn't a firm like NVDIA, MSFT, ORCL, FB, AAPL etc build out their own capacity? For this one project they could by 5000+ solar panels, some acreage in gods country, all the battery packs they need, and tweak the software to run during the optimal times. Do it all in-house for relatively cheap.
Here's what I think pisses people off....they don't do this, rather these firms make false promises about nuclear but then squeeze their way into existing generation and transmission, raising prices for residential customers. In essence arbitraging. "Hey, we can spend XX billion on our own generating capacity, or we can take from existing capacity + transmission, raising rates for everyone and ourselves, but it will be a few billion cheaper than building out our own generating capacity." They have trillions but are still sticking it to society so the SVP of blah blah blah can make his quarterly numbers
It also speaks to how truly invested in this AI race they are. Which is to say not as much as they advertise since they are only putting a fraction of their sizable cash towards renting AI 'stuff'
About the only good thing that may come of this is that the extra capacity and transmission will be available for EV's and residential electrification (away from nat gas) when the AI bubble deflates. Not sure what to do with all the GPU's, maybe artificial coral reefs? Grind into silicon dust and make more solar panels?
Re: (Score:2)
You still need to get that solar power from "God's Country" to the Valley. SVP wasn't set up to own long distance transmission assets, and for private entities to build them to service their own needs has its own whole host of challenges.
So, you look at on-site generation options and ways to free up capacity from other users. General rooftop solar, storage batteries, etc work well to a point. Beyond that centralized generation is pretty much unavoidable.
Ghost towns... (Score:1)
This is their own fault. (Score:2)
American big tech knew that they would eventually run out of power for their clouds, social media, video, etc. AI just hastened the problem. They should have been planning for this years ago. Nuclear power was always the solution. Tech companies should have been funding their own nuke plants and lobbying politicians to loosen America’s regulations on building new plants. But they fucked around, didn't plan ahead, and it bit them on the ass. Meanwhile the tech leaders in China saw this coming and worke
Shut down some Bitcoin miners (Score:2)
Anyone note the narrative change? (Score:2)
Remember last month, when energy supply was scary scarce, but now suddenly it's infrastructure that is the constraint?
Re: (Score:2)
Those are just different faces of the same coin. A gas turbine plant needs the physical equipment, the transmission capacity, a distribution network, gas pipelines, cooling water, land, and a whole host of other dependencies. When you design around one constraint the next one pops up.
Lower power (Score:2)
Maybe this is incentive to help design data centres that are less power demanding, such as using computers that use ARM and are better with how their code is implemented?
Then combine that with roof top renewables.
This is a hard problem, but if the economic incentive is there, then someone will want to address it.
Macroeconomics of power (Score:2)
When I was in college circa 2000, my undergraduate macroeconomics professor pointed out why the US is so attractive to corporations. "If you want to setup shop in China," he said, "you will need to pay for your own backup generator because the power grid isn't reliable like it is here." He went on to explain similarly about labor - how the US constantly produces a supply of educated people (referring to us in the classroom). So when you incorporate in the US, you have everything you need.
For decades peop
Do these people still have jobs? (Score:2)
So if you plan, and then try to do a big thing, and you find out you're a failure because you didn't actually plan... Please tell me these guys didn't just keep getting paychecks.
You know damn well a single full server rack can have enormous power density and simple math says you can't power it .... and you are just 'hoping' some other enormous project comes along you must be hella stupid.
If you don't control it...you can't count on it.
And the bonds? (Score:2)
So let me get this straight: they built the datacenters from investment money... and now, due to lack of power, they're just sitting there doing nothing. And the debt they were built with is earning nothing... actually, given property, etc, taxes, is loosing money.
Let's see if we can bring the world economic collapse on sooner.