China Built Hundreds of AI Data Centers To Catch the AI Boom. Now Many Stand Unused.
- Reference: 0176838291
- News link: https://slashdot.org/story/25/03/27/149238/china-built-hundreds-of-ai-data-centers-to-catch-the-ai-boom-now-many-stand-unused
- Source link:
The rise of DeepSeek's open-source reasoning model R1, which matches ChatGPT o1's performance at a fraction of the cost, has fundamentally altered hardware demand. Computing needs now prioritize low-latency infrastructure for real-time reasoning rather than facilities optimized for large-scale training workloads.
Technical misalignment compounds the problem, as many centers were constructed by companies with little AI expertise, MIT Technology Review reports. The facilities, often built in remote regions to capitalize on cheaper electricity and land, now face obsolescence as AI companies require proximity to tech hubs to minimize transmission delays. GPU rental prices have collapsed, with eight-GPU Nvidia H100 server clusters now leasing for 75,000 yuan ($10,333) monthly, down from peaks of 180,000 yuan, making operations financially unsustainable for many data center operators.
[1] https://www.technologyreview.com/2025/03/26/1113802/china-ai-data-centers-unused/
Seems only logical... (Score:2, Redundant)
China will start using the data centers for Bitcoin mining.
So just like housing (Score:2, Insightful)
Are they going to have a rat renaissance? That's what shitloads of empty buildings usually produce.
Re: (Score:3)
Only if those buildings contain anything rats can use, either for nesting or for food. Empty clean rooms for data centers have none of this.
Re: (Score:2)
They have wires, which rats love to chew on. Once they chew up all the wires and make a huge mess of the place they'll move on to some other shelter.
Re: So just like housing (Score:2)
Hardly. No food there. Most likely they'll be demolished, like those huge empty ghost towns from a decade ago.
Re: (Score:2)
A few were left to effectively rot, but they don't bother demolishing them. A few of the ghost cities have also attracted small populations over time even if they're nowhere near the original capacity envisioned. If given long enough they may become full eventually. I'm sure that these buildings will be repurposed for something else even if they're nowhere near aren't used for data centers. China just overbuilds initially, which is wasting resources that could have been allocated to something else instead,
ghost cities ... "full eventually"? (Score:1)
Chinese buildings are literally falling apart and I'm talking about the ones with people living in them. These uninhabited ones are going to rot and fall down. Also China's population is imploding faster than just about anywhere on earth (not counting populations emigrating like from Ukraine), there is no scenario where these rotting exurbs of are of use for housing.
Re: (Score:2)
They might mothball them. That's what happened with quite a few coal plants that were built but not needed. If the demand ever comes they can be put into action, although it looks like they will be written off unless something dramatically changes.
Re: (Score:2)
Depends on how well they were constructed and whether they are maintained. Poorly constructed buildings with no maintenance degrade much faster so, after just a few years, the costs of resurrecting exceeds the cost of demolition and rebuild. Given the history of modern China, I'd bet on the latter happening.
Re: (Score:2)
I assume most of the costs aren't the building but the hardware therein and presumably that'll be out of date within a few years? Mothballing anything but the core building seems like it wouldn't help much.
Selling off that hardware will have interesting consequences too. It'd push down prices which feel good given the tariff wars going on, but it'd also presumably cause huge amounts of damage to the manufacturers as they suddenly have to compete against hardware they sold last year.
My final concern here, wh
What's the message? (Score:4, Insightful)
Since AI research requires lots of hardware for training and the required trial and error approach, is this saying that in contrast the the US, Chinese companies are abandoning AI research and the search for new and better models, and that instead, the Chinese companies believe they've already found the models they need or that someone what will develop those models? Or maybe this is saying that only the largest Chinese companies are doing AI research while the others will just use whatever models are developed by those large companies.
The article seems to suggest two things. First, AI research, particular with LLMs, is indeed decreasing in China. Apparently development plans have to be registered with the government. Second, the decrease in new data centers appears to be a fallout from small cloud providers who were speculating on demand. It's not clear if the Chinese hyperscalers have scaled back their research. Recent news articles suggest that the hyperscalers are instead ramping up their research efforts.
"infrastructure as a national priority" (Score:2)
"The central government designated AI infrastructure as a national priority, urging local governments to accelerate the development of so-called smart computing centers"
And where did the money come from? Mainly from the State.
"State-owned enterprises, publicly traded firms, and state-affiliated funds lined up to invest in them"
Quoting Hemingway... (Score:3)
Data center. For sale. Never used.
Ghost cities (Score:2)
People mock the Chinese ghost cities, but many (I dare say most)of them actually filled up. Certainly enough that the whole project can't be counted as a waste.Some of those include Pudong, which now has a 400 billion dollar GDP; Kangbashi (150,000 people and growing) Chenggon and more.
Re: (Score:2)
I read one article saying there are over 3 billion homes in China. There are not 3 billion people. There likely never will be, since the country cannot feed 1.4 billion let alone 3.
The herd is going off the cliff (Score:2)
It will be interesting to see what happens in the US.
Microsoft has already cancelled several data centers. When will the other big players realize that they are building obsolete facilities.
Re: (Score:2)
There are two sides to this though, aren't there? Wasn't it basically shown that DeepSeek was trained against already existing LLM models and was basically a condensed model? If that's true, then there still need to be fully trained "full" models to condense. Maybe it means that we need less fully trained models and therefore less giant data centers, but I don't think it means we need zero.
Too much money in the wrong hands (Score:2)
This is an interesting article as it shows how a lot of investments end up going to waste instead of creating something of sustainable use. No country wants to be left behind the "AI revolution", but it's like investing in beachfront development when it's not clear where the beachfront will be in 2-5 years. All of these buildings have a high CO2 cost ultimately, and governments are for the most part not considering balancing technology investments with sustainability. All of the big powers are in an arms r
Sounds like the .Com crash (Score:4, Insightful)
Every company thought that they had the next great idea, only to find 12 other companies had the same idea, and the market would only support 2 or 3. Lots of investment, only to find out you missed the boat.
Re:Sounds like the .Com crash (Score:5, Interesting)
One consequence of the dotcom bubble was the massive buildup of high-speed internet infrastructure that arguably paved the way for the modern internet. It'll be interesting to see what use cases start to make sense with a massive glut in datacenter capacity, even if it doesn't turn out to be AI.
Re: Sounds like the .Com crash (Score:2)
I agree! Assuming the data centers have all the required infrastructure to go live (power, fiber, water, etc...) my hunch is that they will get used. I think a lot of people will be tempted to draw parallels to the Chinese ghost cities, but it might not quite be the same situation.
Re: (Score:2)
Makes me wonder if it's easy to convert them to massive greenhouses, or hydroponics for growing exotic things.
Re: (Score:2)
Every year the computers in the data centers will get less and less relevant.
Re: (Score:2)
As opposed to all the routers and switches installed in the late 90s?
Some of the investment will be in hardware that falls out of date. Some of it will be in infrastructure that can be re-used (buildings, redundant power and cooling). Such is life.
Re: (Score:2)
A large portion of datacentre investment isn't in the computers inside it, and technology doesn't progress anywhere near as fast as you think. Even now in this article the industry is benchmarking on the lowest tier card from the previous generation for their workloads.
Re: (Score:2)
Bitcoin farms
Re: (Score:2)
It's the same thing they did with real estate. There are entire cities with no one living in them in China. And they know it will turn out that way. They use the pointless construction jobs to inflate their economic numbers, one of many ways they manipulate the reports on their economy.