MIT Technology Review · Mar 26, 2025
MIT Technology Review reports China’s data center boom is running into a demand mismatch
Markdown · JSON · Original source
MIT Technology Review’s report, published March 26, 2025, describes a boom in China’s AI data center construction followed by widespread underuse and financial strain. The report says facilities were often built without enough regard for demand or technical feasibility, while the rise of DeepSeek shifted many potential customers away from training models and changed what they need from computing infrastructure. Many centers, particularly those far from major technology hubs or optimized for rather than inference, may not suit those needs. The report describes falling rental prices, projects struggling to attract clients, and energy being wasted; it also reports that some operators pursue government benefits rather than direct profit from computing services. RAND’s Jimmy Goodrich says the central government is likely to take over and consolidate failed projects, while arguing that China will continue pursuing AI infrastructure as an important capability. The report ends with project manager Fang Cunbao’s view that infrastructure is no longer the main obstacle to AI becoming widespread: “solid plans to deploy the technology” are. Whether underused centers will find a second life remains uncertain.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
MIT Technology Review · Article
Source details
Original source date: Mar 26, 2025 (2025-03-26T09:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-14T20:18:23.241Z. First observed: 2026-10-27T20:03:25.507Z. Retrieved: 2026-10-27T20:05:20.695Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
Some references could not be checked. Reload this page to try again.
Limits
Text as retrieved on October 27, 2026; originally published March 26, 2025. The report relies in part on interviews with sources who requested anonymity and on figures attributed to Chinese media and industry sources; their underlying methods are not provided in the retrieved text.
Open question
Will China’s underused data centers find a second life, or will the government consolidate and transfer them to other operators?
Source-stated
MIT Technology Review reports that many newly built Chinese data centers are underused and that companies running them are struggling financially.
most of the companies running these data centers are struggling to stay afloat. The local Chinese outlets Jiazi Guangnian and 36Kr report that up to 80% of China’s newly built computing resources remain unused.
Source-stated
The report says many projects were built without adequate attention to demand or technical feasibility, according to project participants and other sources it interviewed.
the push to build AI data centers was largely driven from the top down, often with little regard for actual demand or technical feasibility, say Fang, Li, and multiple on-the-ground sources
Source-stated
The report says the rise of DeepSeek has changed demand for computing, as many smaller AI companies have given up pretraining or shifted strategy.
Many smaller players have been giving up on pretraining their models or otherwise shifting their strategy since the rise of DeepSeek
Source-stated
RAND’s Jimmy Goodrich predicts that the Chinese government will likely take over failed projects and hand them to more capable operators.
The situation may eventually prompt government intervention, he says: “The Chinese government is likely to step in, take over, and hand them off to more capable operators.”
Source-stated
Fang Cunbao argues that infrastructure is no longer the main obstacle to AI becoming widespread; he points instead to plans for deploying the technology.
“What stands between now and a future where AI is actually everywhere,” he says, “is not infrastructure anymore, but solid plans to deploy the technology.”
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Editorial provenance
Version 1. Publication ID: ebbb0ee0-cd97-e89a-58f3-b5b7fa7b30a8.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-14T20:13:40.809Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: af858b1382e468d022de15261c631ead2f5d60915a74145187c9470125846edc. Source text hash: 231e8d98036ffd72102ea4b0b244d14d2b2bb736a8f02dbb119515693c46ed07.
Revision and source-check history
Version 1 · 2026-11-14T20:18:23.241Z
MIT Technology Review reports China’s data center boom is running into a demand mismatch
MIT Technology Review’s report, published March 26, 2025, describes a boom in China’s AI data center construction followed by widespread underuse and financial strain. The report says facilities were often built without enough regard for demand or technical feasibility, while the rise of DeepSeek shifted many potential customers away from training models and changed what they need from computing infrastructure. Many centers, particularly those far from major technology hubs or optimized for rather than inference, may not suit those needs. The report describes falling rental prices, projects struggling to attract clients, and energy being wasted; it also reports that some operators pursue government benefits rather than direct profit from computing services. RAND’s Jimmy Goodrich says the central government is likely to take over and consolidate failed projects, while arguing that China will continue pursuing AI infrastructure as an important capability. The report ends with project manager Fang Cunbao’s view that infrastructure is no longer the main obstacle to AI becoming widespread: “solid plans to deploy the technology” are. Whether underused centers will find a second life remains uncertain.
Original source
MIT Technology Review · Article
Source details
Original source date: Mar 26, 2025 (2025-03-26T09:00:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-14T20:18:23.241Z. First observed: 2026-10-27T20:03:25.507Z. Retrieved: 2026-10-27T20:05:20.695Z.
Limits
Text as retrieved on October 27, 2026; originally published March 26, 2025. The report relies in part on interviews with sources who requested anonymity and on figures attributed to Chinese media and industry sources; their underlying methods are not provided in the retrieved text.
Open question
Will China’s underused data centers find a second life, or will the government consolidate and transfer them to other operators?
Source-stated
MIT Technology Review reports that many newly built Chinese data centers are underused and that companies running them are struggling financially.
most of the companies running these data centers are struggling to stay afloat. The local Chinese outlets Jiazi Guangnian and 36Kr report that up to 80% of China’s newly built computing resources remain unused.
Source-stated
The report says many projects were built without adequate attention to demand or technical feasibility, according to project participants and other sources it interviewed.
the push to build AI data centers was largely driven from the top down, often with little regard for actual demand or technical feasibility, say Fang, Li, and multiple on-the-ground sources
Source-stated
The report says the rise of DeepSeek has changed demand for computing, as many smaller AI companies have given up pretraining or shifted strategy.
Many smaller players have been giving up on pretraining their models or otherwise shifting their strategy since the rise of DeepSeek
Source-stated
RAND’s Jimmy Goodrich predicts that the Chinese government will likely take over failed projects and hand them to more capable operators.
The situation may eventually prompt government intervention, he says: “The Chinese government is likely to step in, take over, and hand them off to more capable operators.”
Source-stated
Fang Cunbao argues that infrastructure is no longer the main obstacle to AI becoming widespread; he points instead to plans for deploying the technology.
“What stands between now and a future where AI is actually everywhere,” he says, “is not infrastructure anymore, but solid plans to deploy the technology.”
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Pre-training
A model's initial training on a large dataset, before further adaptation for particular uses.
Graphics processing unit
A chip that performs many calculations at once; the main hardware used to train and run AI models.
Editorial provenance
Version 1. Publication ID: ebbb0ee0-cd97-e89a-58f3-b5b7fa7b30a8.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-14T20:13:40.809Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: af858b1382e468d022de15261c631ead2f5d60915a74145187c9470125846edc. Source text hash: 231e8d98036ffd72102ea4b0b244d14d2b2bb736a8f02dbb119515693c46ed07.
uncertain · 2026-11-14T20:13:40.809Z · anthropic/claude-opus-5.5
Publication: ebbb0ee0-cd97-e89a-58f3-b5b7fa7b30a8
attribution — pass: The reading attributes the report to MIT Technology Review. It credits the quoted views to their speakers, Jimmy Goodrich of RAND and project manager Fang Cunbao. Figures from Chinese outlets are kept as the report's attributed claims.
The local Chinese outlets Jiazi Guangnian and 36Kr report that up to 80% of China’s newly built computing resources remain unused.
summary — uncertain: The summary is faithful, but "inference" is central to the demand mismatch and is neither explained nor in the definitions. The summary also does not say why reasoning models changed what customers need: their need for low latency favours sites near tech hubs. Smallest repair: add a plain gloss, such as running trained models to answer users in real time, or add a term-library entry for inference. DeepSeek could also take a short clause, such as maker of the low-cost R1 reasoning model.
rather than for inference, the process of running trained reasoning models to respond to user inputs in real time
date — pass: March 26, 2025 matches the UTC day of originalPublished. The source is a publisher page, so no archive disclosure is needed. The page was modified the same day. The later 'Deep Dive' sidebar items are not used. Forecasts such as government takeover remain attributed predictions, and nothing is presented with hindsight.
support — pass: Each claim matches its quoted excerpt. Goodrich's takeover prediction is presented as a prediction. His 'important capability' framing and the report's line that experts say China is unlikely to scale back support the summary's 'continue pursuing'. Gaming of government benefits, falling rental prices and wasted energy are all source-stated. No unsupported benefits or costs are added.
The Chinese central government will likely see [underused data centers] as a necessary evil to develop an important capability
media — pass: No media is selected, and no verified media candidates were supplied, so there is nothing to misidentify.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A model's initial training on a large dataset, before further adaptation for particular uses.
Definition referenceA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition referenceA model's initial training on a large dataset, before further adaptation for particular uses.
Definition referenceA chip that performs many calculations at once; the main hardware used to train and run AI models.
Definition reference