Mao tried picking industrial locations for survivability in case of war: https://en.wikipedia.org/wiki/Third_Front_(China) Turns out that places that are hard to reach for the enemy are also hard to reach in peacetime, so the project was ultimately abandoned.
Why? I think any attack on China, Russia, India by major power will quickly escalate to nuclear. At this stage everyone will have much bigger problem than worry about datacenters.
It would very slowly escalate toward nuclear, and then all at once.
Nuclear rhetoric would escalate rapidly by the losing / weaker side.
It's a common fallacy that conflict between two nuclear powers would instantly jump to nuclear exchange. That isn't how it would go at all, although the propaganda that it would does likely help to prevent wars. Nobody is interested in vaporizing all of their own people by starting a nuclear exchange from the first bullets fired. There would be various high-intensity inflection points, triggers, that would risk nuclear escalation: when one side or the other starts losing a lot of territory; when one side or the other loses a major city; when one side or the other is at risk of operational collapse (leading to more rapid losses in the field); when one side or the other is at risk of losing their core territory / capital; and so on. The key inflection points would be prime candidates for nuclear threats, to try to get the winning side to stop or back off immediately or else. From there it'd be a complex equation as to when a side would actually finally set off a nuke (would it be a warning nuke first, etc).
From the article: 1.859 billion kWh. Divide by 1 year = 212 megawatt. Definitely not small. Not that gigantic either, compared to current day (US) AI/hyperscalers.
It's about the delta. Similar kinds of buildouts weren't common 3-5 years ago.
China as well as India are seeing a significant buildout of DC capacity which is partially a result of subsidies but also a result of increased demand for services domestically as well as within APAC.
While local politicans in the US are using populist misinformation such as water usage to push back against DC buildouts, you're seeing both China [0] and India [1] offering multi-decade subsidizes and tax shelters to build out DC capacity, which leads to a snowball effect as supply chains and ecosystems start shifting to those centers of gravity - this is what happened to semiconductor fabrication (Upstate NY to Taiwan/SK) and Cybersecurity (Bay Area/Boston to Israel/CEE/India) 15 years ago.
Additionally, this is essentially HPC and best in breed hardware doesn't neccesarily equate to outcompeting a large amount of less performant but scaled out compute, which is the path both China and India are adopting by pushing hyperscalers to (in China's case) use domestic vendors or (in India's case) transfer the entire IP chain to an Indian subsidiary. This is why you see Amazon announce layoffs [2] while committing to spend tens of billions of dollars in India's hyperscaler buildout [3] or Marvell expanding in Vietnam [6] after a similar sized layoff [7].
Barely 10 years ago, states like California, NC, etc used to use similar strategies to compete with each other and build out innovation hubs, but state and local level politicking has led to public-private partnerships stagnating and incentivized beancounters like me to seriously consider offshoring or shutting down "innovation centers" in RTP/Boulder/ATX tier hubs.
CHIPS and the IRA were on the right track, but even Dems [4] and the EU [5] as well as the GOP tried to stymie it.
They really should be building such things deep underground, at least to offer some basic protection against the various adversaries.
I assume we do too...we do right?
Mao tried picking industrial locations for survivability in case of war: https://en.wikipedia.org/wiki/Third_Front_(China) Turns out that places that are hard to reach for the enemy are also hard to reach in peacetime, so the project was ultimately abandoned.
MAD is far too intact with China to bomb random infrastructure with reckless abandon.
Why? I think any attack on China, Russia, India by major power will quickly escalate to nuclear. At this stage everyone will have much bigger problem than worry about datacenters.
Pretty much, they aren't going to trust each other that an incoming missile is not armed with a nuclear payload.
It would very slowly escalate toward nuclear, and then all at once.
Nuclear rhetoric would escalate rapidly by the losing / weaker side.
It's a common fallacy that conflict between two nuclear powers would instantly jump to nuclear exchange. That isn't how it would go at all, although the propaganda that it would does likely help to prevent wars. Nobody is interested in vaporizing all of their own people by starting a nuclear exchange from the first bullets fired. There would be various high-intensity inflection points, triggers, that would risk nuclear escalation: when one side or the other starts losing a lot of territory; when one side or the other loses a major city; when one side or the other is at risk of operational collapse (leading to more rapid losses in the field); when one side or the other is at risk of losing their core territory / capital; and so on. The key inflection points would be prime candidates for nuclear threats, to try to get the winning side to stop or back off immediately or else. From there it'd be a complex equation as to when a side would actually finally set off a nuke (would it be a warning nuke first, etc).
I doubt it. Maybe a full fledged invasion might get nuclear but mutually assured destruction is a pretty solid deterrent.
From the article: 1.859 billion kWh. Divide by 1 year = 212 megawatt. Definitely not small. Not that gigantic either, compared to current day (US) AI/hyperscalers.
It's about the delta. Similar kinds of buildouts weren't common 3-5 years ago.
China as well as India are seeing a significant buildout of DC capacity which is partially a result of subsidies but also a result of increased demand for services domestically as well as within APAC.
While local politicans in the US are using populist misinformation such as water usage to push back against DC buildouts, you're seeing both China [0] and India [1] offering multi-decade subsidizes and tax shelters to build out DC capacity, which leads to a snowball effect as supply chains and ecosystems start shifting to those centers of gravity - this is what happened to semiconductor fabrication (Upstate NY to Taiwan/SK) and Cybersecurity (Bay Area/Boston to Israel/CEE/India) 15 years ago.
Additionally, this is essentially HPC and best in breed hardware doesn't neccesarily equate to outcompeting a large amount of less performant but scaled out compute, which is the path both China and India are adopting by pushing hyperscalers to (in China's case) use domestic vendors or (in India's case) transfer the entire IP chain to an Indian subsidiary. This is why you see Amazon announce layoffs [2] while committing to spend tens of billions of dollars in India's hyperscaler buildout [3] or Marvell expanding in Vietnam [6] after a similar sized layoff [7].
Barely 10 years ago, states like California, NC, etc used to use similar strategies to compete with each other and build out innovation hubs, but state and local level politicking has led to public-private partnerships stagnating and incentivized beancounters like me to seriously consider offshoring or shutting down "innovation centers" in RTP/Boulder/ATX tier hubs.
CHIPS and the IRA were on the right track, but even Dems [4] and the EU [5] as well as the GOP tried to stymie it.
[0] - https://www.reuters.com/world/asia-pacific/china-offers-tech...
[1] - https://techcrunch.com/2026/02/01/india-offers-zero-taxes-th...
[2] - https://www.reuters.com/business/world-at-work/amazon-plans-...
[3] - https://www.aboutamazon.in/news/economic-impact/amazon-econo...
[4] - https://www.nytimes.com/2024/01/10/us/politics/semiconductor...
[5] - https://www.cnbc.com/amp/2022/11/09/europe-shows-a-united-fr...
[6] - https://www.marvell.com/company/newsroom/marvell-accelerates...
[7] - https://www.crn.com/news/computing/semiconductor-firm-marvel...