Yves here. It appears that citizens of other countries are not on board with the Big Tech scheme to outsource data center environmental and added energy costs to what is impolitely called the third world. As an aside, I find it remarkable that data center, as in AI, new energy demands are being foisted onto all customers. Those demanding incremental and higher cost energy should pay more. But instead the population at large is subsidizing AI, both through said energy costs plus also bearing the cost of additional pollution.
Admittedly, citizens in advanced economies have been putting up enough resistance so as to lead the tech overlords to look for other victims, on the Larry Summers assumption that poorer countries will agree to accept the energy generation version of garbage barges. But Jomo below discusses how many if not yet most are quite aware of this neocolonial power play and are not on board.
By Jomo Kwame Sundaram, former UN Assistant Secretary General for Economic Development. Originally published at Jomo’s website
Data centers have been proliferating quickly, driven by the popularity of artificial intelligence.
Who Are Data Centers For?
Already, the AI boom has overwhelmed other ‘cloud’ uses and drives the rapid growth of data centers (DCs), imposing fast-expanding resource demands. This has triggered a bipartisan public backlash in the US due to higher energy, water, and land use, as well as rising prices.
In October 2024, McKinsey projected that global energy demand by DCs would rise between 19% and 22% annually through 2030, reaching an annual demand between 171 and 219 gigawatts.
This greatly exceeds the “current demand of 60 GW”. “To avoid a [supply] deficit, at least twice the [DC] capacity built since 2000 would have to be built in less than a quarter of the time”!
As tech companies are not paying for the additional energy generation capacity, consumers and host governments are, whether they benefit from AI or not.
As DCs increasingly faced growing pushback in the North, developers have turned to developing countries, outsourcing problems to poorer nations with limited resources.
Understanding these energy- and water-guzzling facilities is necessary to better protect economies, societies, communities, and their environments.
Energy Needs
With growing corporate and consumer demand for AI, DC growth will continue, and even occasionally accelerate.
Increased AI usage will significantly increase energy and water consumption, accelerating planetary heating both directly and indirectly.
As demand for AI and DCs increases, supporting computers will require significantly more electricity. This will generate heat, needing the use of water and energy for cooling. Much energy used by DCs, from 38% to 50%, is for cooling.
Electricity generation, whether from fossil fuels or nuclear fission, requires more cooling than renewable energy sources such as photovoltaic solar panels or wind turbines.
A small-scale DC with 500 to 2,000 servers consumes one to five megawatts (MW). For tech giants, a ‘hyperscale’ DC, hosting tens of thousands of servers, consumes 20 to over 100MW, like a small city!
Data Centers Not Cool
As the popular focus is on DCs’ enormous energy requirements, their massive water needs to cool equipment tend to be ignored, understated and overlooked.
Locating new DCs in developing countries will further heat local microclimates and the planetary atmosphere. Worse still, heat is more environmentally threatening in the tropics, where ambient temperatures are higher.
Establishing more DCs will inevitably crowd out existing and other possible uses of freshwater supplies, besides reducing local groundwater aquifers.
Unsurprisingly, DC investors rarely warn host governments about the amount of locally supplied energy and water required.
DCs require much freshwater to cool servers and routers. In 2023, Google alone used almost 23 billion litres to cool DCs. In cooling systems using evaporation, cold water is used to absorb severe heat, releasing steam into the atmosphere.
Closed-loop cooling systems absorb heat using piped-in water, while air-cooled chillers cool down hot water. Cooled water recirculated for cooling requires less water but more energy to chill hot water.
Investors Expect Oubsidies
Like other prospective investors, DCs have relocated to areas where host governments have been more generous and less demanding.
Led by US President Trump’s powerful ‘tech bros’, many foreign investors have profited from subsidised energy, cheap land and water, and other special incentives.
Prospective host governments compete to offer tax and other incentives, such as subsidised energy and water, to attract foreign direct investment in DCs.
The US pressured Malaysia and Thailand to stop Chinese firms from using them as an “export-control backdoor” for its AI chips. Washington alleges that DCs outside China buy chips to train its AI for military purposes. So far, only Malaysia has complied.
This limits Chinese firms’ access to such chips. Washington claims that Chinese substitutes for US-made chips are inferior and seeks to protect US technology from China.
High-Tech DC Jobs?
Data centers are emerging everywhere, but not many jobs will be created. Advocates claim DCs will provide high-tech jobs.
DCs are largely self-operating, requiring minimal human intervention, except for maintenance, which they determine independently. Thus, job creation is minimised.
Construction and installation work will be temporary, with most managerial functions being performed remotely from headquarters. A Georgetown University reportestimates only 27% of DC jobs are ‘technical’.
While the DC discourse mainly focuses on foreign investments, there is little discussion on growing national desires for data sovereignty.
Acceding to so many foreign requests will inevitably block national capacity ambitions to develop end-to-end DC capabilities and not just host them.
Thus far, there is limited interest in the ‘afterlife’ of DCs, such as what happens after they have outlived their purpose, or the disposal of waste materials.
Higher energy and water costs, subsidies, tax incentives and other problems caused by DCs are hardly offset by their modest employment and other benefits.
When I worked at Oracle as a system administrator[1] one of the things I worried about was a disgruntled tape operator breaking open a laser printer cartridge and dumping the toner[2] into the crawl space beneath the computer room floor, where the cold air circulated – my concern was that the toner (which is a very fine powder, containing a magnetic element) would be distributed by the air currents and lodge in the power supplies and connectors and components and interfere with the flow of electricity and bring 4OP4[3] to a halt.
Nowadays I would be more worried about finely powdered carbon[4] or graphite[5], both of which are conductive and extremely common elements, as well as magnetite[6] and probably others.
One baggie full of finely powdered minerals in the wrong spot could bring down an entire data center for weeks.
But, don’t worry, I’m sure the H-1Bs running our data centers are aware of the danger and on guard 24 hours a day.
[1] https://ca-civ393104.org,
https://salanave-runyon.org/herbie.html#06oracle
[2] https://www.echemi.com/community/why-is-laser-printer-toner-magnetic_mjart2204272249_141.html
[3] ‘4OP4’ refers to 4 Oracle Parkway, 4th floor, where the Data Center was located.
[4] https://en.wikipedia.org/wiki/Carbon
[5] https://en.wikipedia.org/wiki/Graphite
[6] https://en.wikipedia.org/wiki/Magnetite
News 🗞️ you can use!
What I miss in these “data center analysis” stories is the great big “security” question. IF (and it’s a big IF) the AI phenomenom becomes as important and all-encompassing as their proponents want to make us believe, these data centers become very attractive targets for terrorists and hostile regimes. A strike on some vital AI datacenter will then paralyze all activity it supports, which might very well have fatal consequences.
So, if this is the case, why do I see so little discussion about the protection of these AI datacenters? I have not seen any indication of wanting to put them underground or in heavily reinforced bunkers, or protect them against attack from the air or even from the ground. Especially if these datacenters are on foreign territory, at least a small army or extensive protection measures seem “de rigueur” to me.
It’s a curious dichotomy. Either AI is crucial and its data centers must be ultra-secure, or AI is ho-hum and any ordinary datacenter setup will do. I have asked this question several times. Answer: crickets.
‘these data centers become very attractive targets for terrorists and hostile regimes’
Are you kidding? If I was in charge of a terrorist organization or a hostile government, I would be sending in combat teams to protect those data centers. Those centers are helping wreck America’s water, electrical and financial systems so I would want them working as long as possible for them to perform their work.
While Rev Kev has a valid point, I want to point out that it isn’t just the data center itself that would need protection. People forget how fragile our electrical grid system really is, and as was discovered in Ukraine, how long it can take to get replacement parts.
Call me paranoid, but a group takes out a series of substations leaving a sizable region without power. Only one or at most two can be repaired without waiting months for the parts. What happens? The data center, because of all the functions that require it that support our oligarchy gets power first. The rest of the region, homes, hospitals, stores, other businesses are left to flounder and in many cases die. Think the aftermath of disasters in Haiti and Puerto Rico, or even the Carolinas only on steroids.
It would be a terrorist’s dream since they would only be the first blow. The killer attack would be the data center(s) being more important than the people.
An EMP would wreck a lot of systems. Multiple EMPs set off at the right locations would do a lot of damage. Oh and there are plenty of instructions online for how to build them
Ah, the return of the EMP! With so many vehicles relying on computer chips,
an adversary could shut down the country with an overhead series of
explosions. It would take days and days to get all the abandoned vehicles
off the roads. Shipping via truck would cease. Even things like kitchen stoves
(I’m looking at YOU, Wolf!) and TV would be effected. But would it count as
a ‘Nuclear’ attack if no one was killed? The stock market would crash,
and, uh, if Congress was out of session, no laws could be passed.
There are also plenty of instructions online for how to build a perpetuum mobile (including an ad running on Rumble/Bitchute). None of them work as advertised.
I sense a multi-billion dollar opportunity here for someone to sell “hardened data center” services. You may not survive the apocalypse but your selfies and AI porn stash will! Some Democrat billionaire should jump on this.
Hardened data centers exist but are expensive and costly to build, and are typically limited to highly classified, top secret, or mission critical functions. Examples include SCADA systems for controlling critical infrastructure like energy transmission or assets, or national security (I’m betting Defense and the CIA don’t just buy compute on AWS). Often they are not connected to the public Internet.
As far as regular cloud data centers go, the usual answer is redundancy. AWS is organized into regions, each of which has availability zones consisting of multiple data centers. If a data center goes out, the system is designed to simply shift load to another one. In practice this can range from seamless to awkward and painful, depending on how well designed the apps are to cope with it, but it’s a fundamental capability of the infrastructure.
Outages at higher levels (availability zones or regions) are more problematic, though rarer. Even in those cases, customers who want extra resilience can pay to deploy in multiple regions and have failover available.
In practice, taking out a single or even multiple data centers would be barely a blip in availability terms. The real vulnerabilities tend to be in software. The Crowdstrike outage ended up taking down practically all of the cloud infrastructure in the world at once, because it was ubiquitous on the underlying devices, and the update with the defect was pushed automatically to all of them at once (actually not quite, and some of the ones who got it later were able to successfully head it off, but it was certainly the intention). In hindsight it was a huge vulnerability and obvious single point of failure, but it wasn’t obvious or easy to see before the fact. Bombs, natural disasters and other region-specific events, on the other hand, are easy to foresee and are typically part of BAU planning for most large enterprises (business continuity and disaster recovery).
A complement:
Many Latin Americans Living Near Data Centers Don’t Feel Welcome in the Future
“Time and again this investigation found that promises of near-future solutions were being made to push through decisions that do not make sense in today’s technology and resource context.”
Thank you, this is an excellent companion read. I’m curious about this,
,
We also reached out to Amazon, a company which announced this year that it would launch a data center region in Querétaro, to ask about the water use of its data centers in this state. The company replied saying that this data center region will use a design that will not use water for cooling. They added that this will get them closer to being water-positive by 2030.
Do waterless cooling systems exist? If they do, I imagine they are expensive.
Here is a global map of data center locations.
What is “water-positive”? Is that a new euphemism for burning hydrocarbons?
It means that we will be getting water credits, to go with our carbon credits.
What’s your water footprint?
Replacing air-to-water with air-to-air reversible heat pumps?
Thanks for the map! I was astounded to see that our pokey little county data center was on it!
At least, there’s a dot close to the location but not exact, and where the dot is, there no building that could house one.
In Cedar Rapids, Iowa, where two DCs are in the works–involving total investment of $1.25T–leaders tout construction jobs over several years and tax payments that lurk somewhere beyond a 30 yr horizon. Meanwhile local subsidies to, um, attract the DC business are in excess of $55M in 20 yr tax breaks.
Data centers: Socialize the costs and impacts. Privatize the profits … but there haven’t been any profits … and there are many more knowledgeable that am I who say the whole thing is bubble. Take a look at the investment streams and contracts. It has been described as a self-licking ice cream cone. There is a graphic circulating on the “intertoobs” that presents this notion. What do we know of data centers from experience? They are power hogs, water hogs, and the noise is to say the least a constant irritation. They are brought to us by our friendly and public spirited “tech bros.” Remember this if, when the lights go out and the tap runs dry.
Hmmm – a bunch of header bullet points followed by short paragraphs, concluding with a list of sources (at the original website).
I may be mistaken here, but this article on the pushback against “AI” data centers appears to have been written by “AI”.
I live a bit south of the Data Center Ground Zero, but the Northern Virginia Data Center virus is spreading rapidly south. My county just approved three out in the western portion of the county where a lot of wealthier residents have moved in the past decade. Lots of outrage growing there. But two counties in my vicinity have both recently declined new data center facilities. Charles City County scrapped its plans after major public outcry and Henrico recently rejected one in the rural east end of its territory. Both were rejected because the communities came out so fiercely against these facilities.
Unfortunately, both of the candidates for Virginia governor do not seem to have any intention of trying to slow or roll back the data center parasites and I expect if the Blue side wins, much of the present opposition will quiet down since data centers are presently perceived as a Youngkin creation.
There was a real flurry of activity for semi-rural data center proposals in western North Carolina. So far the most rural locations have not gotten past county govt. , but there have been a few, nothing in actual development yet. I expect the demand to evaporate with the next crisis level event in marketlandia.
This writer seems to have definitely drunk the Kool-Aid. He sees an ever-growing and expanding demand for data centers, as AI backers would like us to see. My understanding is that there is little to no additional uptake in actual AI use, and that worthwhile applications for this technology have been few. Furthermore, the AI industry as a whole is astoundingly unprofitable, and seems unlikely to be anything else in the next few decades, if ever.
I suspect the problem of data center proliferation is one that will solve itself in short order. In fact, it seems very likely that we will have a tremendous oversupply of unused, half-built, and obsolete data centers beginning in the next few years.
Perhaps an alternative use will become apparent after the AI crash: Air conditioned goat farms? Indoor paintball arenas? Petting zoos?