<nettime> A Different AI Is Possible. Resist Hyperscalers

[Felix Stalder via nettime-l <nettime-l@lists.nettime.org>]

If you come to Ars Electronica this year, join us at a demonstration 
against the Google Hyperscale Data Center that is being built close by.

When: Fri, 11.09.2026, 15:00 – 17:00 CET
Where: Landhaus, Linz, Landhausbrücke, Altstadtviertel, Innere Stadt, 
Linz, Upper Austria, 4020, Austria
https://www.openstreetmap.org/way/156478885

Below is a text I wrote for Versorgerin, the newspaper of Stadtwerstatt. 
The German text was written without any AI. The version here is an 
edited AI translation. If you think this is a contradiction, read on.

In the text, I try to formulate a position that is less "against AI", 
but more about US-Tech giants. Admittedly, a European Tech giants would 
only be a marginal improvement, but we are so far from having one, that 
we can worry about that later.


all the best. Felix





# A Different AI Is Possible. No Google in Kronsdorf.

Resistance is stirring against the Google data center currently under 
construction in Kronsdorf, Upper Austria, bundled together in the 
Bürger:inneninitiative Rechenzentrum Kronstorf. 
(https://www.koogle.at/). This resistance comes rather late, since the 
project has been in planning for a long time. But only slowly is it 
becoming clear just how massive its scale really is — no longer a 
conventional data center, but a hyperscale facility of the kind built 
specifically for AI applications. The project is gigantic not only in 
terms of land use (up to 50 hectares), it also has an energy demand of 
up to 2 TWh per year — equivalent to roughly three additional large 
gas-fired power plants, or five times all the wind energy installed in 
Austria to date — and a water demand for which there are no binding 
upper limits.

This opens up a field of contestation that is unusual for politics of 
digital technologies. Unlike questions of surveillance, manipulation, or 
monopolization, the problems here are not diffuse, distributed, and only 
cumulatively perceptible. On the contrary: a data center is concrete, 
localized, and immediately visible. The cloud materializes on the 
ground, becoming visible to everyone for what it really is, contrary to 
all the rhetoric of immateriality: a gigantic, industrial 
infrastructure. That makes it something conventional politics can 
actually get a grip on. How this fight is waged could have consequences 
that reach far beyond a single data center, because it touches on 
fundamental questions surrounding AI itself.

Supporters like to portray resistance to the hyperscaler as a refusal of 
the future, an unreflective rejection of everything new. They emphasize 
the scale of the investment, the strengthening of (Upper) Austria as a 
location, the contribution to Europe's "digital sovereignty," and the 
boost for the local IT sector, which is now getting "its own" data 
center and thus more direct access to the future technology that AI 
supposedly is.

## Hollow promises

On closer inspection, many of these promises turn out to be hollow. The 
scale of the investment is indeed gigantic — likely several billion 
euros (exact figures are not available) — but the overwhelming majority 
of that money flows into imported NVIDIA chips, which contribute nothing 
to local value creation. And beyond the servers, data centers are little 
more than banal industrial sheds with oversized power, water, and 
fiber-optic connections. Nor is a data center any kind of commitment to 
a location. Even though the investments are enormous and have an 
infrastructural character, the lifespan of the central component is 
extremely short. Under heavy use, the specialized chips burn out quickly 
and have to be replaced every one to three years. That makes it easy to 
relocate data centers. They can be built quickly, but they can also 
disappear again just as quickly. Everything that would actually be 
needed for sustainable local development — local rootedness, embedding 
in existing systems — is largely absent from the Google project. That 
gives the whole dynamic something of a colonial land grab, in which the 
local population is neither involved nor really benefits. The profits 
are funneled out of the country (Google Europe is headquartered in 
Ireland and pays hardly any taxes), while the (environmental) costs stay 
local. Such a data center contributes nothing to digital sovereignty — 
quite the opposite, thanks to the US "Cloud Act." Here, for once, the 
name says it all: an acronym for "Clarifying Lawful Overseas Use of 
Data." The law stipulates that US companies must grant American 
authorities access to data no matter where the servers are located. 
Strictly speaking, many uses within such data centers are incompatible 
with the European General Data Protection Regulation, especially where 
personal data is involved. And even where that wouldn't be a problem for 
local IT use, physical proximity to an AI data center offers hardly any 
advantage, because for most AI applications the delay caused by data 
transmission is negligible compared to the time needed to process the 
request itself. And in the cases where it isn't negligible — industrial 
automation, for example — processing mostly happens locally anyway. AI 
data centers are used for load balancing, meaning tasks get pushed 
wherever computing capacity happens to be available. Unlike, say, a new 
rail line, an AI data center cannot be expected to give local economic 
development any real boost.

## Real burdens

The burdens, by contrast, are very concrete and very real. The rushed 
expansion of this infrastructure is driving a reversal of the energy 
transition. Many such facilities run on gas turbines. The planned Amazon 
data center in Pecos County, Texas, will, at full build-out, become the 
single largest CO₂ emitter in the United States, emitting more than any 
coal-fired power plant. And even though Kronsdorf will run entirely on 
hydropower, that energy will then be missing elsewhere. New hydropower 
plants cannot simply be built on (Upper) Austria's already heavily 
dammed rivers. Soil sealing, too, comes with serious follow-on costs in 
an era of droughts and floods. The water needed for cooling is drawn 
from an already heavily strained water cycle, or else contributes to 
heating waterways beyond their ecological carrying capacity. This past 
summer we saw power plants shut down one after another because river 
water for cooling was either unavailable or already too warm. Will 
Google shut down its data center too, when that happens?

## Not for or against AI

The fight over this US hyperscale data center is not a simple matter of 
being "for" or "against" AI. Given the breadth of the technology's 
development and the sheer diversity of its applications, such a framing 
makes no sense — it only serves to portray the big tech corporations and 
their visions as the only alternative. But they are not. This fight is 
about how AI gets shaped, so that it works for us rather than against us.

To do that, we need to lay down some basic requirements that must be 
upheld. First: new infrastructure must contribute directly to the energy 
transition — not through some future miracle, but here and now. That 
means it must run on renewable energy and storage systems that are 
additionally provided for the purpose. It helps no one if green 
electricity simply goes missing somewhere else. In addition, the waste 
heat generated must be put to use in a way that displaces fossil heat 
sources, for instance in district heating networks. Only then can we 
speak of a genuinely innovative project. A transparent approval process 
and environmental impact assessments are therefore indispensable. 
Second, these facilities must contribute to Europe's independence. They 
must not be subject to the US Cloud Act. As long as critical 
infrastructure is not operated by companies regulated within Europe, 
new, uncontrollable forms of dependency and incompatibility with 
European law will keep emerging. Hardly any European business could 
otherwise use them in a way that complies with the law. Dependencies 
would be further reduced if we consistently favored reasonably 
transparent, open models over closed ones. Third, resource-intensive AI 
applications must be reserved for those uses that generate real, 
substantial benefit to society. Computing power, too, is a finite 
resource, and it must be handled with care. AI does not need to be 
everywhere, and it must certainly never be allowed to undermine 
democracy by flooding public communication channels with junk and 
rendering them useless.

These basic requirements do not by themselves guarantee that AI will be 
put to sensible use. But they do limit the power of the US tech giants, 
and they open up the chance to develop a different kind of AI — one that 
helps us confront the extreme challenges we face with the tools we 
actually need. And that would be real progress.


--
| |||||||||||||||| http://felix.openflows.com |
| |||||||||| https://tldr.nettime.org/@festal |
| for secure communication, please use signal |

--
# distributed via <nettime>: no commercial use without permission
# <nettime> is a moderated mailing list for net criticism,
# collaborative text filtering and cultural politics of the nets
# more info: https://www.nettime.org
# contact: nettime-l-owner@lists.nettime.org