Data centers are the hot political topic of the moment. They seem to touch on a lot of nerves. There’s the issue of land, affecting rural folks as well as indigenous people. There’s the issue of water, during this age of droughts. There’s the sound pollution and the light pollution, and pollution-pollution (poisoning the air, poisoning the water, poisoning our children, etc).

Don’t get me wrong, I am mad too. The supply chains that allow for the development of data centers rely on the resources and exploitation of the global south. While we bear a burden here, the greatest burden falls on the people of Chile, the Philippines, Kenya, Sudan, the Congo, etc etc etc.

the empire of ai

I just finished reading the book Empire of AI, written by Karen Hao – a journalist who has been covering AI development over the past decade. I have a lot of feelings about generative AI, and Hao’s book left me feeling validated in those feelings. I cannot adequately summarize her work within this post, but I’ll bullet point some of my essential takeaways:

• AI development has taken a very narrow approach that requires an enormous amount of energy. This approach, known as deep learning, comes from the connectionism paradigm within AI research (as opposed to the symbolic paradigm). Hao makes the case that deep learning’s success comes from the fact that models can be trained fast enough to returns on investments, although this training is what requires a huge amount of energy, water for cooling, and countless gpus running across large swaths of land.
• Tech companies like Alphabet, Microsoft, and OpenAI take advantage of special economic zones and other neoliberal schemes in Global South countries like Chile and Kenya to secure access to mining, land for data center development, and cheap labor for training models.
• Tech companies and their CEOs (like Sam Altman) are raising capital on a promise of AGI, even though there are no agreed-upon definitions of intelligence, or any true understanding of what AGI would even look like. They are blindly promising that this technology will solve climate change and all other crisis.
• Tech companies and their lobbyists have put a great deal of effort into promoting a narrative that “rogue” AGI is the greatest threat to humanity, and that the US must develop AGI before a rival state, like China. This narrative has been made dominant over conversations on the real threats AI poses, such as energy and water consumption, ecological destruction, and its destabilizing effects on the economy.
• On the topic of the economy, there’s a lot I’d like to say, but this article by Tabea Krug in Disconjunctions Mag sums it up well: Speculating our Way Through Crisis

This book raised some questions for me. What would AI development look like if it weren’t primarily motivated by profit? What does it look like when technological development serves the people, and not the billionaires? Could a data center be a good thing?

I’m not the only person asking these questions. But you are on my blog! And today, I want to think about my dream data center. It’s kinda like a dream barbie mansion, but better because it houses the cloud.

what is a data center

Data centers have been around for a while, well before the rise of generative AI. They typically contain many humming computers, running services, receiving requests and sending responses. Data centers exist in office buildings, university campuses, and government buildings. They can be as large as several football fields, or as small as a modest room in an apartment.

When we think about our modern software environment, it’s common to refer to the cloud. We store our photos and videos in the cloud, our social media accounts and messages with our loved ones live in the cloud, routing a trip to a new location in a maps application requires the cloud, our email inbox lives in the cloud, so on and so forth.

When we describe these actions as existing within the cloud, it’s easy to look at those fluffy cotton balls floating overhead in the sky and think it’s the same thing. But the cloud is a real place that occupies space on our lovely planet. That is a data center.

Data centers are tucked out of view. They don’t require many employees, so not many people have ever stepped foot into a data center. Truthfully, the obscenely large data centers that we associate with AI development are not pleasant to be in or near, but I find it curious that there are so many unremarkable buildings with nothing but rows of servers in racks, generating heat that does nothing but dissipates.

When I dream about information technology infrastructure, I imagine physical spaces occupied by people in community. Perhaps coworking spaces, libraries, or hackerspaces could coexist alongside the machines. Perhaps the machines could exist within urban greenhouses that utilize the dissipating heat. Perhaps the machines could be utilized in heated walkways during colder months. The point is that we don’t necessarily need to put the machines inside empty warehouses that serve no other purpose.

As I share these imaginings, I do want to warn against seeing a reorientation of what a data center could be as a solution to the problems posed by data centers. Some construction firms that build campuses for AI data centers propose designs that incorporate ideas I mention above, but this does not resolve the issue of resource extraction from the global south, and power and water consumption. It comes down to the fact that these centers are owned by tech oligarchs whose interests are opposed to the interests of the majority of people. The services that they are building infrastructure for serve to further enrich them.

The data centers that I dream about are collectively owned by the masses, and they serve to provide services for the masses. What does this mean?

Find out in part II :->