The Promise and Perils of a Cooperative Data Center in Berlin
Photo credits: Julian Pütz
Artificial intelligence (AI) is expanding virtually everywhere (pun intended), even in physical spaces. Here, AI manifests in data centers. You can think about data centers as ‘the means of production for AI’. They are the buildings that house the servers which host the data and computational resources needed to develop, train and use modern AI systems. Data centers are usually ugly, loud windowless buildings that consume enormous amounts of energy and water. However, because data centers are really, really important for AI they will continue to grow as AI becomes more ubiquitous. In fact, McKinsey estimates that the data center industry will need to grow by around 20% by 2030. Put differently: in 5 years we may require three times the data center capacity we have today!
Currently, the data center industry is dominated by publicly traded, shareholder owned and governed companies. The usual suspects, such as Amazon Web Services (AWS), Microsoft Azure, Baido and Google Cloud are leading the construction of new data centers world wide. Often, these companies use their data centers to develop and train their own AI models, and host them for users. However, usually they also rent out their computational resources to other AI startups. For smaller organizations, it is often a lot cheaper and easier to rent dedicated virtual server space on a service like AWS than to set up their own dedicated hardware, which is incredibly costly and complex to maintain.
The fact that large corporations own the lion share of data centers and that the majority of AI startups are beginning to rely on them for their computational resources, entrenches the concentration of power in the tech industry. It also makes it a lot harder to develop sustainable, inclusive and accountable AI from the bottom up. Ask yourself this; When the computational resources required to train ever more sophisticated AI models begin to consume more energy than entire countries, will financially motivated shareholders choose to curtail growth in the AI porn industry to reduce the environmental burden on our planet amidst a climate crisis? Also, as AI startup demand for scarce computational resources increases, will shareholders choose to keep their cost structure accessible for organizations and initiatives from the global south who often don’t have access to vast pools of capital? And if, in a not too distant future, an AI goes rogue, autonomously undermining democratic discourse or disproportionately targeting people of color in autonomous policing, will shareholders choose to switch it off?
While it’s unclear what the answers to these questions will be, it is clear that we should not solely be relying on shareholder owned and governed businesses to make these decisions. We need more diversity in the ownership and governance of data centers. While huge public initiatives are already responding to this need, I want to explore a third option, i.e. the potential for cooperative data centers in more detail.
A vision of cooperative data centers
What if data centers were cooperatively owned and governed by the people who live in their direct vicinity and who are often also directly served by them? Citizen operators could receive a share of the profit generated by the data center (similar to citizen owned wind farms). This would lead to an indirect redistribution of profits that are made from online user data – an idea that is becoming even more important as user data becomes ever more valuable in the age of AI. Cooperative data centers would also give citizens more control over how these infrastructures impact their local physical environment. For example, citizens could decide if and when the amount of energy or water used by the data center conflicts with other local needs and may decide to reduce the energy supply to the data center in critical situations.
Alternatively, what if data centers were cooperatively owned and governed by other aligned organizations, for example those building cooperative AI, platform and data coops as well as allied public and research institutions like the ICDE? Guided by the cooperative values of equity and solidarity, such a data center may decide to fogo higher profits in order to design its cost structure to be more accessible to organizations from the global south and other cooperatives. This in turn may encourage a broader diversity in AI development and incentivize new AI startups to adopt a cooperative identity. Furthermore, by building with and on cooperatively owned data center infrastructure, cooperative AI startups strengthen their autonomy and independence, giving their members more control over the whole technology stack, including the means of production of AI.
Overall, cooperative data centers could take many different forms and are likely to provide a variety of benefits in terms of environmental sustainability and accessibility. Most importantly, cooperative data centers would add diversity to the market for computational resources, which is essential to reduce global dependency on a few, powerful actors, and create new measures of accountability for AI development.
But, if there’s so many benefits, why don’t we hear about cooperative data centers more? Are they realistic? Do they exist? At scale? And successfully?
GAD eG: learning from the historical case of a cooperative data center
In a somewhat ironic twist, I look to the past to learn more about the viability of cooperative data centers in the futuristic world of AI. Specifically, during my ICDE research fellowship I will conduct a historical case study of GAD eG by drawing on historical primary sources and stakeholder interviews. GAD (Gesellschaft für automatische Datenverarbeitung) eG, was a cooperative that operated various data centers for a network of cooperative banks in Germany between 1963 and 2015. GAD eG makes for a compelling case study, as it operated successfully for over 50 years, scaling both in terms of physical locations and revenue, and adapting to technological innovation. Moreover, GAD eG is interesting as it operated for the financial services industry, thus in a highly regulated sector requiring high standards for computing, data availability and data protection. After growing through several acquisitions itself, GAD eG was dissolved in 2015 after merging with Fiducia IT AG, a corporate competitor.
Through this case study I aim to demonstrate that cooperative data centers are an economically viable and scalable option that is (or was) able to reliably operate in critical industries such as financial services. Furthermore, by conducting a series of interviews with former stakeholders in GAD eG, I aim to explore the unique challenges and opportunities facing cooperative data centers in order to propose a set of policy recommendations for data center startups and regulators, which can facilitate the emergence of more cooperative data centers.
Overall, through this research project I aim to argue for the importance of creating and supporting cooperatively owned and governed data centers, in order to enable cooperative AI to be developed from ‘the bottom-up’ of the tech stack. I am eager to learn with and from anybody who has experience in the field of cooperative data centers or is curious about the idea and its viability.
Learn more about the author here.