New infrastructure loading 2- Computing capacity: The strategic resource of AI

“Every communication revolution is, at the same time, a new infrastructure revolution.”

In the first article, I argued that the era of artificial intelligence (AI) is not just about producing new software or more advanced algorithms, but represents a new infrastructural transformation built upon data centers, energy systems, semiconductors, and digital networks. If this transformation truly forms the basis of a new regime of capital accumulation, then we must ask the following question: What is the fundamental factor of production for this new infrastructure?

For many years, one of the most common discourses explaining the digital economy was the phrase “data is the new oil.” With the rise of digital platforms, it was argued that economic power depended on the capacity to collect and process data. However, the rise of generative AI shows that this narrative is not sufficient on its own. Because data cannot turn into economic value without the computing capacity to process it. Today, what is decisive in the AI economy is the ability to build the computing infrastructure that will train, process, and continuously run data.

Here, “computing capacity” is not just a technical concept expressing the processing power of computers. Computing capacity is an integrated production capacity that encompasses advanced semiconductors, graphics processing units (GPUs), data centers, high-speed networks, energy infrastructure, cooling systems, and the capital that finances them. Therefore, the element that creates value in the AI economy is the material infrastructure upon which these algorithms run.

At this point, it is necessary to rethink an important reminder that communication and media studies have long emphasized. Communication technologies are often discussed through content, platforms, and user experiences. Yet, as media infrastructure researchers such as Lisa Parks, Nicole Starosielski, and Shannon Mattern have shown, the real transformation of communication often takes place in infrastructures that remain invisible. Digital communication is impossible without submarine fiber optic cables, data centers, satellite systems, base stations, and network architectures. Content is visible; however, the infrastructures that make them possible are often invisible.

We are facing a similar situation in the AI era. Public debates largely revolve around how smart the models are, which company is developing more advanced artificial intelligence, or the social impacts of new applications. However, behind this visible layer, a much more comprehensive infrastructural transformation is taking place. Data centers are growing, electricity consumption is rising rapidly, semiconductor production is gaining strategic importance, and global communication networks are now carrying not only information but also large-scale computing processes.

However, this infrastructure is also a story of invisible labor. As communication researcher Christian Fuchs' “digital labor” approach shows, digital technologies do not rely solely on the labor of software developers; they are built upon a global labor chain extending from mines to chip factories, from data labeling processes to the operation of data centers. Although AI is often described as a symbol of automation, there is a digital labor ecosystem in the background that is wider and more complex than ever before. The worker extracting cobalt in the Congo, the miner producing copper in Chile, the technician producing chips in Taiwan, the worker packaging chips in Malaysia, the employee labeling data in Kenya, the content moderator in Dublin, the data center engineer in Virginia, and the AI researcher working in Silicon Valley all form different links of the same digital labor chain. No matter how digital artificial intelligence may appear, there is a highly material geography of production and labor organized on a global scale behind it.

For this reason, the difference between the internet and AI is not just technological. The internet was primarily a communication infrastructure that accelerated the circulation of information. AI, on the other hand, creates a new computing layer rising above this communication infrastructure. Thanks to cloud computing systems, distributed data centers, and high-performance computing clusters, digital networks are becoming global production infrastructures that perform computation. In other words, communication infrastructure is increasingly transforming into computing infrastructure.

The transformation that began with technology entrepreneurship is also changing the way capitalism is organized. Carlota Perez's work on technological revolutions shows that every major technological transformation creates a new cycle of infrastructure and investment. Just as railways, electrical grids, and the internet became the fundamental infrastructures of their own eras, computing capacity is assuming a similar role today. When considered alongside David Harvey's “spatial fix” approach, this helps explain why global capital is turning toward data centers, energy infrastructure, and semiconductor production. By investing in new technologies, capital is simultaneously building the infrastructure that will enable the production capacity of the future. For this reason, reading the competition in the AI field solely through the models developed by companies like OpenAI, Anthropic, Google, or xAI will remain incomplete. The real competition is knotted around the question of who possesses the computing capacity to train these models. Microsoft's Azure investments, Amazon Web Services' data center expansions, Alphabet's energy deals, or NVIDIA's strategic position are all different parts of the same infrastructural transformation. In the AI economy, power is concentrated in those who control the infrastructure that makes the operation of algorithms possible. Therefore, explaining computing capacity with the oil metaphor seems increasingly inadequate. Oil is a strategic raw material used in specific sectors; whereas computing capacity is increasingly resembling electricity. Electricity is a general-purpose infrastructure that permeates almost every sector of the modern economy. As AI becomes a general-purpose technology spreading to very different fields from healthcare to finance, from defense to education, computing capacity stands out as the fundamental input of this transformation.

Perhaps we now need to ask the question “Who will win the AI race?” in a different way. Before asking who will develop the more advanced model, who will build the computing capacity to run these models? Because in the AI era, economic and geopolitical power is increasingly concentrated in the infrastructure that makes software possible. For this reason, in the next part of the series, I will address data centers, one of the most visible but least discussed elements of this infrastructure, as the new factories of digital capitalism.