We’re in Abilene, in West Texas.
At first glance, this might seem like a rather unusual destination for a journey into the world of artificial intelligence. We’re not in Silicon Valley, and we’re not standing in front of the headquarters of one of California’s major tech companies.
And yet it is precisely here that one of the most important infrastructures of the new era ofartificial intelligence is taking shape.
Behind the streets and open spaces of West Texas lies the vast Stargate campus of OpenAI and its partners: an infrastructure designed to provide an enormous amount of computing power to artificial intelligence systems.
The Abilene project covers approximately 1,100 acres, includes eight buildings, and spans approximately 4 million square feet, with a planned capacity of approximately 1.2 gigawatts.
This does not mean that the campus continuously consumes 1.2 gigawatts; rather, it refers to the projected capacity at full development.
But why build a facility of this size right here?
To understand this, we need to step inside—at least in our minds— the Abilene AI Datacenter Campus Stargate
Abilene’s Stargate Campus, where artificial intelligence takes shape
The name ” Stargate ” is now associated with one of the most ambitious programs to expand the infrastructure needed to support artificial intelligence.
The goal is not simply to construct new buildings filled with computers, but to create a network of large-scale infrastructure capable of providing the computing power needed for the next generation of AI systems.
Abilene is one of the first concrete examples of this new reality.
The campus covers a vast area and includes buildings specifically designed to house ultra-high-density IT infrastructure. Its sheer size helps illustrate a transformation that often goes completely unnoticed when we use a chatbot on a computer or smartphone every day.
When we type a question into an artificial intelligence program, all we see is a window on the screen.
Behind that window, however, lies a complex physical infrastructure. It requires buildings, servers, networks, power and cooling systems, fiber-optic connections, and an enormous amount of energy.
This is why Abilene has also become of interest to those who want to understand the future of artificial intelligence without limiting themselves to looking at the software.
Here, AI takes on a tangible form; it is no longer just an algorithm, but has become an industrial infrastructure. And it is poised to grow even further.
OpenAI has, in fact, designated the campus Stargate campus as a project designed to gradually expand computing capacity in the United States through new sites and new investments. The choice of locations such as Abilene also shows how the geography of technology is changing.
New artificial intelligence hubs do not necessarily have to be located in the heart of major tech hubs. Above all, they must have access to space, energy, digital infrastructure, and conditions suitable for building large-scale facilities.
Abilene, with its wide-open spaces and the potential to develop infrastructure on an industrial scale, perfectly embodies this new landscape. But what do we actually find inside these enormous buildings?
Inside the Data Center: The Data Center of the Future
To understand what a modern AI data center is, we can start with something familiar:the data center, which many companies use to house servers and IT systems.
The difference lies mainly in the size.
In a traditional data center, we can find servers, storage systems, networking equipment, and devices used for data management.
In an infrastructure designed for artificial intelligence, however, the amount of computing power required completely changes the scale.
At the heart of the system are computing accelerators—processors designed to perform enormous amounts of mathematical operations in parallel.
A single processor would not be enough. To train and use the most advanced models, it is necessary to connect large numbers of accelerators via ultra-high-speed networks so that they can function as a giant distributed machine.
It’s a concept that’s hard to visualize.
However, we can imagine a data center as a gigantic computer center in which, instead of just a few rows of servers, there are enormous numbers of machines that must operate simultaneously and communicate with one another in a very short amount of time.
The network therefore becomes almost as important as computing power.
Data must, in fact, move quickly between different systems, and the accelerators must be able to work together without creating bottlenecks.
But there is another key factor: all this computing power generates heat.
And this is precisely where an AI data center begins to stand out even more clearly from a typical server room. As computing density increases, so does the amount of heat concentrated in the same space. The cooling system, therefore, is a critical part of the infrastructure.
For this reason, modern AI data centers are increasingly adopting liquid cooling systems, which are capable of transferring heat directly from the components that generate it to the cooling systems.
For the Stargate campus in Abilene, OpenAI opted for a closed-loop, non-evaporative cooling system.
The coolant circulates continuously to transfer the heat generated by computing systems, without relying on the continuous consumption of water through evaporation for cooling. This technical detail becomes important when discussing the physical impact of artificial intelligence.
Because a data center isn’t just made up of computers. It also includes power infrastructure, electrical distribution systems, backup equipment, networks, cooling systems, and control systems.
In other words, a facility like Stargate no longer resembles a corporate computer room. It resembles a large-scale industrial infrastructure built around computing. And this helps us understand why artificial intelligence is also changing the geography of the tech industry.
The New Geography of Artificial Intelligence
For decades, we have associated technological innovation primarily with large cities. Silicon Valley, Seattle, New York, London, and other major urban centers have become synonymous with technology.
However, artificial intelligence is shaping a new landscape.
A large data center cannot simply be built wherever there are many programmers. It requires electricity, space, fiber-optic connections, network infrastructure, and systems capable of continuously cooling the equipment. This is why enormous campuses can spring up even far from traditional technology hubs.
The Abilene AI Datacenter Campus Stargate is one of the most striking examples.
As a result, the new landscape of AI isn’t just made up of offices and labs. It also includes industrial campuses, power lines, substations, fiber-optic cables, and cooling systems.
But this new technological geography isn’t just about the locations where data centers are built. Artificial intelligence also depends on a long chain of raw materials and minerals needed to produce chips, electronic equipment, and energy infrastructure. Another journey—from the dunes of China to the frozen lands of Greenland, tells the story of the race for rare earths and strategic resources in the new tech economy.
Artificial intelligence, which from our perspective seems almost intangible, therefore requires an increasingly larger physical presence.
And the next question is inevitable: How much does it cost, in terms of energy and resources, to keep a facility like the one in Abilene up and running?
How much does it cost to operate the Abilene AI Datacenter Campus Stargate?
The first element is, of course, electricity.
A facility like the Abilene AI Datacenter Campus Stargate—designed to reach a capacity of approximately 1.2 gigawatts —is on a completely different scale than that of a typical corporate data center.
But once again, we must distinguish between the maximum rated capacity and actual consumption.
Electricity is used to power the accelerators and all other IT systems, as well as the networks, storage systems, power supply systems, and the systems necessary to ensure business continuity.
And then there’s the issue of heat.
In fact, almost all of the energy used by computing systems ends up being converted into heat, which must be dissipated from the equipment.
Hence the importance of liquid cooling.
At the Abilene AI Datacenter Campus Stargate, the system described by OpenAI is a closed-loop, non-evaporative system. The liquid is circulated to transfer heat, reducing the need for the continuous water consumption typical of evaporative systems.
OpenAI explained that initially filling the cooling system of each building requires an amount of water comparable to about two Olympic-sized swimming pools. Once the system is filled, the water requirement for cooling is much lower.
But energy consumption isn’t limited to servers.
A campus of this size must have redundant power systems, electrical distribution, emergency infrastructure, and a network capable of ensuring business continuity.
We can therefore understand why a simple question typed into a smartphone is just the latest link in a much larger chain.
So how much does a single request to ChatGPT consume?
There is no one-size-fits-all answer.
A short text-based question requires a different amount of computation than a long and complex task, especially when reasoning, document analysis, images, or videos are involved.
A study published by Microsoft in 2026 estimated a median power consumption of approximately 0.31 Wh per query in the systems analyzed. However, this estimate applies to specific conditions and cannot be generalized to the power consumption of every AI service.
The difference becomes even more apparent when we move from text to images and, above all, to videos.
In fact, generating a video requires a much greater amount of computing power than a simple text-based response. The point, therefore, is not to assign a specific number of watt-hours to each of our questions. It is to understand that behind every artificial intelligence service there is a physical machine that must be powered, cooled, and maintained.
The Water Hoax: The Fake Claim of Half a Liter for Every Question
There’s a saying that has become almost a cliché in recent years: every question asked of ChatGPT supposedly uses up half a liter of water.
It’s one of the most well-known pieces of fake news about artificial intelligence.
However, the figure of half a liter isn’t made up.
This figure is based on a scientific estimate by the University of California, Riverside, which calculated the water consumption associated with using a model like GPT-3 under certain conditions. The estimate was based on a session of approximately 20–50 questions, depending on the conditions and the location where the calculation was performed.
The problem arises when that estimate is converted into a much simpler formula:
One question to ChatGPT = half a liter of water.
Put that way, it’s a fake.
In fact, there is no fixed amount of water that can be associated with each individual request.
Power consumption depends on the cooling technology used, the climate, the data center design, the water source, and how the heat is dissipated.
And, above all, not all of the water in a cooling system is necessarily consumed.
In the case of the Abilene AI Datacenter Campus (Stargate / Lancium), the system described by OpenAI is a closed-loop, non-evaporative system. This does not mean that the campus does not use water.
This means we cannot take an estimate obtained under certain conditions and automatically apply it to the amount of water associated with every ChatGPT query.
The real question, then, is much more interesting: How are the systems designed to dissipate the heat generated by the enormous amount of computing power required to run AI?
From Problem to Resource: District Heating
But that very heat—which is normally something to be eliminated—could become a resource.
The principle is simple: if data centers consume enormous amounts of electricity and almost all of that energy is converted into heat, why not try to recover it?
District heating may be the answer.
The heat generated by IT systems can be transferred via heat exchangers to a water network and, when necessary, brought to the required temperature using large heat pumps. From there, it can be distributed to the buildings connected to the network.
It’s not just a theoretical hypothesis.
In Finland, Microsoft is working with Fortum to develop a system in which waste heat from data centers is recovered to power the district heating networks in Espoo, Kauniainen, and Kirkkonummi. Once fully operational, the project is expected to cover approximately 40% of the annual demand of a network serving about 250,000 customers, with an estimated reduction in emissions of about 400,000 metric tons of CO₂ per year.
The International Energy Agency considers the waste heat from data centers to be one of the sources that can be integrated into district heating networks, especially when the facilities are located close enough to the areas that need heat.
The future of data centers could therefore be different from what we imagine.
Not just massive structures that consume electricity and generate heat, but infrastructure capable—at least under certain conditions—of returning some of that heat to the cities.
The data center could thus become not only a computing machine, but also a component of the urban energy system.
And it is precisely when we talk about water that one of the most common oversimplifications about artificial intelligence comes to light.
Can we tour the Abilene AI Datacenter Campus Stargate facility?
And here’s a question that almost naturally comes to mind when talking about a trip to Abilene. If we were to visit the city, could we tour the large Stargate campus?
As far as the interior is concerned, the answer is no. A data center that houses critical infrastructure is not open to the public. Inside, there are computer systems, networking equipment, power systems, and cooling systems that must be protected both physically and digitally.
The Stargate Abilene AI Data Center Campus does not offer a regular guided tour program open to the public.
However, this does not mean that the trip is uninteresting.
In fact, there is a significant difference between visiting a data center and seeing the location where it is situated. The Stargate campus in Abilene is located in an area that, when viewed from the outside, looks very different from what we imagine when we think of artificial intelligence.
There isn’t a giant sign that says “AI lives here.” But there are large industrial buildings, electrical infrastructure, service roads, and extensive technical areas.
And it is precisely this contrast that makes the place so interesting.
The artificial intelligence we use every day seems like something intangible, confined within a browser window.
In Abilene, however, it takes the concrete form of a massive industrial complex.
To understand what a data center actually contains, one must therefore rely primarily on the materials provided by the operators themselves.
One example comes from Google, which does not offer public guided tours of its data centers for security reasons, but does provide a 360-degree virtual tour of its data center in The Dalles, Oregon, along with images and informational materials.
It is a solution that allows you to do—at least virtually—what would be very difficult to do in person: enter one of the places where the computing power of the new digital era is being built.
For those arriving in Abilene, then, the journey remains primarily an experience from the outside. But perhaps that is precisely what makes it interesting. You don’t need to step inside a server room to understand what’s going on. It’s enough to observe the scale of the buildings, the infrastructure surrounding them, and the amount of energy required to keep them running.
Artificial intelligence, once a seemingly intangible phenomenon, suddenly becomes something physical.
Prices and Artificial Intelligence Tools
If there is a physical cost to running artificial intelligence, there is naturally also an economic cost for those who use it.
The major services have opted for different models, offering free tiers and paid plans.
ChatGPT, for example, offers a free plan and several paid tiers. OpenAI currently lists prices of $8 per month for ChatGPT Go, $20 for Plus, and $200 for Pro, with prices adjusted for certain markets.
Higher-tier plans don’t just mean having a paid version of ChatGPT; they can also offer greater functionality, access to more advanced models, and additional tools.
Claude, developed by Anthropic, also offers a Pro plan for $20 a month in the United States.
Google is taking a similar approach with itsGoogle AI plans, integrating Gemini with other services in the Google ecosystem. In Italy, Google AI Pro is priced at 21.99 euros per month.
In addition to chatbots, there are also services specialized in generating images, audio, and video, often based on credits or monthly limits: the business model changes, but the underlying logic remains the same: Behind every function lies computing power. And behind that computing power lies an infrastructure like the one we just saw in Abilene.
Where will this journey with artificial intelligence take us?
The journey began in Abilene, in front of enormous buildings that, from the outside, might simply look like part of a new industrial district. We took a virtual tour of the data center and discovered that behind artificial intelligence lie energy, networks, processors, cooling systems, and massive infrastructure.
But at the end of this journey, the most interesting question is no longer about what’s inside those warehouses. It’s about what AI will do in our lives. It will be able to help us access knowledge, create, communicate, travel, learn, and have experiences that today require time and resources.
But it could also become capable of creating images, music, videos, conversations, and virtual environments that increasingly align with our desires.
And that’s where the hardest question arises.
Will an artificial life be more fulfilling than reality?
Reality is imperfect, unpredictable, and often different from what we would like it to be. Artificial intelligence, on the other hand, could come to build an increasingly personalized world around us. Perhaps the real challenge, then, will not be merely understanding what artificial intelligence is capable of, but deciding how much we are willing to let it do.
We set out to find where AI lives, and we found data centers, power, cooling, processors, and networks.
But in the end, we discovered that the most important question isn’t where artificial intelligence exists, but where we want to bring it into our lives: if we can build an artificial reality exactly as we want it, will we still need reality as it is?
