The New Front in the AI War: Data Centers

Regular readers of this column know that I’ve long been skeptical about the appeal of data center investments. I used to argue that data centers consumed an excessive amount of Turkey’s scarce energy resources and created little employment beyond security and cleaning staff. However, my perspective has changed. John Maynard Keynes once said, “When the facts change, I change my mind. What do you do?” Let me explain why I’ve changed mine.

In recent weeks, several countries have announced substantial investments in data centers. For example, U.S. President Donald Trump unveiled a Stargate project shortly after taking office. Most of the $500 billion allocated will go toward building data centers. Twenty facilities, each measuring 45,000 square meters — approximately the size of seven football fields — will be constructed. Meanwhile, the United Arab Emirates has revealed plans to invest between $30 and $50 billion in data centers in France.

During the Asia-Pacific Economic Cooperation summit in Peru, Chinese President Xi Jinping stated that China’s domestic laws would align with global standards to attract data center investments. Imagine — even China, which restricts access to Google and WhatsApp, is loosening up regarding data centers. In India, states are competing to offer the best incentives. Morocco recently announced the establishment of a unique industrial zone to attract Europe’s data center investment businesses.

New AI Models Are Based on Reasoning

So, what suddenly made data centers such a hot investment? The answer lies in the growing demand created by large language models — like ChatGPT — in the past two years. These models operate on massive data centers. Until recently, these facilities were primarily used to train AI systems with text and image data. Now that much of the training is done, new models — like the latest versions of ChatGPT — function on reasoning-based logic.

What does “reasoning” mean? It involves solving math problems, understanding algorithms, and writing computer code. In other words, it goes beyond simply repeating the texts used in training. These models are now beginning to outperform humans even in math olympics.

Models that reason through inference now use data centers not for training but for responding to user queries. The model should be near a well-functioning data center with robust processors to ensure rapid responses. While planning the computing power needed for training is possible, predicting when and how many real-time inference requests will come is impossible. That’s why data centers need to be constantly ready to go.

What happens if they’re not? Response times increase. Maybe you’ll tolerate a delay when asking your chatbot, “Where’s my food?” (Though I don’t.) But for autonomous driving systems, decisions must be made instantly.

Also, building a data center isn’t just about stacking bricks. These centers contain chips for various purposes. NVIDIA makes the most advanced ones — and they’re not something you can pick up off a store shelf. First, you must pay and get in line. More importantly, some chip models are subject to U.S. export restrictions. Under an executive order issued in the final days of the Biden administration, Turkey did not make the list of 18 “trusted partners and allies.” For countries on that list, chip sales are unrestricted. North Korea, China, and Iran are outright banned. Turkey and 17 NATO members are placed in a category where sales are allowed under certain conditions and in limited quantities.

Right now, Sheikh Tahnoun — the brother of the UAE ruler — is lobbying to remove his country from these restrictions. Last year, the UAE-based company G42 was forced by the U.S. government to partner with Microsoft. This partnership is now building its first major data center in Kenya. As you can see, even when you sit at the table with a private company like Microsoft, you’re sitting across from the U.S. government. That’s how the data center game works.

Companies That Learn to Build Data Centers Can Work Worldwide

So, if we are serious about our AI ambitions, we cannot afford to overlook the construction side of this business. Data center projects could receive low-interest project financing from the public sector. We should also attract investment from allied countries. Public-private partnerships must take nimble diplomatic steps to avoid the impact of export restrictions.

If Turkey learns to build data centers, our construction firms could compete worldwide. For once, in the realm of AI, the question came from a familiar area — construction. Let’s not miss this opportunity.

This article is a translated version of “Yapay zekâ savaşının yeni cephesi: Veri merkezleri” which was initially published in Economic Daily (Nasıl Bir Ekonomi Gazetesi) in Economic Daily on March 21, 2025.