They're building data centers pretty much everywhere these days. The sprawling, windowless buildings are the physical engines of the internet and the cloud; more are now under construction than ever before, and the new breed is bigger and hungrier. A typical center used to consume 10 megawatts of electricity; now they're being built to suck up 10 times that much. Last year all the data centers in the world had room for 10.1 zettabytes of information — roughly 456 billion Wikipedias. And with the rise of artificial intelligence, which requires vast quantities of data and power, the global capacity of data centers is expected to double by 2027. If you don't live near a data center, you will soon.
But cloud computing and AI aren't the only things driving the push for "hyperscaled" data centers. About 65% of the capacity in global centers is owned by just three companies: Amazon, Google, and Microsoft. Like the railroad magnates of old, they're racing to control the market, because they understand something that has eluded the rest of us. Data centers are more than just vast digital warehouses. They're the essential infrastructural technology on which pretty much every other company in the world must run.
When companies need pretty much any computing service these days — networking, security, data processing, platforms, you name it — it's easier and cheaper to just rent it from Amazon Web Services, Google Cloud, or Microsoft Azure. The more data centers those companies have, the more of those services they can offer, and the more storage and number-crunching capacity they can provide. By trying to corner the market on data centers, they're not just creating bigger warehouses for data — they're aiming to be a one-stop shop for all of the tech a company needs.
That's even more true of AI startups. When an innovative newcomer needs access to the large language models that are required to train and run generative AI, they pretty much have to go through Big Tech to get them. And now the tech giants are making venture investments in those startups by offering them "credits" for using the company's cloud. That's how Microsoft made a chunk of its investment in OpenAI, for example — by giving the startup access to its data centers. It's a lucrative inducement to join a proprietary ecosystem.
"This is where the real business is," says Cecilia Rikap, an economist who is the author of a new report called "Dynamics of Corporate Governance Beyond Ownership in AI." "The more AI is consumed, there's more cloud consumption, and therefore not only more money for these companies but more digital technology that is intertwined and tangled inside their infrastructure."
And that entanglement is what worries many economists and legal scholars. Regulators call the problem "locking in." Changing from one data ecosystem to another isn't like moving your office to a new building; the programming interfaces between Microsoft Azure, say, don't just port over to Amazon Web Services. Getting into one is easy, but like the Hotel California, you can never leave. Once a tech giant gives a startup access to its cloud services and its large language models, it has pretty much assured itself a form of control over a fledgling firm that might one day have grown into a competitor. "Market leaders benefit from early-mover advantage coupled with network effects and high switching costs that lock-in customers," a congressional subcommittee warned in a 450-page report back in 2020. The rush to build data centers is, in no small part, a move by Big Tech to secure the keys to the coming AI kingdom.
In the short term, the rise of data centers has actually been a good thing for startups. "Until recently, the perception among academics was that the rise of cloud computing was great for startups and innovation," says Matthew Wansley, a law professor at Yeshiva University who studies competition and regulation. "It used to be that if you were a startup, you had to build your own servers. That's a huge, fixed up-front cost."
That's not true anymore. The price of cloud-computing services has fallen every year since 2006, when Amazon opened its cloud. And it absolutely crashed in 2014, as a team of economists noted, when Microsoft and Google started advertising their competitive prices. From 2010 to 2014, AWS database prices dropped by 11%. Over the next two years, they plunged by 22%.
Cloud computing also made it easier for startups to get funding. Venture capitalists adopted a "spray and pray" approach to investing, meaning they placed bets on more companies but put less money into each one. They also ratcheted back their direct involvement in running the companies, trusting the marketplace to sort out the winners from the losers.
The whole scene has been especially great for AI startups. "Smaller companies like us could get access to compute power and the scalability that the larger service providers offer," says Jonas Jacobi, CEO and cofounder of ValidMind, a fintech company. "You have a few large players dominating the AI space, but there are startups trying to compete with them as well. The only reason they can is because of the cloud vendors."
The trick, Jacobi says, is to write code that can work with any of the three providers, so you don't get locked in to a single company. You have to stay "neutral to the tech stack," he says. Sure, one of the tech giants can always swoop in and build their own version of your software. There's data suggesting that Amazon has made it a standard operating procedure to "engulf" the products of small, open-source competitors and repackage them as part of its own suite of services, as it did with the Elastic search engine. "But that's part of the journey as a startup," Jacobi says. "It's just up to us as a company to be faster and nimbler."
But over time, economists warn, nimble won't be enough. In the battle to create foundational tech — the "key complementary assets" of the business — AI startups will inevitably lose out to the tech giants that control the data centers. "AI is a general-purpose technology," says Rikap. "It's being applied to everything. But what type of AI we get and what type we don't get is going to be affected by the power of just three companies. It's an intellectual monopoly. What they are controlling is data and knowledge." By locking startups into their systems, Google and Amazon and Microsoft can effectively play favorites, offering better deals and cheaper services to the companies in which they have the largest stake.
Over time, economists warn, AI startups will inevitably lose out to the tech giants that control the data centers.
Rikap has also found that their growing control of data centers also gives Big Tech an incentive to work together to share information and protect their joint interests. In a paper with Bengt-Åke Lundvall, an economist at Aalborg University in Denmark, Rikap notes that articles in technical and academic journals from researchers at Microsoft, Google, and Amazon consistently had coauthors employed by their competitors. Now, for sure, computer science is a small world. But the joint authorship, Rikap says, is "a pure way to tell they are collaborating and know what each other are doing" — a hallmark of anticompetitive behavior.
For the moment, there is still reason to hope that innovation can win out over monopolization. Amazon, Google, and Microsoft are still competing on price and features, which is good for everyone. And in Europe, where regulators are taking a more aggressive approach to tech generally and cloud computing in particular, the Big Three are busy pointing fingers at one another. A Google Cloud exec recently denounced Microsoft as a "monopoly" and a "walled garden," and a trade group that includes Amazon filed an antitrust complaint over Microsoft's cloud-computing licenses. As they vie for market share, the companies aren't in lockstep yet — and that creates an opening, albeit a small one, for nimble, faster competitors.
There's also a tendency, over time, for mature technology companies to shift from trying to innovate themselves to simply charging other people who innovate. Among economists, that's known as "rent-seeking behavior," and it looks an awful lot like what Amazon, Google, and Microsoft are doing with cloud computing and data centers.
So what's the best way to make sure Big Tech doesn't use data centers to short-circuit innovation? Researchers point to Google, which is offering a friendlier kind of partnership to startups. "The Google Cloud Division partners with promising database start-ups, contributes to open-source projects, and collaborates with open-source foundations," two scholars recently observed. It's an "architecture of participation," they say, that enables Google to profit while fostering the growth of new companies and ideas.
Even more important, the Federal Trade Commission, aware of the threat posed by data centers, has ordered the Big Tech companies to hand over information on their AI investments. Just as new laws eventually caught up to the pricing practices of the railroads in the 1880s, today's regulators may well catch up to the futuristic, technological tangles of cloud computing. One reason to think so: The lead author of that 450-page House subcommittee report about Big Tech's anticompetitive behavior was a lawyer named Lina Khan. Today's she's the hard-charging head of the FTC.
Adam Rogers is a senior correspondent at Business Insider.