Cloud Needs More than Fee Waivers to Be Competitive

Luke Hogg is a senior fellow and Sophia Bulla is a research fellow at the Foundation for American Innovation.

Shortly after coining the term "artificial intelligence" in 1956, Stanford University professor John McCarthy spoke of a time when “computing will be packaged as a public utility just as the telephone system," wherein "each subscriber needs to pay only for the capacity he actually uses, but he has access to all programming languages characteristic of a very large system." His prescient vision describes today's cloud industry, where anyone can buy computing power and storage on demand. As McCarthy predicted, the cloud offers today’s businesses and consumers a breadth of turnkey tools and services that, 20 years ago, companies would have had to build and maintain themselves.

Unfortunately, while businesses benefit from the infrastructure and network effects cloud offers, the similarities to earlier telecommunications systems that were hamstrung by heavy concentration and monopolization do not end there. Like the telegraph and telephone markets before it, a few companies dominate a market that underwrites much of modern commerce. Such concentration, combined with anticompetitive business practices, erodes customer choice, creates lock-in, and leads to higher prices for businesses and their customers. Early efforts to reform the industry have been to little effect, but lawmakers can encourage competition by cutting through the various product bundles hyperscalers employ.

For years, regulators concerned with this obvious market concentration have investigated ways to improve competition. One early focus was the data transfer fees companies charge when customers switch cloud providers. Typically, companies face "egress fees" any time they move large amounts of data around or off the cloud. Downloading, streaming, and sending data are all forms of egress—and so is moving to a competitor. In 2024, to get ahead of EU mandates and alleviate pressure from the Federal Trade Commission, three major cloud service providers—AWS, Microsoft's Azure, and Google Cloud Platform (GCP)—all began offering fee exemptions for customers looking to exit their cloud. But this move, which at first glance seems like a major win for competition, actually has little practical effect for businesses or their consumers.

Cloud providers have historically charged a high markup for moving data around. While the cost of bandwidth and storage has plummeted over the last several decades, hyperscalers' egress charges have dropped only modestly. Still, the effect of such fees on deterring customers from moving to a different cloud are minimal. Small businesses have always been able to move data freely; large businesses can typically either negotiate egress fees down or employ techniques such as data compression to drastically reduce the bandwidth of large data transfers. A report prepared for the European Commission concluded that one-time data transfer costs represent less than 1 percent of yearly cloud expenses. This is hardly enough to dissuade a customer from leaving when compared to the costs of reconfiguring data, running systems in both clouds during the transfer, and paying for duplicate software licenses. Even if data transfer fees really did inhibit switching clouds, companies move data slowly to ensure continuity, not within the small window the egress fee rebates mandate. And most move only the data components that could benefit from rivals' proprietary tools, not the full cloud exit the rebates require.

If eliminating egress fees is proven to have had practically no effect, policymakers will do well to focus on the more insidious anticompetitive practices in the industry, particularly the way products are bundled and sold. Unlike attacks on data transfer fees or calls to improve interoperability, which can be technically complicated, the bundling of services offered by highly vertically integrated companies is primarily designed to ensure customer loyalty.

For example, major cloud providers run marketplaces for third party software. Software companies pay a small commission on sales made through the marketplace in exchange for simplified billing and customer acquisition. Customers can purchase software outside the marketplace, but it usually makes no sense to do so, since hyperscalers offer tiered discounts the more customers purchase through their cloud marketplace. These marketplaces themselves are not anticompetitive, but the product bundling can be: hyperscalers force the hand of software companies to host on their clouds while tying the success of competitors in the adjacent software-as-a-service market to their own.

In addition to bundling complementary products, some cloud providers also tie their own legacy products. For example, Microsoft's popular database and office productivity software are notoriously cumbersome to run in public clouds other than Microsoft’s own Azure cloud. Microsoft products on AWS or GCP are subject to higher, more complicated price structures and suffer from more limited security updates compared to the same products running on Azure. Waiving egress fees is all well and good, but it does nothing to address bad licensing practices that trap customers into being locked into a particular cloud environment.

AI's contribution to market dynamics is still playing out, and OpenAI's recent expansion to clouds other than its partner Azure are a welcomed move away from tying. A crowd of neoclouds have burst onto the scene, becoming key contributors to frontier labs' ability to perform the taxing compute needed for model training and inference. They join more established tech companies, such as Meta and SpaceX, who have recently expanded into the cloud space as they ramp up capacity needed for their own AI ventures. The shake up is an encouraging sign for those concerned about market concentration, but AI could just as easily cement incumbents if models become another means of product bundling through marketplace discounts and exclusive or priority model access.

Much digital market regulation hinges on how companies should compete in the markets they platform and whether they must offer their products to rival platforms on equal terms. In that sense, Microsoft charging high cross-cloud licensing fees for its software and Amazon replicating its marketplace's most successful products are nothing new. Yet cloud, which sits at the intersection of the digital and the physical, also faces the competition constraint common among hardware suppliers—namely, that multibillion dollar infrastructure build-outs raise the cost of entry to a point where the only viable competitors are firms already dominant in adjacent tech markets. That makes it all the more important that policymakers crack down on bundling, whether it be third-party software in marketplaces, legacy products through restrictive licensing, or AI models through priority partnerships.

Requiring easy and free switching between telecom providers was insufficient to address the concentration of the Baby Bells during the 1980s and 1990s. At best, eliminating egress fees is benignly unproductive in the effort to improve competition in the cloud. If the visionaries of cloud saw it as an evolution in the telephone system, then it may require all the more imagination to ensure that it is more resilient and open than the networks that preceded it. This means making sure competition and merit, not market position, drive the software and services up and downstream of the cloud.