The Missing Link in the AI Stack: Why Digital Infrastructure Is Essential to U.S. Leadership

In his first major policy speech, Vice President JD Vance outlined the administration’s initial vision for AI development. His remarks covered key elements of the AI stack from his perspective, including “high-quality semiconductor design and manufacturing facilities, reliable power, computing power, [and] frontier applications.” These are, indeed, key enablers of the stack that will be necessary for the United States to win the global AI race.

Vance’s speech, however, ignored another key enabler of the AI stack, and it is one on which the United States is vulnerable: the networks over which AI traffic travels. This oversight may be attributable to the fact that the non–People’s Republic of China (PRC) telecommunications equipment sector has faced challenges in recent years. When quarterly profits shrink, it is easy to adopt an out-of-sight, out-of-mind reaction. The irony in this development is that the reason for the Western telecom infrastructure industry’s issues is that the PRC undertook a systematic, years-long campaign to kill off non-PRC vendors. Thus, our neglect of a technology that may be critical to winning the AI race against the PRC is due, in large part, to actions by the PRC.

The United States underestimates the key role of networks—and the connectivity that they enable—in the AI race at its own risk, particularly when considering the centrality of networked infrastructure to the diffusion of previous technologies. To grasp this point, we need only try to imagine the Industrial Revolution without the railroad; the modern, global financial industry without the telegraph; or social media, e-commerce, or streaming without the internet.

Shifts in the AI race caused by the release of the DeepSeek chatbot in January 2025 highlight the key role of networks and connectivity. AI is already undergoing a profound transformation as large language models, once believed to be the preserve of massive national champions and the locus of tectonic economic gains, are being open-sourced and commoditized. Now that entrepreneurs will be able to affordably leverage AI applications in bespoke ways, applications run on and information gathered on edge devices will be the battleground on which the AI race is won or lost. Networks will collect data from multiple, geographically dispersed locations, and then transmit that data to locations where model training and model interference occur—and, in many cases, between locations where training and inference occur. These developments will require the United States to have a more advanced digital infrastructure to handle all the data that will be sent to—and from—phones, data centers, vehicles, business computer networks, and all manner of other devices and locations. To maximize the development of AI, the United States must capitalize on the generative interplay between compute and communications. This dynamic is the continuation of a technology trend that began with the development of the Internet and continued with the growth of cloud computing: Gains appear when there is dense clustering of computational and communications infrastructure. There is truth to the belief that the United States’ dominance in these technologies and the industries that grew up around them owes to the fact that the Advanced Research Projects Agency Network was first invented here. It is less well understood, however, that the high-density clustering of communications infrastructure and compute enabled these technologies to penetrate every level of the economy in the United States before the rest of the world.

By way of example, in the 1990s, Northern Virginia hosted the first major cluster of internet exchanges—facilities where different internet service providers (ISPs) could exchange data operated by each other’s networks. For these exchanges to serve as points of interconnection, ISPs needed to lay fiber. The concentrated group of exchanges and fiber made it economically attractive for Internet companies to locate servers in the same area. This created a virtuous cycle of agglomeration in which “more server farms attracted more high-speed fiber, which in turn attracted more server farms, in a self-reinforcing loop.” It is no accident that when Amazon opened its first data center in 2006, it did so in Northern Virginia. This area, known as “data center alley,” still contains the most dense cluster of such facilities in the world.

For AI to flourish, the United States will need communications infrastructure—located in proximity to compute—of substantially greater density, complexity, and scale. This infrastructure will need to be multimodal, reaching every edge device that can potentially leverage AI applications, including phones, cars, drones, wearables, and home appliances. Further, reaching the billions of devices wherever they are located will require a heterogeneous assortment of interoperable communications infrastructure, including mobile networks, fiber, satellite constellations, high-altitude platforms, optical and infrared technology, and eventually quantum communications. Just as we need valves, arteries, capillaries, venules, and veins for blood to reach every part of our body, our society will need an assortment of communication methods to feed compute, regardless of where it resides.

As it currently stands, the United States and its allies and partners have fallen behind the PRC in providing digital infrastructure. The PRC has poured massive subsidies into its national champions, such as Huawei, ZTE, SMIC, Guowang, and Alibaba, investing tens of billions of dollars in Huawei alone. Indeed, Huawei’s research and development (R&D) budget—which will be critical to developing new-generation communications capabilities—is roughly five times the size of two of its major non-PRC competitors. Further, the PRC has proved adept at packaging infrastructure—including telecom, cloud, and various other technology offerings—in ways that non-PRC companies are simply not set up to do.

In contrast, Western companies have acted as individual market participants, and while that is a more efficient model in a pure market economy, that approach has run into challenges in competing with subsidized, state-controlled competitors, particularly in emerging markets. This provided the PRC with an inherent advantage in selling digital infrastructure in third countries, where it could offer a comprehensive, turnkey solution that its non-PRC competitors lack.

As a result, the United States is in desperate need of a comprehensive strategy to bolster its digital infrastructure, as well as that of its allies and partners. While such a strategy will need a variety of pillars and policies, three elements are indispensable. First, the administration should articulate how the United States will build and maintain an industrial base outside of the PRC that can produce the necessary networks and capabilities. Such an industrial base will be composed of numerous, specialized components, including radio frequency equipment, network infrastructure, and power supply equipment, as well as services such as network design, planning, management, and monitoring; installation and deployment of equipment; and data services and cloud integration. Building such an industrial base will require us to consider key enablers such as workforce and education, R&D, capital markets, and strategic finance. Any strategy to build such a supply chain should incorporate close cooperation with allies and partners—even if the United States could, in time and at great expense, construct such networks alone (which is doubtful), it is inconceivable that the United States could prevail over companies such as Huawei and ZTE without the expertise of telecom firms in Europe, South Korea, and Japan.

The United States cannot develop these capabilities unless it adopts policy approaches that support them, including existing initiatives that can be used to advance digital infrastructure and unleash AI innovation. These initiatives include the U.S. Innovation Fund (under the CHIPS and Sciences Act), which the federal government should continue to deploy to advance open and interoperable standards for telecom networks; spectrum auctions, which provide a critical tool in spurring the development of wireless technology in the United States; the Export-Import Bank of the United States financing of 5G deals in third countries, which relaxed content restrictions so that non-U.S. suppliers could obtain financing for digital infrastructure projects; and significant efforts to support research and development of advanced telecommunications capabilities by agencies such as the Defense Advanced Research Projects Agency, the National Science Foundation, and the National Institute of Standards and Technology.

Second, such a strategy must explain how the United States, in cooperation with its allies and partners, will advance its position in digital infrastructure outside its borders—particularly in the Global South. To achieve a global scale in AI, the United States will need to drive AI solutions in major sectors such as agriculture, fintech, and health care in these third countries. And that, in turn, will depend on providing the backbone on which these AI applications will run, including digital infrastructure, chips, and energy. In the case of digital infrastructure, the United States will need a backbone that includes both terrestrial communications (e.g., submarine cables and wireless) and satellites, particularly so that it can reach rural and remote areas.

In the PRC’s effort to accomplish this, it is already leveraging its Digital Silk Road effort—for instance, by reaching an agreement with 26 African countries last summer on a joint action plan with those countries for ten digital transformation demonstration projects. At the same time, we are beginning to see similar deals from U.S. and U.S.-friendly firms. Most notably, G42—a United Arab Emirates–based company that has a partnership with Microsoft, incorporates U.S. tech, and has agreed to rip out PRC gear—recently reached a deal to build a data center powered by geothermal energy in Kenya. The fact that this deal involves a non-U.S. company points to the necessity of incorporating allies and partners, and companies based in those countries, to build the infrastructure backbone the United States will need in the Global South. Given the scale of the investments needed and the hesitance of the U.S. government and some of its largest companies to consistently take risks in such markets, such participation is a necessity.

Facilitating such deals will require the United States to integrate new types of international institutions and commercial partnerships so that non-PRC companies can put together attractive, packaged solutions that are competitive with the PRC offerings. While businesses need to make individual decisions about whether deals make commercial sense, the United States needs mechanisms to enable them to explore all the ways they could work together to offer competitive and comprehensive solutions.

Finally, to facilitate the development of a full AI stack in third countries, the United States will need to update its exercise of one of the oldest tools of statecraft: diplomacy. The United States needs a first-class, strong cadre of tech diplomats. While there have been moves in that direction in recent years, such as the creation of the Bureau of Cyberspace and Digital Policy at the Department of State, the scale of the challenge described above will require a massive retooling. The United States needs diplomats in the Department of Commerce’s Foreign Commerce Service, as well as the Department of State, who are experts on the different components of the AI stack, including telecommunications and data centers, commercial partnerships, and ecosystems for it to flourish outside of the United States, and are able to use their expertise to thoughtfully deploy resources (such as financing) for projects in third countries that will succeed. As part of this diplomatic effort, the United States should push multilateral development banks to prioritize U.S.-friendly standards in exchange for additional financial assistance or more flexible funding. For example, the United States should work with the World Bank to ensure that U.S. companies are competitive with PRC companies when they bid on digital infrastructure using World Bank–financed contracts.

The scale of the challenge the United States faces is significant, and reforms like this will not be easy to advance through Congress or the federal bureaucracy. Unless the U.S. government gives serious attention to finding and implementing these—and other—solutions soon, however, it runs the risk of losing an AI contest in which it has numerous other advantages because it missed the key digital infrastructure ingredient.

Navin Girishankar is president of the Economic Security and Technology Department at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Matt Pearl is the director of the Strategic Technologies Program at CSIS.