Infrastructure Is Destiny: The Geography of the Next Technology Race
Executive Summary
The real technological race is not over who invents the next breakthrough application. It is over who builds the infrastructure that allows innovation ecosystems to form, scale, and persist. Communications and computing infrastructure increasingly function as general-purpose economic inputs, much like electricity or transportation networks. Their value depends not only on deployment, but on complementary assets such as skilled labor, capital, energy availability, and regulatory environments that support experimentation and growth.
This issue brief argues that next-generation technological leadership will depend on a converging infrastructure chain that includes wireless spectrum, terrestrial broadband, fiber backhaul, satellite systems, edge computing, hyperscale data centers, and AI accelerators. These layers are interdependent: spectrum without backhaul creates stranded capacity; data centers without fiber become isolated facilities; edge computing without wireless access cannot effectively serve users.
Innovation also remains geographically concentrated. Breakthrough inventions, venture capital, technical talent, and specialized infrastructure continue to cluster in jurisdictions that reduce deployment friction and support complementary ecosystems. Artificial intelligence (AI), augmented reality (AR), virtual reality (VR), and immersive computing intensify these dynamics because they require low latency, high-capacity connectivity, large-scale compute resources, and reliable power infrastructure.
At the same time, permitting delays, fragmented rights-of-way rules, environmental-review requirements, pole-attachment disputes, and energy constraints can significantly slow infrastructure deployment. As networks densify toward 5G, 6G, and edge-computing architectures, these transaction costs become more consequential. Indeed, reducing deployment friction may produce larger long-run innovation gains than targeted subsidies for favored technologies. Performance-based, technology-neutral policies are generally better suited to this challenge.
The central policy lesson is that infrastructure investment is the most effective form of horizontal industrial policy. Jurisdictions that make it easiest to build, connect, compute, and experiment—and that cultivate the complementary ecosystem of skills, capital, and institutions—will attract the firms and innovation clusters that define the next digital economy.
I. Infrastructure and the New Geography of Innovation
Public debate frequently warns that the United States risks “losing the race” to China in technologies such as 5G, artificial intelligence (AI), augmented reality (AR), and next-generation wireless systems. That framing often implies that a rival nation can “capture” a technology market, as though 5G or AI were finite prizes one jurisdiction wins and another loses. But these technologies are not zero-sum product markets. They are general-purpose infrastructure layers whose value depends on deployment breadth and ecosystem depth.
The “race” metaphor nevertheless points to a real concern. Jurisdictions that fail to deploy the underlying infrastructure risk losing the developers, investors, and firms that build on top of it. The danger is not merely losing market share. It is losing the locus of innovation—the places where the next generation of products and services is conceived, tested, and scaled. Technological leadership depends not only on who invents a technology, but also on where the infrastructure exists to commercialize it.
That dynamic has become more important as the computing frontier converges around common infrastructure requirements. Generative AI, agentic systems, and immersive augmented- and virtual-reality applications increasingly depend on ultra-low latency, distributed computing capacity, and high-capacity wireless networks. Where infrastructure gets built—and how quickly—now helps determine who participates in the next wave of innovation.
This is not a new insight. Despite decades of falling communication costs, technological innovation remains geographically concentrated. Roughly 56% of the most consequential new technologies emerge from just two U.S. corridors: Silicon Valley and the Northeast. The geographic diffusion of related employment then takes roughly 50 years.[1] Far from dissipating, this concentration has intensified. Tech clusters capture an increasing share of U.S. patenting activity—not because knowledge itself is rivalrous, but because the complementary inputs required to produce innovation, including talent, capital, and specialized infrastructure, remain scarce and geographically concentrated.[2]
The persistence of place-based clustering despite dramatic improvements in connectivity suggests that infrastructure alone cannot explain innovation geography. The interaction among infrastructure availability, deployment regulation, and the self-reinforcing dynamics of agglomeration also matters.[3] Innovation ecosystems cluster in jurisdictions that make it easier to deploy communications and computing infrastructure, experiment with new architectures, and scale digital platforms.
Knowledge spillovers—the external benefits that arise when ideas and know-how spread among nearby firms, workers, and institutions—also remain geographically localized. They fade slowly over time and remain bounded not only by physical distance, but also by jurisdictional lines.[4] Differences in permitting requirements, energy regulation, spectrum allocation, and procurement practices create policy discontinuities that can segment innovation networks even among nearby regions. As deployment frictions rise, investment and experimentation migrate toward jurisdictions that impose fewer barriers to value creation.
Next-generation computing architectures—including AI inference, immersive computing, and edge computing—depend on dense infrastructure capable of supporting low-latency connectivity and massive computational throughput. But the relevant infrastructure extends well beyond the endpoint connection. Communications networks, including fiber backhaul, satellite systems, and wireless access, increasingly converge with computing infrastructure, including data centers, edge nodes, and accelerator clusters. Technological leadership therefore depends on the full infrastructure stack, not merely one layer.
The familiar “race to China” framing from industrial-policy debates is therefore not entirely wrong. But infrastructure’s economic value does not come from its mere existence. It comes from the local complements—human capital, organizational capacity, specialized services, and supporting institutions—that turn capacity into productivity. A jurisdiction that deploys fiber, spectrum, and edge computing capacity, but fails to create the conditions for those complements to emerge, will not capture the innovation gains that infrastructure enables.
In that sense, losing a technological race is not principally a question of whether a country builds infrastructure. It is a question of whether it builds in ways that allow local ecosystems to scale, coordinate, and reinforce one another.
This issue brief maps how these infrastructure layers interact and identifies the regulatory and economic dynamics that will determine whether the United States remains at the center of the next generation of technology ecosystems. It is the first in a series examining the infrastructure domains—including spectrum, deployment barriers, satellite connectivity, and broader policy design—that will shape the emerging competitive landscape.
II. Infrastructure as a General-Purpose Technology
Communications and computing infrastructure increasingly function as general-purpose economic inputs, analogous to electricity or transportation networks.[5] The general-purpose technology (GPT) framework helps explain both the breadth of infrastructure’s economic effects and the long lag between deployment and measurable returns. Communications networks exhibit all three canonical GPT characteristics: they are pervasive across sectors, capable of sustained technical improvement, and highly complementary with downstream applications. The internet’s status as a GPT is now well-established.
What distinguishes the current transition from earlier rounds of edge-network evolution—such as content-delivery networks, flatter peering arrangements, or Differentiated Services-era traffic management—is its scale and integration. Computation and intelligence are migrating outward from centralized facilities toward the network edge. AI inference at cell sites, programmable network slicing, and compute-aware routing increasingly distribute computational capacity throughout the network itself. Communications infrastructure is evolving from a passive conduit for transmitting data into an active computing platform.
Whether this converged “network-as-compute-platform” model has fully achieved GPT status remains debatable. What is clear is that the infrastructure chain’s geographic dependence is intensifying. Each layer increasingly relies on physical proximity to the others to satisfy latency, bandwidth, and computational demands.[6] These characteristics create increasing returns and a coordination problem: the network’s value depends on the density of applications built on top of it, while developers’ incentives depend on the network’s quality, scale, and reach.
New infrastructure therefore rarely generates immediate productivity gains. Firms and workers must reorganize production, retrain employees, and redesign business processes around the new technology. Those complementary investments take years to complete and even longer to appear in measured output.
Economists describe this pattern as the “productivity J-curve,” which helps explain the “Solow Paradox” identified by Robert Solow in 1987: one could “see the computer age everywhere but in the productivity statistics.”[7] The paradox becomes less puzzling once the lag between deployment and measurable productivity gains is understood as a predictable feature of GPT adoption, rather than evidence of technological failure.[8]
A complementary explanation may also exist. Organizations often appear to absorb efficiency gains into coordination, task-switching, and workplace slack, rather than converting them directly into measured output. Researchers have documented this pattern across multiple waves of automation technology.[9]
Historical experience supports both explanations. The productivity gains from electrification emerged only decades after initial deployment, and jurisdictions differed substantially in how quickly they captured those gains. Organizational restructuring, competitive conditions, and regulatory environments all shaped adoption.[10] The same pattern appeared with steam power, electricity, and information and communications technology (ICT): each produced substantial economic gains, but only after long adjustment periods, with innovation activity concentrating in early-deploying jurisdictions.[11]
These complementarity effects are also deeply geographic. The inputs necessary to exploit a GPT—including skilled labor, organizational capital, specialized suppliers, developer communities, and venture financing—are themselves geographically concentrated. The returns from infrastructure deployment therefore depend partly on the local density of these complementary assets.
Jurisdictions that deployed earlier generations of communications infrastructure enter each successive technology wave with compounding advantages.[12] Industry-level evidence confirms that ICT-using industries—not merely ICT-producing industries—drove the productivity acceleration after 1995, and that total factor productivity growth correlated with ICT investment during the preceding decade.[13] Cross-country evidence on intangible investment further suggests that jurisdictions with greater organizational-capital formation extract disproportionately greater value from equivalent levels of infrastructure deployment.[14]
A. Agglomeration and Innovation Geography
Digital technology reduces five categories of economic friction: search, replication, transportation, tracking, and verification.[15] But not all frictions matter equally for innovation geography. Agglomeration economies—the benefits firms derive from geographic concentration—operate through three primary mechanisms: sharing, matching, and learning.[16] Each has distinct empirical implications.
Of these mechanisms, learning appears most resistant to digital substitution. Frontier innovation depends heavily on tacit knowledge transfer, informal exchange, and iterative feedback loops that rely on trust and high-fidelity interaction.[17] Infrastructure that lowers latency and enables real-time collaboration may therefore have different geographic effects than infrastructure that merely increases bandwidth.
The empirical evidence on broadband’s economic effects—including employment growth, productivity, and innovation—is similarly nuanced. Broadband consistently supports marginal product and process innovation through technology adoption and diffusion, even if it does not directly generate breakthrough invention. That distinction matters. Breakthrough technologies may emerge disproportionately from dense innovation clusters, while the marginal gains from deploying communications infrastructure may be greatest in areas where connectivity was previously weak or unavailable.
Broadband’s economic effects also vary substantially by region. Employment gains tend to be strongest in lower-density areas, rather than major urban markets. Productivity effects, meanwhile, depend heavily on pre-existing human capital and technical expertise.[18] Broadband investment generally produces stronger productivity gains in regions with deeper knowledge stocks and more highly skilled workforces.
Communications infrastructure is often evaluated primarily through a bandwidth lens—how much data can move through a network per second. That focus risks obscuring other network characteristics that shape innovation geography. Latency, reliability, edge-computing capacity, and network density can matter just as much for the formation of innovation clusters.
Broadband deployment can increase innovation activity even without contemporaneous productivity gains.[19] That pattern is consistent with the J-curve dynamics associated with GPTs, in which complementary investments and organizational adaptation precede measurable productivity improvements. Innovation clusters may therefore begin forming before conventional economic metrics register meaningful gains.
Historical evidence points in the same direction. A 10% increase in mobile penetration has been associated with a 0.59% to 0.76% increase in GDP per capita, with larger effects in countries with stronger human-capital bases.[20] Venture-capital geography likewise illustrates how infrastructure and complementary ecosystem inputs reinforce spatial concentration.
Venture-capital investment remains highly clustered because investors face substantial information asymmetries and benefit from proximity-based monitoring. Reduced travel time between venture-capital firms and portfolio companies can improve innovation output and increase exit probabilities.[21] Conversely, when venture-capital supply contracts in a region, startups often relocate elsewhere, suggesting that capital availability remains a binding constraint on ecosystem formation.[22]
The COVID-19 pandemic briefly appeared to weaken these geographic constraints by enabling more remote investment. Yet the resulting “death of distance” proved incomplete. Remote deal-making still depends heavily on high-quality digital infrastructure, while physical proximity remains especially important for large transactions and ventures marked by severe information asymmetries.[23]
Computing infrastructure shows similar concentration dynamics. Third-party data centers tend to cluster near urban demand centers, while hyperscale cloud providers often locate facilities in lower-density regions optimized for land, electricity, and cooling costs.[24] At the regional edge layer, peering locations and Internet Exchange Points carry substantial economic consequences. Sparse interconnection points or restrictive peering policies can degrade content quality and increase costs for platform providers.[25]
Specialized human capital further reinforces these dynamics. Broadband investments correlate with faster productivity growth primarily in labor markets with complementary technical skills.[26] In some cases, faster connectivity can even reduce productivity if firms and workers lack the expertise needed to integrate new technologies into existing workflows.[27] Infrastructure investment without complementary ecosystem inputs may therefore generate expectations that outpace economic results.
Jurisdictions that combine communications infrastructure with strong talent pipelines, skilled labor markets, and complementary institutions are far more likely to capture the productive and innovative gains these technologies enable.[28]
The broader pattern is clear: innovation clusters emerge where three conditions align. First, jurisdictions must minimize deployment frictions for communications and computing infrastructure. Second, they must support dense connectivity networks that enable experimentation and collaboration. Third, they must sustain capital markets capable of scaling successful technologies.
Clusters thus operate as bundles of complements, including specialized suppliers, deep labor pools, supporting institutions, and infrastructure networks. Infrastructure alone is rarely sufficient to create a new innovation center. Industry-life-cycle research consistently finds that firms cluster geographically near early industry leaders, from automobiles to semiconductors.[29] Historical examples include the development of Wi-Fi and the U.S. unlicensed-spectrum ecosystem, broadband infrastructure’s role in cloud-computing growth, and divergent deployment environments across the United States, Europe, and China.[30]
B. The Converging Infrastructure Chain
The geographic concentration of technological innovation is becoming more pronounced as next-generation applications demand increasingly sophisticated infrastructure. These applications do not depend on any single network layer. They rely instead on a converging “infrastructure chain”—the layered set of physical and logical systems that support modern digital services.
Historically, communications networks operated in discrete domains: cable networks delivered television, fiber served enterprise connectivity, wireless networks enabled mobility, and satellites provided remote coverage. Those boundaries are rapidly dissolving. Nearly all networks now support Internet Protocol (IP)-based connectivity and cloud-computing services.[31] At the same time, the distinction between communications and computing infrastructure is fading. Data centers, edge nodes, and AI accelerators increasingly function as core infrastructure components, no less essential than the networks connecting them.[32]
This converging infrastructure chain consists of five interdependent layers. The first is the local-access layer, which includes Wi-Fi, 5G, ethernet, and emerging 6G standards. It enables devices to communicate with nearby computing resources. The second is the terrestrial broadband layer, including fiber and cable networks that transport large volumes of data between local access networks and regional or hyperscale data centers.
The third layer consists of backhaul and edge infrastructure, which enables low-latency processing closer to end users. The fourth is satellite connectivity, particularly low-Earth-orbit (LEO) constellations. Satellite systems cut across the infrastructure chain rather than occupying a single layer: the same network can provide consumer broadband in underserved regions while supplying backhaul capacity for terrestrial networks.
Satellite infrastructure nevertheless raises distinct policy concerns. Spectrum allocation, orbital coordination through the International Telecommunication Union, and the economics of globally shared infrastructure create regulatory issues with no direct terrestrial analogue.
The fifth layer is compute itself: data centers, graphics-processing-unit (GPU) clusters, and edge servers that provide the processing power required for AI systems, immersive applications, and cloud services.
The economic stakes are substantial. One recent industry study estimated the value of Wi-Fi in the United States at roughly $1.3 trillion in 2023, with projections reaching $2.4 trillion by 2027. The study attributed much of that growth to enterprise-productivity gains, expanded unlicensed use of the 6 GHz spectrum band, and increasing augmented- and virtual-reality adoption.[33]
Licensed mobile services generate comparable effects. The U.S. wireless industry contributes more than $825 billion annually to gross domestic product, while 5G deployment is projected to generate approximately $1.5 trillion in economic growth and 4.5 million jobs over the course of this decade. Globally, the GSM Association estimates that mid-band 5G alone will contribute $610 billion to global GDP by 2030, accounting for nearly 65% of the total socioeconomic value generated by 5G.[34] Realizing those gains depends heavily on adequate spectrum allocation and timely infrastructure deployment.
Mobile edge computing further raises the geographic stakes by moving computational workloads from centralized data centers closer to end users. That architecture enables latency-sensitive and computation-intensive applications on devices with limited local processing capacity.[35]
The transition toward 6G intensifies these constraints. Proposed 6G standards target sub-millisecond latency and terabit-per-second throughput.[36] At those speeds, physics becomes economically consequential. Because light travels through fiber-optic cables at roughly 200,000 kilometers per second,[37] applications requiring round-trip latencies below one millisecond cannot tolerate compute resources located much farther than roughly 100 kilometers away. Jurisdictions unable to support dense, low-latency edge deployments will therefore face structural disadvantages in hosting the most demanding next-generation applications.
The convergence of wireless infrastructure and AI computing is already underway. NVIDIA’s AI-RAN architecture deploys GPU-accelerated platforms at cell sites capable of simultaneously supporting wireless-network and AI workloads on shared infrastructure. Under this model, base stations evolve from single-purpose radio-access points into multi-purpose edge-computing platforms capable of handling voice, video, data, and AI inference simultaneously.
In October 2025, NVIDIA and a coalition of U.S. firms—including T-Mobile, Cisco, Nokia, and Booz Allen—announced an “all-American” AI-RAN stack built on NVIDIA’s AI Aerial platform. The initiative aims to accelerate the transition toward 6G through AI-driven spectrum management and integrated sensing-and-communications systems relevant to public safety and national-security applications. The U.S. Defense Department (DOD), for example, is exploring integrated sensing-and-communications technologies for drone detection and tracking.[38]
In March 2026, T-Mobile and NVIDIA announced the deployment of “physical AI” applications—including computer vision, video search, and autonomous-system coordination—on AI-RAN-ready infrastructure.[39] That same month, a broader international coalition—including BT Group, Deutsche Telekom, Ericsson, Nokia, SK Telecom, SoftBank, and T-Mobile—committed to building 6G systems on open, secure, AI-native platforms. NVIDIA Chief Executive Officer Jensen Huang described the effort as transforming “the world’s telecom networks into AI infrastructure everywhere.”[40]
Beyond AI-enabled cell sites, firms are also developing “AI Grid” architectures that distribute computing workloads across geographically dispersed regional hubs, central offices, metro nodes, and edge data centers. These systems do not necessarily require direct co-location with wireless infrastructure. NVIDIA’s AI Grid reference design allows telecom providers and distributed cloud operators to route AI-inference workloads dynamically to the location offering the best combination of latency, cost, and available capacity.
Early benchmarks suggest these distributed architectures can reduce cost per token by as much as 76% relative to centralized deployments. Major operators—including AT&T, Comcast, and Indosat—have begun deploying AI-grid systems, leveraging a global network of roughly 100,000 distributed network data centers that may eventually support more than 100 gigawatts of additional AI-computing capacity.
III. The Infrastructure Demands of Next-Generation Computing
Infrastructure’s economic value depends largely on the applications built on top of it. Over the past two decades, smartphones, app-based platforms, and gig-economy services built on fourth-generation (4G) wireless networks and high-speed broadband helped drive U.S. leadership in the digital economy. Those applications turned communications infrastructure into a foundation for new markets, business models, and forms of economic coordination.
The next generation of computing technologies will impose even greater infrastructure demands. AI, AR, virtual reality (VR), and related immersive-computing systems require more than bandwidth. They also depend on ultra-low latency, high-capacity wireless connectivity, edge computing, reliable fiber backhaul, and access to large-scale compute resources.
These technologies differ from earlier internet applications in important ways. Many rely on continuous real-time inference, bidirectional data flows, deterministic network performance, and geographically distributed compute architectures. Their workloads place simultaneous pressure on communications networks, data centers, power systems, and spectrum resources.
As the previous section explained, innovation ecosystems form where complementary infrastructure layers and supporting institutions co-locate. The same dynamic applies here. AI systems, immersive applications, and edge-computing architectures increasingly depend on dense clusters of compute, connectivity, power, and specialized human capital operating as an integrated infrastructure stack.
This section examines two dimensions of that transformation: how AI and hyperscale computing are reshaping the geography and economics of data-center infrastructure, and how AR, VR, and immersive computing are changing the technical requirements of communications networks.
A. Compute Infrastructure and the Geography of AI
The computing demands of the digital economy are compounding rapidly. For the past two decades, centralized cloud services—storage, web hosting, and software-as-a-service (SaaS) platforms—have run on general-purpose processors housed in a relatively small number of hyperscale data centers. That model persists and continues to expand. Large-scale AI, however, has introduced a qualitatively different level of resource demand.
Training frontier AI models requires thousands of specialized accelerators operating in parallel for weeks or months. Inference—the deployment of trained models to generate real-time outputs—creates continuous, geographically distributed demand for computing resources that differs substantially from traditional cloud workloads. AI systems are evolving from isolated cloud-based tools into persistent, multimodal agents operating across phones, vehicles, wearables, and AR and VR platforms. Those applications generate continuous, latency-sensitive, and increasingly upload-heavy traffic flows.[41]
In this sense, data centers have become the factories of the digital economy. Their geographic distribution shapes innovation much as the location of steel mills shaped industrial development a century ago. Specialized hardware requirements—including GPUs, tensor-processing units, and custom AI accelerators—have created supply-chain bottlenecks and concentrated compute capacity among the small number of firms capable of deploying infrastructure at scale.
The location of that compute capacity is not economically neutral. Firms developing AI applications need low-latency access to inference infrastructure and reliable high-bandwidth connectivity to deliver outputs to users. Training workloads are more geographically flexible, tending to cluster in regions with inexpensive land, abundant electricity, and sufficient water for cooling. Inference workloads, by contrast, depend heavily on proximity because they must respond to user requests in real time.
The emergence of frontier AI models outside the traditional U.S. hyperscaler ecosystem—most notably DeepSeek—does not undermine this geographic-concentration thesis. Those models still depend heavily on hardware designed and manufactured through U.S.-led supply chains. China’s aggressive investment in AI infrastructure likewise reinforces the central point: compute geography matters.
Data-center infrastructure exhibits three economic characteristics that shape these geographic dynamics.
First, massive capital requirements create substantial barriers to entry. A single hyperscale data center can cost between $500 million and $1 billion to construct, while large campuses often require multibillion-dollar investments.[42] Those costs create strong economies of scale that favor concentration among relatively few operators. The global data-center industry is now valued at roughly $250 billion, with approximately 500 major facilities concentrated in the United States, Europe, and China.[43]
Second, energy consumption has become a binding growth constraint. Data centers already account for a substantial and growing share of electricity demand, and AI workloads consume significantly more energy than traditional cloud computing. Some projections suggest global data-center electricity consumption could double by 2030 under current growth trajectories.[44] Access to affordable and reliable electricity increasingly determines where new computing infrastructure can be deployed. Energy policy and grid capacity have therefore become core infrastructure-policy concerns.
Third, computing infrastructure is consolidating around hyperscale firms that combine physical infrastructure, software platforms, and AI models into vertically integrated stacks. This integration reflects the economics of frontier AI. Training runs that cost hundreds of millions of dollars naturally favor firms with the balance sheets to finance them. Access to high-performance computing is becoming a critical input to competitiveness: the availability, cost, and proximity of compute resources increasingly shape which firms, industries, and jurisdictions can participate in AI-driven innovation.
Newer techniques such as model distillation and mixture-of-experts architectures—including DeepSeek’s R1 model—show that capable systems can be derived at far lower cost than frontier training runs. But distillation still depends on frontier models from which knowledge can be extracted. These techniques reduce the cost of deploying advanced capabilities; they do not eliminate the concentrated compute requirements necessary to advance the frontier itself. The asymmetry in high-performance compute therefore persists at the layer most relevant to technological leadership.
Data-center economics also reflect a persistent tension between concentration and distribution. Hyperscale cloud providers cluster facilities in locations optimized for land, power, and cooling efficiency.[45] That concentration improves scale economies, but it also creates resilience risks and geographic disparities in access. Regions far from hyperscale clusters often face higher latency, reduced redundancy, and less competitive cloud-service pricing.
Edge computing provides a partial counterweight. By distributing computational capacity closer to end users, edge infrastructure can support latency-sensitive applications that centralized cloud architectures cannot efficiently serve. But edge deployment introduces tradeoffs. Edge nodes sacrifice some efficiency advantages of large-scale data centers in exchange for physical proximity to users and applications. Their viability depends on sufficient local demand and supporting infrastructure, including fiber backhaul and reliable power.
AI infrastructure also depends on the communications networks connecting data centers to users and one another. Fiber backhaul functions as the circulatory system of the AI economy. Training data must move from collection points to training clusters, trained models must be distributed to inference endpoints, and inference traffic must travel between users and compute resources with minimal latency. The scale of this traffic is increasing rapidly as AI-enabled applications spread across sectors.
AI-driven network demand differs from traditional internet traffic in several respects. Training workloads require massive, sustained data transfers capable of saturating even high-capacity links. Inference traffic, while modest per query, creates continuous demand across distributed endpoints. Many emerging AI applications—particularly those involving real-time sensor data, video analysis, and multimodal agents—also produce symmetrical or upload-heavy traffic flows that invert the traditional download-heavy structure of consumer internet use.
Many AI applications also require guaranteed quality of service: predictable latency and bandwidth rather than the best-effort delivery model that historically characterized internet connectivity.[46]
B. Immersive Computing and Network Architecture
AR, VR, and mixed-reality applications place unusually demanding requirements on communications and computing infrastructure, even relative to AI. The primary constraint is the human perceptual system. Visual-vestibular conflicts caused by latency above roughly 20 milliseconds can induce motion sickness, creating a hard engineering threshold that network architecture must satisfy.
Latency is only part of the challenge. Immersive applications also require sustained high-bandwidth throughput for high-resolution stereoscopic rendering, deterministic performance rather than best-effort delivery, and continuous bidirectional data flows as devices constantly upload sensor data and receive rendered outputs.
Legacy communications networks were not designed for these traffic patterns. Consumer internet usage historically followed heavily asymmetrical download-to-upload ratios, often approximating 90/10. Streaming video, web browsing, and file downloads dominated demand, while uploads remained comparatively limited. Cable and digital-subscriber-line (DSL) architectures were optimized accordingly, allocating most network capacity to downstream traffic.
Immersive computing fundamentally alters that model. AR and VR devices continuously transmit sensor data, camera feeds, positional-tracking information, and environmental maps to edge or cloud-computing systems while receiving rendered video frames, spatial audio, and haptic feedback. The resulting traffic patterns are often symmetrical or upload-heavy—precisely the type of demand for which many legacy access technologies are poorly suited.
This is not a marginal engineering issue, but a structural limitation of existing network architecture. Fiber-optic networks, with their inherently symmetrical capacity, and advanced wireless systems such as 5G and Wi-Fi 6E/7 are better suited to support immersive-computing workloads at scale.[47]
Meeting these demands requires coordination across multiple layers of the converging infrastructure chain. 5G wireless ultra-reliable low-latency communications capabilities provide the wireless performance guarantees needed for untethered AR and VR devices. Wi-Fi 6E and Wi-Fi 7, operating in the 6 GHz band, support high-density indoor deployments with the bandwidth and latency characteristics needed for immersive applications in enterprise, education, and health care.
Edge computing provides another critical layer by offloading computationally intensive tasks from wearable devices to nearby infrastructure. Network slicing—the creation of virtualized dedicated networks within shared physical infrastructure—allows providers to guarantee the deterministic performance immersive applications require.
The Federal Communications Commission’s (FCC) 2023 authorization of Very Low Power (VLP) devices in the 6 GHz band may prove especially significant. VLP operations allow untethered AR and VR headsets to achieve sub-millisecond latency over 160 MHz channels without requiring incumbent-protection mechanisms. Economic spillovers associated with AR and VR applications enabled by expanded Wi-Fi capacity are projected to reach $47.8 billion by 2027.[48]
The offload architecture itself is equally important. Next-generation wearable devices face unavoidable constraints involving heat, battery capacity, weight, and device size. Performing all computation locally is often impractical. The emerging solution distributes intensive tasks—including rendering, AI inference, and spatial mapping—to nearby edge-computing resources.
These architectures require the simultaneous availability of high-bandwidth wireless connectivity, low-latency edge compute, and reliable fiber backhaul—in other words, the full converging infrastructure chain described above.
AR and VR technologies function simultaneously as infrastructure drivers and beneficiaries. As applications, they generate demand for network upgrades and edge-computing deployment that benefit the broader digital economy. As platforms, they create new markets in enterprise collaboration, surgical training, industrial design, remote education, and immersive entertainment.
The economic potential is substantial. Realizing it, however, depends on infrastructure readiness. Immersive applications cannot achieve widespread adoption if the underlying networks cannot support them.
IV. Infrastructure Policy and Deployment Frictions
Getting infrastructure policy right today will shape innovation geography for decades. Infrastructure investments are long-lived, their returns emerge slowly, and jurisdictions that attract early deployment often capture cumulative advantages that become hard to dislodge. Private capital does not automatically flow where investment would produce the greatest social returns. It responds to regulatory environments, permitting timelines, and subsidy structures that can accelerate, distort, or foreclose deployment.
Infrastructure deployment is constrained not only by engineering costs and market demand, but also by the regulatory transaction costs between investment decisions and operational infrastructure. Because these projects are capital-intensive and largely irreversible, time matters. Permitting delays impose carrying costs, foreclose alternative investments, and increase the risk that market conditions shift before deployment is complete. In some cases, delay can determine whether a project proceeds at all.
These problems are especially acute for network infrastructure, which often requires sequential approvals from multiple governmental bodies with different authorities and timelines. A single fiber route may require federal environmental review, state highway-right-of-way permits, county or municipal road-crossing approvals, and negotiations with private property owners. Each process applies different standards, follows its own timeline, and is administered by officials with varying technical familiarity with telecommunications infrastructure.
The United States’ layered system of government compounds these coordination problems. Federal lands may involve overlapping jurisdiction from agencies such as the Bureau of Land Management, the U.S. Forest Service, and the DOD, each with separate environmental-review and permitting regimes. State highway rights-of-way are governed by 50 different regulatory systems with varying fee structures, technical requirements, and processing timelines. Local zoning and permitting add thousands of jurisdiction-specific rules. Infrastructure permitted by right in one municipality may require a conditional-use hearing in the next.
These frictions become more consequential as networks evolve toward dense 5G and 6G wireless deployments. Traditional macro-cell networks required relatively few tower permits per jurisdiction. Small-cell architectures require orders of magnitude more attachment points on utility poles, buildings, and street furniture. Cost-and-coverage modeling suggests that reaching roughly 90% population coverage is generally achievable under baseline assumptions, while the final 10% becomes exponentially more expensive and highly sensitive to regulatory design.[49] Pole-attachment disputes alone may account for as much as 25% of rural broadband deployment costs.[50]
Federal environmental review under the National Environmental Policy Act (NEPA) and historic-preservation review under Section 106 of the National Historic Preservation Act can serve legitimate public interests. But they can also impose substantial friction on next-generation infrastructure deployment.[51] The policy challenge is to calibrate review to the actual environmental or historical impact of the activity. Routine deployments—such as attaching equipment to existing structures, installing fiber within existing rights-of-way, or collocating antennas on existing towers—usually present minimal environmental or preservation concerns, yet they often trigger review frameworks designed for major construction projects.[52] Categorical exclusions and streamlined review pathways for routine deployments can reduce timelines without materially undermining environmental or preservation goals.
The FCC’s wireless-siting rules address part of this coordination problem through “shot clock” requirements that obligate local governments to act on siting applications within specified timeframes.[53] These rules create a federal backstop against indefinite local delay. Enforcement, however, remains difficult. Some local governments have circumvented shot-clock requirements by declaring applications incomplete on technical grounds, imposing temporary moratoria, or requiring supplemental studies that restart the clock.[54] Similar timeframe mechanisms for wireline deployment—including fiber permitting, pole attachments, and rights-of-way access—could reduce another major source of friction, though reforms should balance speed against meaningful procedural review.
Regulatory frictions are not the only way public policy shapes deployment. Subsidy programs also play a central role. Subsidies carry familiar economic risks: they can distort investment incentives, invite rent-seeking, and miss their intended beneficiaries. They nevertheless remain a persistent feature of infrastructure policy, both because some market failures may justify public support and because political economy makes subsidy programs hard to avoid. Given that subsidies will continue, program design matters enormously. Poorly designed subsidies can lock in inferior technologies, suppress competition, and create dependencies that outlive their purpose.
A recurring design question is whether subsidy programs should mandate particular technologies or allow providers to select deployment approaches based on local conditions. The case for technology neutrality rests on a straightforward insight: regulators are poorly positioned to predict which technologies will prove most cost-effective across the country’s varied geography and demand conditions. Fiber, fixed wireless, and LEO satellite systems each have comparative advantages in different environments. Mandating one technology risks foreclosing better solutions.[55]
Recent federal programs illustrate these tradeoffs. The Rural Digital Opportunity Fund (RDOF) revealed the risks of inadequate performance verification when some winning bidders proved unable to meet deployment commitments.[56] The Broadband Equity, Access, and Deployment (BEAD) program likewise encountered implementation challenges after initially prioritizing fiber in high-cost regions where projected deployment costs exceeded available funding.
In June 2025, the National Telecommunications and Information Administration (NTIA) issued a restructuring policy notice—described as the “Benefit of the Bargain” reform—that eliminated the prior administration’s fiber preference, removed rate-regulation mandates, and required states to conduct at least one additional competitive subgrantee-selection round open to all technologies meeting statutory performance benchmarks of 100/20 megabits per second and 100 milliseconds latency.[57] The reform required states and territories to reopen competition on a technology-neutral basis, allowing fixed wireless, cable, and LEO satellite providers to compete alongside fiber providers on equal terms. NTIA estimated the restructuring would reduce taxpayer costs by at least $21 billion through greater competition and increased private-sector matching contributions.
The BEAD restructuring illustrates a broader principle: subsidy programs that mandate particular technologies risk inflating costs and foreclosing better deployment options. Performance-based standards—which set minimum service thresholds while allowing providers to choose deployment methods—better fit the geographic diversity of the United States, where mountainous terrain, remote areas, and widely varying population densities make no single technology universally cost-effective.
V. Conclusion
Four conclusions emerge from the preceding analysis.
First, infrastructure policy likely matters more for innovation than direct industrial targeting. Telecommunications infrastructure has a positive causal relationship with economic growth. But its effects are not uniform. Infrastructure investment can produce both concentration and dispersion, depending on the type of connectivity and the local stock of complementary assets, including skilled labor, capital, and institutions. Even technologies that sharply reduce distance frictions can reinforce existing economic centers. Submarine fiber-optic cables, for example, reduced certain spatial frictions by as much as 80%, yet increased London’s share of global trading activity by roughly one-third.[58]
Second, the converging infrastructure chain requires integrated policymaking. Spectrum policy, broadband deployment, data-center siting, edge computing, and satellite connectivity are typically governed by separate agencies under separate regulatory frameworks. Yet each layer depends on the others. Spectrum without backhaul creates stranded capacity; data centers without fiber connectivity become isolated facilities; edge computing without wireless access networks cannot serve users effectively.
Demand for one layer also increases the value of adjacent layers. The relationship between licensed and unlicensed spectrum illustrates the point. Wi-Fi offloads an estimated $33 billion annually in cellular-network capital expenditures, while expanded unlicensed access to the 6 GHz band generated an estimated $870 billion in incremental economic value during its first two years.[59] The growth of advanced wireless services outside the home also increases demand for high-speed residential fiber and in-home wireless networking. Policymaking that treats these technologies in isolation risks creating coordination failures that market participants cannot resolve on their own.
Third, reducing deployment friction across the connectivity and compute stack may generate greater long-run innovation benefits than targeted subsidies for favored technologies. The productivity J-curve associated with general-purpose technologies implies that jurisdictions must sustain investment through extended periods of apparently low returns. Policies that reduce deployment costs—including permitting reform, streamlined environmental review, technology-neutral subsidy design, and energy-policy modernization—may therefore produce larger gains than efforts to select technological winners directly.
Fourth, recent federal actions suggest an emerging bipartisan consensus that next-generation infrastructure requires coordinated federal strategy rather than piecemeal regulatory adjustment. Executive Order 14179 and the subsequent AI Action Plan identified maintaining global AI leadership as an explicit federal policy objective and directed agencies to reduce barriers to AI deployment.[60] The December 2025 executive order establishing a national AI policy framework went further, directing the U.S. Justice Department (DOJ) to challenge state AI laws deemed inconsistent with federal policy and conditioning certain BEAD funds on states’ AI regulatory posture.[61]
Whether one views these measures as necessary coordination or federal overreach, they reflect the jurisdictional-competition dynamics described throughout this brief. The federal government is trying to prevent cross-jurisdictional regulatory fragmentation that economic-geography research associates with slower infrastructure deployment and weaker innovation clustering. The central policy question is not whether coordination is necessary, but how best to design it—balancing regulatory uniformity against state experimentation and the risk that federal preemption may suppress legitimate local objectives.
Public debate over whether the United States risks “losing the race” in technologies such as 5G, AI, and AR often focuses too narrowly on the technologies themselves. Technological leadership depends more fundamentally on the environment in which new technologies can be deployed, scaled, and commercialized.
Next-generation computing architectures depend on a converging infrastructure chain that includes wireless spectrum, terrestrial broadband, edge and hyperscale computing, and satellite systems. The infrastructure challenge no longer concerns connectivity alone. Data centers, AI accelerators, and edge-computing nodes now join fiber networks, towers, and spectrum as core inputs to the innovation economy. Their deployment depends on regulatory environments, subsidy structures, energy policy, and workforce availability—policy domains historically treated separately, but which increasingly function as parts of a single system.
Technological leadership ultimately follows infrastructure. Infrastructure investment is therefore best understood as horizontal industrial policy: it creates broad enabling conditions rather than selecting specific firms or technologies for support.[62] Vertical interventions require policymakers to predict future winners, a task regulators have historically performed poorly.[63] Horizontal infrastructure investment expands the complementarity base for innovation. It lowers entry barriers across sectors, enables applications policymakers cannot anticipate, and allows competitive markets rather than political allocation to determine which technologies succeed.
Jurisdictions that make it easiest to deploy and experiment with next-generation connectivity and compute—and that cultivate the complementary ecosystem of skills, capital, and institutions—will attract the developers, investors, and firms that define the future digital economy.
The technological race is real. Winning it requires treating infrastructure deployment not as a discrete capital expenditure, but as the foundation on which the complementary inputs that generate innovation can emerge and scale. The papers that follow in this series examine these infrastructure domains in greater detail: spectrum policy, the transaction costs of fiber and wireless deployment, satellite connectivity, and policy frameworks for efficient infrastructure expansion.
[1] Aakash Kalyani et al., The Diffusion of New Technologies, 140 Q.J. Econ. 1299 (2025), https://doi.org/10.1093/qje/qjaf002.
[2] Brad Chattergoon & William R. Kerr, Winner Takes All? Tech Clusters, Population Centers, and the Spatial Transformation of U.S. Invention, 51 Research Pol’y 104418 (2021), https://doi.org/10.1016/j.respol.2021.104418. Patenting activity, however, does not necessarily correspond to greater innovation. See, e.g., Andrew W. Torrance, Lisa C. Friedman & Tanya Singh, Is China’s Patent Boom a Bust?, 63 Hous. L. Rev. 453 (2025), https://houstonlawreview.org/article/154481-is-china-s-patent-boom-a-bust. Patent counts are an imperfect proxy for innovation and historically have correlated only weakly—or even inversely—with total factor productivity growth. They nevertheless remain useful for tracking the geographic distribution of formal intellectual-property activity.
[3] See, e.g., Chris Forman, Avi Goldfarb & Shane Greenstein, The Internet and Local Wages: A Puzzle, 102 Am. Econ. Rev. 556 (2012), https://doi.org/10.1257/aer.102.1.556 (finding that advanced internet adoption correlated with wage growth only in the 6% of U.S. counties that were already wealthy, educated, and IT-intensive, suggesting that improved connectivity reinforced rather than dispersed existing geographic concentration).
[4] Adam B. Jaffe, Manuel Trajtenberg & Rebecca Henderson, Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations, 108 Q.J. Econ. 577 (1993), https://doi.org/10.2307/2118401. Knowledge spillovers are not the only force driving agglomeration. Asset complementarities—including shared labor pools, specialized suppliers, common infrastructure, and accumulated industry experience—also exhibit strong geographic concentration. See, e.g., Alfred Marshall, Principles of Economics bk. IV, ch. X (8th ed. 1920) (identifying localized skilled-labor pools, specialized input suppliers, and knowledge spillovers as the three classical sources of agglomeration); Michael E. Porter, The Competitive Advantage of Nations 148–54 (1990) (describing how factor conditions, demand conditions, related industries, and firm rivalry cluster geographically); Maryann P. Feldman & David B. Audretsch, Location, Location, Location: The Geography of Innovation and Knowledge Spillovers, WZB Discussion Paper No. FS IV 96-28 (1996) (discussing how complementary innovation inputs co-locate and reinforce one another). The relevant point here is not that knowledge spillovers alone drive geographic concentration, but that they are the externality most directly tied to the policy question of where innovation emerges and why it persists there.
[5] Describing communications and computing infrastructure as a “general-purpose input” is an economic observation about its role in enabling a broad range of downstream activities, not a legal claim about the proper regulatory framework. General-purpose inputs vary substantially in both market structure and regulation. Electricity and freight rail, for example, are subject to common-carrier or public-utility obligations, while commercial real estate and cloud computing are not. Whether a particular layer of digital infrastructure warrants common-carriage regulation depends on market-specific conditions—including contestability, switching costs, and the availability of substitutes—not merely on the input’s generality.
[6] See, e.g., Yong Mao, Changshuo You, Jun Zhang, Kaibin Huang & Khaled B. Letaief, A Survey on Mobile Edge Computing: The Communication Perspective, 19 IEEE Commc’ns Surveys & Tutorials 2322 (2017), https://doi.org/10.1109/COMST.2017.2745201 (explaining that edge computing’s latency constraints require physical proximity between computing and radio resources).
[7] See Erik Brynjolfsson, Daniel Rock & Chad Syverson, The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, Nat’l Bureau of Econ. Rsch., Working Paper No. 25148 (2018), https://www.nber.org/papers/w25148.
[8] See Daron Acemoglu, David Autor, David Dorn, Gordon H. Hanson & Brendan Price, Return of the Solow Paradox? IT, Productivity, and Employment in U.S. Manufacturing, 104 Am. Econ. Rev. 394 (2014), https://doi.org/10.1257/aer.104.5.394.
[9] See Kristian Stout & Ian Adams, The Hype Cycle Meets Malpractice Law: Why the Jobs Persist, Truth on the Mkt. (Apr. 2, 2026), https://truthonthemarket.com/2026/04/02/the-hype-cycle-meets-malpractice-law-why-the-jobs-persist; Daron Acemoglu & Pascual Restrepo, Automation and New Tasks: How Technology Displaces and Reinstates Labor, 33 J. Econ. Persps. 3 (2019), https://doi.org/10.1257/jep.33.2.3; Brynjolfsson et al., supra note 7.
[10] Paul David & Gavin Wright, General Purpose Technologies and Surges in Productivity: Historical Reflections on the Future of the ICT Revolution, in The Economic Future in Historical Perspective (2006), https://doi.org/10.5871/bacad/9780197263471.003.0005.
[11] Nicholas Crafts, Artificial Intelligence as a General-Purpose Technology: An Historical Perspective, 37 Oxford Rev. Econ. Pol’y 521 (2021), https://doi.org/10.1093/oxrep/grab012.
[12] See Steven Klepper, Industry Life Cycles, 6 Indus. & Corp. Change 145 (1997), https://www.jstor.org/stable/2696374; Susanto Basu & John G. Fernald, Information and Communications Technology as a General-Purpose Technology: Evidence from US Industry Data, 8 Ger. Econ. Rev. 146 (2007), https://doi.org/10.1111/j.1468-0475.2007.00402.x (finding that total factor productivity growth during the post-1995 acceleration correlated with prior-decade ICT investment intensity, suggesting that earlier infrastructure deployment predicted later productivity gains).
[13] Basu & Fernald, supra note 12.
[14] Carol Corrado et al., Innovation and Intangible Investment in Europe, Japan, and the United States, 29 Oxford Rev. Econ. Pol’y 261 (2013), https://doi.org/10.1093/oxrep/grt017.
[15] Avi Goldfarb & Catherine E. Tucker, Digital Economics, 57 J. Econ. Lit. 3 (2019), https://doi.org/10.1257/jel.20171452.
[16] Gilles Duranton & Diego Puga, Micro-Foundations of Urban Agglomeration Economies, in 4 Handbook of Regional and Urban Economics 2063 (J. Vernon Henderson & Jacques-François Thisse eds., 2004), https://doi.org/10.1016/S1574-0080(04)80005-1.
[17] Taiwan’s semiconductor cluster illustrates the point. The tacit process-engineering knowledge concentrated in Hsinchu Science Park—built through decades of real-time collaboration among fab engineers, equipment vendors, and design houses—reflects precisely the sort of learning externality that resists replication through remote coordination, as ongoing U.S. reshoring efforts have demonstrated. See Sujai Shivakumar & Charles Wessner, The Role of Industrial Clusters in Reshoring Semiconductor Manufacturing, Ctr. for Strategic & Int’l Studs. (Oct. 2024), https://www.csis.org/analysis/role-industrial-clusters-reshoring-semiconductor-manufacturing.
[18] See Irene Bertschek, Daniel Cerquera & Gordon J. Klein, More Bits—More Bucks? Measuring the Impact of Broadband Internet on Firm Performance, 25 Info. Econ. & Pol’y 190 (2013), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1852365 (finding that broadband adoption significantly improved firms’ capacity for process and product innovation, but had no statistically significant direct effect on labor productivity); Jed Kolko, Broadband and Local Growth, 71 J. Urb. Econ. 100 (2012), https://www.sciencedirect.com/science/article/abs/pii/S0094119011000490 (finding a positive relationship between broadband expansion and local employment growth, particularly in lower-density areas); Stefanie A. Haller & Seán Lyons, Broadband Adoption and Firm Productivity: Evidence from Irish Manufacturing Firms, 39 Telecomm. Pol’y 1 (2015), https://www.sciencedirect.com/science/article/abs/pii/S0308596114001554 (finding no statistically significant effect of broadband adoption on firms’ overall productivity or productivity growth); Elizabeth A. Mack & Alessandra Faggian, Productivity and Broadband: The Human Factor, 36 Int’l Reg’l Sci. Rev. 392 (2013), https://journals.sagepub.com/doi/abs/10.1177/0160017612471191 (concluding that broadband improved regional productivity only in areas with substantial human capital and highly skilled workforces).
[19] See Bertschek, Cerquera & Klein, supra note 18.
[20] Kalvin Bahia, Pau Castells & Xavier Pedrós, The Impact of Mobile Technology on Economic Growth: Global Insights from 2000–2017 Developments, Int’l Telecomms. Soc’y Conf. Paper (2019), https://www.econstor.eu/bitstream/10419/205164/1/Bahia-et-al.pdf.
[21] Shai Bernstein, Xavier Giroud & Richard R. Townsend, The Impact of Venture Capital Monitoring, 71 J. Fin. 1591 (2016), https://doi.org/10.1111/jofi.12370.
[22] Jun Chen & Michael Ewens, Venture Capital and Startup Agglomeration, 80 J. Fin. 2343 (2025), https://doi.org/10.1111/jofi.13451.
[23] Pengfei Han, Chunrui Liu, Xuan Tian & Kexin Wang, Invest Local or Remote? The Effects of COVID-19 Lockdowns on Venture Capital Investment Around the World, 70 Mgmt. Sci. 3640 (2025), https://doi.org/10.1287/mnsc.2024.04374.
[24] Tommy Pan Fang & Shane Greenstein, Where the Cloud Rests: The Economic Geography of Data Centers, Harv. Bus. Sch. Working Paper No. 21-042 (2025), https://www.hbs.edu/ris/Publication%20Files/21-042_47feacb8-215d-408f-81e2-8b101b9a8f0a.pdf.
[25] Stefan Bauer, William H. Lehr, Georgios Smaragdakis & Volker Stocker, The Growing Complexity of Content Delivery Networks: Challenges and Implications for the Internet Ecosystem, 41 Telecomm. Pol’y 1003 (2017), https://doi.org/10.1016/j.telpol.2017.08.008.
[26] Mack & Faggian, supra note 18.
[27] Id.
[28] Id.
[29] Michael E. Porter, Clusters and the New Economics of Competition, 76 Harv. Bus. Rev. 77 (1998), https://hbr.org/1998/11/clusters-and-the-new-economics-of-competition.
[30] See, e.g., Giuseppe Colangelo, Regulatory Myopia and the Fair Share of Network Costs: Learning from Net Neutrality’s Mistakes, Int’l Ctr. for L. & Econ. (2023), https://laweconcenter.org/resources/regulatory-myopia-and-the-fair-share-of-network-costs-learning-from-net-neutralitys-mistakes.
[31] See, e.g., Eric Fruits, Ben Sperry & Kristian Stout, The Competitive Effects of the Proposed Charter/Cox Transaction, Int’l Ctr. for L. & Econ. (2025), https://laweconcenter.org/resources/the-competitive-effects-of-the-proposed-charter-cox-transaction (describing intermodal competition in broadband connectivity); see also Kristian Stout & Geoffrey A. Manne, Promoting Competition and Innovation in the Evolving Video Sector, Int’l Ctr. for L. & Econ. (2025), https://laweconcenter.org/resources/promoting-competition-and-innovation-in-the-evolving-video-sector (discussing the convergence of video communications networks).
[32] This process has been underway since the introduction of computer equipment in the 20th century. The 1956 consent decree in United States v. W. Elec. Co. barred AT&T from entering any business other than common-carrier communications services, effectively preventing the Bell System from manufacturing or selling computer equipment. That artificial separation between communications and computing persisted for more than two decades, until the 1982 Modified Final Judgment dissolved it. See United States v. W. Elec. Co., No. 17-49 (D.N.J. 1956) (final judgment); see also United States v. Am. Tel. & Tel. Co., 552 F. Supp. 131 (D.D.C. 1982) (Modified Final Judgment).
[33] Raul Katz, Juan Jung, Fernando Callorda & Ramiro Valencia, Assessing the Economic Value of Wi-Fi in the United States, Telecom Advisory Servs. (2024), https://wififorward.org/wp-content/uploads/2024/09/Assessing-the-Economic-Value-of-Wi-Fi.pdf.
[34] GSM Ass’n, Press Release, GSMA Research Finds Mid-Band 5G Could Deliver $610B GDP Growth by 2030 (Oct. 13, 2022), https://www.gsma.com/newsroom/press-release/gsma-research-finds-mid-band-5g-could-deliver-610b-gdp-growth-by-2030.
[35] Yong Mao, Changshuo You, Jun Zhang, Kaibin Huang & Khaled B. Letaief, A Survey on Mobile Edge Computing: The Communication Perspective, 19 IEEE Commc’ns Surveys & Tutorials 2322 (2017), https://doi.org/10.1109/COMST.2017.2745201.
[36] Walid Saad, Mehdi Bennis & Mingzhe Chen, A Vision of 6G Wireless Systems: Applications, Trends, Technologies, and Open Research Problems, 34 IEEE Network 134 (2020), https://doi.org/10.1109/MNET.001.1900287.
[37] Klint Finley, Brits Approach (True) Speed of Light Over Fiber Cable, Wired (Mar. 20, 2013), https://www.wired.com/2013/03/internet-at-the-speed-of-light.
[38] NVIDIA, Press Release, NVIDIA and US Telecom Leaders Unveil the All-American AI-RAN Stack to Accelerate the Path to 6G (Oct. 28, 2025), https://nvidianews.nvidia.com/news/nvidia-us-telecom-ai-ran-6g.
[39] NVIDIA, Press Release, T-Mobile and Partners Integrate Physical AI Applications on AI-RAN-Ready Infrastructure (Mar. 16, 2026), https://nvidianews.nvidia.com/news/t-mobile-nvidia-ai-ran-physical-ai.
[40] NVIDIA, Press Release, NVIDIA and Global Telecom Leaders Commit to Build 6G on Open and Secure AI-Native Platforms (Mar. 1, 2026), https://nvidianews.nvidia.com/news/nvidia-and-global-telecom-leaders-commit-to-build-6g-on-open-and-secure-ai-native-platforms.
[41] Shashank Modi, Simonas Matulionis, Michael Kleeman, Ernesto Wandeler, John Fleury, Jaison Justin, Heinz T. Bernold & Braden Holstege, 6G: The Network for the Future of AI and Immersive Connectivity, Bos. Consulting Grp. & Qualcomm (Feb. 26, 2026), https://www.bcg.com/publications/how-6g-networks-will-shape-the-next-era-of-ai.
[42] Ari Natter & Will Wade, SoftBank Plans Giant Ohio AI Data Center Powered by Gas Plants, Bloomberg (Mar. 20, 2026), https://www.bloomberg.com/news/articles/2026-03-20/softbank-plans-giant-ohio-ai-data-center-powered-by-gas-plants; see also Nicol Turner Lee & Darrell M. West, The Future of Data Centers, Brookings Inst. (Nov. 5, 2025), https://www.brookings.edu/articles/the-future-of-data-centers.
[43] Konstantin Pilz & Lennart Heim, Compute at Scale: A Broad Investigation into the Data Center Industry, arXiv:2311.02651 (Nov. 22, 2023), https://arxiv.org/abs/2311.02651.
[44] Int’l Energy Agency, World Energy Outlook 2025 (2025), https://www.iea.org/reports/world-energy-outlook-2025.
[45] Jingxi Liang, Dong Chen & Shiyu Xu, Energy-Constrained Optimization of Data Center Layouts: An Integer Linear Programming Approach, 18 Energies 5040 (2025), https://doi.org/10.3390/en18185040.
[46] Khaled B. Letaief, Yuanming Shi, Jianmin Lu & Junshan Lu, Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications, 40 IEEE J. on Selected Areas in Commc’ns 5 (2022), https://doi.org/10.1109/JSAC.2021.3126076.
[47] Modi et al., supra note 41.
[48] Raul Katz, Juan Jung, Fernando Callorda & Ramiro Valencia, Assessing the Economic Value of Wi-Fi in the United States 71–78, Telecom Advisory Servs. (Sept. 2024), https://wififorward.org/wp-content/uploads/2024/09/Assessing-the-Economic-Value-of-Wi-Fi.pdf.
[49] Edward Oughton & Zoraida Frias, The Cost, Coverage and Rollout Implications of 5G Infrastructure in Britain, 42 Telecomm. Pol’y 636 (2018), https://doi.org/10.1016/j.telpol.2017.07.009.
[50] Kristian Stout & Spencer Kahn, Comments of ICLE In the Matter of Accelerating Wireline Broadband Deployment by Removing Barriers to Infrastructure Investment, WC Docket No. 17-84, Fed. Commc’ns Comm’n (2022), https://laweconcenter.org/resources/comments-of-the-international-center-for-law-economics-in-the-matter-of-accelerating-wireline-broadband-deployment-by-removing-barriers-to-infrastructure-investment.
[51] See, e.g., Eric Fruits & Kristian Stout, ICLE Comments to FCC on CTIA Petition for Rulemaking, CTIA Petition for Rulemaking, RM-12003, Fed. Commc’ns Comm’n (2025), https://laweconcenter.org/wp-content/uploads/2025/04/2025-CTIA-NEPA-Comments.pdf.
[52] Id.; see also Jeffrey Westling & Kristian Stout, Reply Comments of the International Center for Law & Economics, WT Docket No. 25-276, Fed. Commc’ns Comm’n (2026), https://laweconcenter.org/wp-content/uploads/2026/01/Build-America-Wireless-infrastructure-Comments.pdf.
[53] Westling & Stout, supra note 52.
[54] Accelerating Wireless Broadband Deployment by Removing Barriers to Infrastructure Investment, Notice of Proposed Rulemaking, FCC 25-67, WT Docket No. 25-276 (Sept. 30, 2025).
[55] Kristian Stout, Eric Fruits & Ben Sperry, ICLE Comments on the NTIA’s Proposed BEAD Alternative Broadband Technology Guidance, Nat’l Telecomms. & Info. Admin. (2024), https://laweconcenter.org/resources/icle-comments-on-the-ntias-proposed-bead-alternative-broadband-technology-guidance.
[56] Wireline Competition Bureau, Rural Digital Opportunity Fund Bid Defaults Announced, Public Notice, DA 21-910 (rel. July 26, 2021), https://docs.fcc.gov/public/attachments/DA-21-910A1.pdf.
[57] Nat’l Telecomms. & Info. Admin., BEAD Program Restructuring Policy Notice (June 2025), https://broadbandusa.ntia.doc.gov/news/latest-news/bead-restructuring.
[58] Barry Eichengreen, Romain Lafarguette & Arnaud Mehl, Cables, Sharks and Servers: Technology and the Geography of the Foreign Exchange Market, Nat’l Bureau of Econ. Rsch., Working Paper No. 21884 (2016), https://www.nber.org/papers/w21884.
[59] Katz et al., supra note 48.
[60] Exec. Order No. 14,179, Removing Barriers to American Leadership in Artificial Intelligence, 90 Fed. Reg. 8,741 (Jan. 23, 2025), https://www.federalregister.gov/documents/2025/01/31/2025-02172/removing-barriers-to-american-leadership-in-artificial-intelligence.
[61] Exec. Order, Ensuring a National Policy Framework for Artificial Intelligence, White House (Dec. 11, 2025), https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy.
[62] Dani Rodrik, Industrial Policy for the Twenty-First Century 3–11, John F. Kennedy Sch. of Gov’t, Faculty Rsch. Working Paper No. RWP04-047 (2004), https://drodrik.scholar.harvard.edu/publications/industrial-policy-twenty-first-century.
[63] See, e.g., Scott Wallsten, The Effects of Government-Industry R&D Programs on Private R&D: The Case of the Small Business Innovation Research Program, 31 RAND J. Econ. 82 (2000).