Big Tech’s Electricity Bill Is a Problem. Why the “Pay Your Way” Principle Should Apply to Broadband and Data Centers

The recent White House statement that hyperscale data centers should “pay their own way” for electricity captures a broader, overdue principle: when large, U.S.-based digital platform and cloud companies drive large-scale demand for essential infrastructure, they should bear the costs they create. Today, the rapid expansion of AI-driven data centers is straining electric grids and broadband networks and shifting costs to consumers, communities, and network providers. These pressures are not driven by consumer choice, but by platform business models that monetize artificial intelligence, advertising, and bandwidth-intensive services. Without a credible framework to allocate cost, infrastructure investment becomes politically unstable and economically distorted. The emerging electricity policy debate offers a clear template for broadband and telecom markets, where similar cost-shifting dynamics are already entrenched. This note examines the unfolding U.S. policy response to hyperscale data centers and power generation and why its logic extends to broadband cost recovery for mobile and telecom markets worldwide. Telecom policymakers have documented comparable cost-shifting patterns for almost two decades, as detailed in Strand Consult’s reports and research notes on broadband cost recovery and universal service.

For clarity, this note distinguishes between hyperscale infrastructure providers that operate large data centers and procure power (including Microsoft, Meta, Alphabet, and Amazon through AWS), and platform service providers whose business models generate downstream traffic and monetize AI, advertising, and content (including those same firms and other major digital platforms). In practice, these functions are often housed within the same corporate groups but give rise to distinct cost drivers and policy considerations. Amazon illustrates this dual role: it operates as a hyperscale infrastructure provider through AWS, and as a platform service provider through Prime Video and its advertising ecosystem.

From “Weightless Software” to Heavy Industry

For decades, digital platforms portrayed themselves as “weightless”—pure software firms operating at the edge of networks—while telecommunications and utilities were treated as regulated infrastructure. That fiction has expired. Hyperscale data centers are capital-intensive industrial facilities with large footprints, heavy power requirements, land and water needs, and direct impacts on broadband networks, local grids, communities, and ratepayers.

According to industry tracking, globally there are approximately 1300 hyperscale facilities, generally defined as having load capacity of approximately 50 megawatts or more. A single 50-megawatt data center can consume as much power as a small city of 50,000 homes, every hour of every day. Amazon Web Services (AWS), Google, and Microsoft account for nearly two-thirds of global hyperscale capacity, with Meta operating roughly twenty hyperscale facilities. Different sources track the number of data centers by country with the US estimated for just under half of all such facilities worldwide.

Based on multiple public estimates, the four largest hyperscale operators—Amazon (through AWS), Microsoft, Alphabet (Google), and Meta—likely consume on the order of 100 terawatt-hours (TWh) of electricity per year in the United States. A single 50-megawatt hyperscale data center uses roughly 0.4–0.5 TWh of electricity annually. Recent research by Harvard scholars suggest that U.S. data centers, in aggregate and increasingly driven by AI workloads, accounted for approximately 4 percent of total U.S. electricity consumption in 2024. At prevailing average electricity prices, this level of consumption corresponds to a retail-equivalent cost of roughly $20–25 billion per year.

While hyperscalers often negotiate lower wholesale rates, the scale of demand explains why regulators and communities are increasingly focused on who pays for new power plants, transmission lines, and grid upgrades. This parallels the broadband investment discussion over rights-of-way, building permissions, network build-out, and capacity upgrades — where policymakers increasingly question whether a small number of large users should be allowed to scale without contributing financially to the infrastructure on which they depend.

Digital Transformation and Systemic Infrastructure Demand

Digital transformation has reshaped media, commerce, and communications. Advertising revenues have migrated from traditional outlets to digital platforms, content consumption has shifted to on-demand streaming, and “free” services funded by data extraction have replaced paid models. These changes are well understood. What receives less attention is the infrastructure required to sustain them.

Digital services depend on constant investment in fiber and mobile networks, data centers, and energy generation. Yet many infrastructure providers remain bound by legacy regulatory frameworks that limit pricing flexibility and require service even when demand spikes unpredictably. As digital activity grows, strain on these systems increases, raising a basic question of cost allocation: who should pay when large-scale investments are made primarily to serve a small number of dominant firms?

Telecom operators have lived with this asymmetry for some time, as traffic volume, a primary driver of capacity expansion and cost driven by a small number of platforms, has grown exponentially while revenue recovery mechanisms remained constrained by legacy rules.

Do Hyperscalers Pay Their Own Way?

The section examines the four hyperscalers Microsoft, Google, Meta, and AWS, their power generation strategy, and each firm’s public commitments to “pay its way”. It provides a reference to total share of US internet traffic volume for fixed and mobile networks based on reports by AppLogic Networks.

Microsoft has received notable attention for its stated commitment to “pay its way” on data-center impacts, including support for rate structures in which data centers cover their own power and infrastructure costs. Microsoft accounts for 6% of internet traffic volume on fixed networks; 3% on mobile. Microsoft’s product mix is diversified across subscription software, cloud services, and advertising-supported services, and its consumer bandwidth-heavy offerings like Xbox and gaming downloads/streaming) are significant.

Microsoft likely has the greatest revenue exposure to US government entities, given its significant deployment of both cloud services and enterprise software, and this could drive its incentive to lead publicly on “pay-your-way” expectations.

AWS is the world’s largest standalone cloud provider and probably the most visible target in the data center debate in the US. It is by far the largest energy user of the hyperscalers, possibly thrice as large as Microsoft.

Amazon Inc. accountsfor 5% of internet traffic volume on fixed networks; 3% on mobile. Amazon Prime Video is one of the top video streaming services in the U.S. and contributes significantly to peak broadband usage. Amazon Prime Video has secured rights to a wide range of high-profile live sports, including the NFL’s Thursday Night Football, select NASCAR Cup Series races, NBA and WNBA games, regional MLB broadcasts, major AEW wrestling pay-per-view events, and ONE Championship MMA fights. This is indicative of an arbitrage trend in certain streaming platforms which access premium sports rights without the regulatory, cost, or public-interest obligations historically imposed on broadcasters—while shifting infrastructure costs on to broadband networks. This asymmetry matters not because of content choice, but because live sports drive traffic peaks, imposing non-optional capacity costs on broadband networks.

In contrast to its reluctance to engage in broadband cost-recovery discussions, Amazon has long charged shipping fees and annual Prime membership fees that explicitly help cover the cost of transporting physical goods through logistics networks. Industry estimates suggest Amazon produces a large share of US Postal Service parcel volume and revenue, making it one of the postal service’s most important commercial customers.

Meta builds and operates large data centers and—unlike a pure enterprise software vendor—its revenue model remains overwhelmingly advertising-based. Meta pursues dedicated, long-duration “clean” power arrangements, including reported nuclear-related deals/PPAs aimed at supporting AI-driven load growth. While Meta may sidestep the power generation discussion with its bespoke data centers, it depends heavily on mobile operators’ offering internet access for free or at a deep discount.  

Meta’s internet traffic footprint is also substantial. It accounts for 6% of fixed and 32% of mobile network traffic volume. Meta’s core products—Facebook, Instagram (including Reels), WhatsApp, and Messenger—are among the most data- and AI-intensive applications on the internet, driven by video delivery, real-time messaging, and continuous AI-based ranking, moderation, and advertising systems. Although Meta does not charge end users subscription fees, these services rely on hyperscale data centers, large GPU clusters, high-capacity networks, and substantial power generation to support constant inference, encryption, and content distribution.

Alphabet has one of the most visible renewable-energy procurement strategies among hyperscalers, relying heavily on large-scale Power Purchase Agreements (PPAs) and detailed sustainability reporting. This highlights an important distinction within hyperscaler energy procurement that is often overlooked: not all PPAs are equivalent in their impact on system capacity and cost causation. Some PPAs merely reallocate existing generation or rely on certificate-based accounting, while others are structured to finance genuinely additive capacity that expands supply and reduces system stress. Market-driven developers such as Better Energy illustrate one version of this latter model, developing large-scale generation projects on commercial terms without reliance on feed-in tariffs or operating subsidies.

A different model called Guarantees of Origin (GOs) and similar certificate schemes, critics say, amount to little more than green accounting, failing to add actual renewable capacity to local grids or address the physical energy consumed by data centers and network infrastructure. For example, Vodafone was forced to retract a claim that its service was powered by 100% renewable energy.

Alphabet/Google accounts for significant internet traffic volume: 12% of fixed and 20% of mobile networks. This is especially related to YouTube and its broader ecosystem (Search, Maps, Gmail, Docs/Workspace, etc.) which contribute both traffic and ads. While it appears to moderate on the pay its way discussion for electricity, it is repeatedly unwilling to engage in broadband cost recovery.

Google and other large digital platforms characterize network usage fee proposals as “double billing,” arguing that consumers already pay broadband providers for internet access. While it is true that consumers consent to platform terms and conditions, those agreements place no meaningful upper bound on the volume of advertising, data collection, bandwidth consumption, or electricity use generated by the platforms themselves.

Consumers also pay for electricity, yet they have not consented to absorb the incremental power, infrastructure, and network costs imposed by hyperscale platforms, particularly as AI-driven services dramatically increase traffic and energy consumption. These costs are externalized onto households, rather than transparently attributed to the entities that generate them.

What is notably absent is platform-level transparency: clear disclosure of a service’s operational footprint, including advertising load, measurement and tracking technologies, bandwidth intensity, and electricity consumption. Without such transparency, claims of “double-billing” obscure the real issue—cost causation without cost responsibility.

Recent bipartisan pushback in the US signals that this platform-driven model of passing costs downstream is no longer politically or economically tenable. The policy question is no longer whether these costs exist, but who should bear them.

Another emerging policy issue is the sheer scale of power generation required to support AI inference. On artificial intelligence, Google’s Gemini family of models competes directly with OpenAI’s GPT models, while Microsoft is tightly linked to OpenAI through deep partnership, capital investment, and exclusive cloud distribution. As these firms race to deploy ever more compute-intensive models, the energy demands of AI inquiries—many of them advertising-driven, automated, or marginal in user value—are rising rapidly.

High Electricity Bills Driving Political Breaking Point

High electricity bills are pushing consumers toward a political breaking point after a sustained period of price inflation, with energy costs remaining elevated ahead of the midterm elections. The rapid expansion of hyperscale data centers—driving unprecedented electricity demand—has necessitated new generation and grid investments, the costs of which are increasingly being shifted onto ordinary consumers who did not request this additional power but are nonetheless required to pay for it. This dynamic has sharpened the “cost-causer” expectation that hyperscalers, rather than households, should bear incremental system costs.

Other actors including utilities and grid operators operate under differing incentives and are governed by regulatory frameworks that vary significantly by state, alongside hyperscalers with distinct contractual and market positions. These actors often rely on long-term PPAs that pre-fund generation but do not always cover transmission, congestion, or reliability costs. State energy policies and planning assumptions were not designed for sudden, concentrated hyperscale demand, creating a policy gap that now requires adaptation to prevent consumer backlash and preserve regulatory legitimacy.

On January 16, the White House, together with its National Energy Dominance Council housed at the Department of Energy, hosted a bipartisan signing of a statement of principles with governors from 13 states: Pennsylvania, Maryland, Delaware, Illinois, Indiana, Kentucky, Michigan, New Jersey, North Carolina, Ohio, Tennessee, Virginia, and West Virginia. The statement sets out a shared policy position emphasizing cost causation and consumer protection in the face of rapidly growing electricity demand. The principles affirm that costs associated with new large loads—particularly hyperscale data centers—should be assigned to those loads rather than broadly socialized across residential and small commercial customers. They underscore the responsibility of state utility regulators to use existing rate-setting and oversight authority to prevent disproportionate impacts on households, while preserving system reliability and affordability. The statement also reflects a coordinated state–federal intent to adapt legacy regulatory frameworks that were not designed for sudden, concentrated demand growth within the footprint of PJM, the nation’s largest grid operator.

PJM Interconnection’s board letter from the same day outlines operational and market-oriented actions emerging from its Competitive Infrastructure Funding Process focused on large-load additions. The board directs PJM to improve load forecasting and planning processes to better account for large, discrete demand sources and their timing. It endorses new integration pathways for large loads, including mechanisms that align load interconnection with new generation development and establish clearer curtailment and reliability expectations. The letter supports voluntary frameworks in which large loads or their load-serving entities bring incremental generation capacity to offset demand growth, while also authorizing accelerated backstop procurement to address near-term reliability risks. It further signals a broader review of market incentives and capacity mechanisms to ensure they can attract timely investment under conditions of rapid load expansion.

These developments are notable because the key actors involved are interconnected but not uniformly or directly regulated. The Federal Energy Regulatory Commission (FERC) oversees interstate transmission and wholesale power markets, including regional transmission organizations such as PJM, while day-to-day rate setting and cost recovery for retail customers remain under the authority of state utility regulators. Power generators participate in FERC-regulated wholesale markets, selling electricity and capacity to utilities and other load-serving entities, while wholesalers and marketers intermediate these transactions. Large technology firms, by contrast, typically enter the system as end-use customers or through affiliates, negotiating PPAs, interconnection terms, and reliability arrangements that may fall partly under federal jurisdiction and partly under state oversight. As a result, hyperscalers can significantly influence demand, investment signals, and grid planning without being directly regulated as utilities, creating governance gaps in which costs, risks, and accountability are distributed across multiple regulatory regimes.

The President’s earlier signaling in favor of an emergency auction provides important context for these developments, framing recent actions as part of a broader effort to prevent reliability risks created by rapid, concentrated load growth from overwhelming existing market mechanisms. The White House–led signing of principles with PJM-state governors reflects political concern that traditional planning and cost-allocation frameworks are no longer keeping pace with hyperscale demand, increasing the likelihood that PJM may need to rely on expedited, out-of-cycle capacity procurement to maintain reserve margins. PJM’s board letter issued the same day responds from the system operator’s perspective, acknowledging these near-term reliability pressures and outlining pathways—including backstop procurement—that could culminate in an emergency auction if capacity shortfalls persist.

Why Markets Need Real Price Signals

AI services are delivered through hyperscale data centers, but they rely on downstream broadband and electricity networks that must continuously expand capacity to support them. These facilities influence land use, grid planning, water consumption, and broadband investment decisions at a systemic level. The resulting costs are not incidental; they are induced by platform strategies that concentrate demand and accelerate infrastructure requirements across multiple sectors.

Consumers do not demand ever-increasing volumes of advertising or AI-generated content, yet roughly one-fifth of U.S. internet traffic consists of advertising alone—equivalent to approximately 80 GB of data, or nearly two months of a typical mobile subscription—based on AdGuard’s measurement of ad-related network requests, trackers, and embedded advertising traffic across web and mobile applications.

Users pay for this traffic on equal terms regardless of how much they value it. This outcome is not a market accident but the result of earlier policy choices that allowed digital platforms to externalize infrastructure costs onto networks and end users, a pattern now emerging in electricity markets.

Unlike electricity, where rising costs are passed through quickly to consumers, broadband providers have largely absorbed growing infrastructure costs internally—masking cost pressures while increasing long-term investment risk. While energy prices have remained elevated, U.S. broadband prices have fallen 43% since 2015, even as traffic volumes have grown. This reflects sustained competition between fixed and wireless technologies, as well as large-scale investment and network upgrades, in part facilitated by consolidation that allows providers to operate more efficiently at scale. Maintaining affordability and efficiency in the face of hyperscaling is increasingly difficult without policy reform that addresses cost allocation directly.

Capitalism functions best when prices reflect real costs. When firms face the full cost of their inputs, investment decisions improve and resources are allocated more efficiently. In both electricity and broadband markets, however, these price signals have become muted.

Hyperscalers invest in private assets—data centers, PPAs, private fiber—but they do not consistently internalize system-wide costs such as grid reinforcement, reserve capacity, or last-mile broadband upgrades. Those costs are in many cases, shifted to ratepayers, taxpayers, and local communities, creating an unintended subsidy that concentrates benefits while dispersing financial and political risk.

Exporting American Tech Requires Domestic Discipline

More than elections are at stake. Last year the US Administration issued executive orders to accelerate AI and associated data center permitting, designed to strengthen US dominance in the AI field. The US has a legitimate strategic interest in exporting its AI technology stack—American platforms, standards, chips, cloud services, and tools are preferable to China’s on both economic and security grounds. But this stack is not abstract or weightless software. It is physically instantiated in hyperscale data centers, advanced semiconductors, power generation and transmission, and high-capacity broadband networks.

Current policy frameworks frequently assign the infrastructure costs required to support this stack to utilities, telecom operators, and local communities, allowing platform providers to optimize globally and avoiding many domestic costs of scale. This dynamic echoes the earlier offshoring wave, in which manufacturing value accrued to shareholders while adjustment costs were borne by workers and communities. A similar pattern is now emerging as AI is treated as intangible despite its heavy dependence on energy, broadband, and capital-intensive infrastructure.

American competitiveness does not require shielding domestic champions from real costs. On the contrary, firms that can export AI services while fully paying for the power and broadband networks they rely on are more resilient, more innovative, and better positioned to compete globally over the long term.

The “pay your way” principle applies with even greater force in the Caribbean.

These distortions are magnified in small and developing markets, where infrastructure costs cannot be spread across large populations—a point illustrated starkly in the Caribbean case, which Strand Consult has observed for years. The large digital platforms generate an estimated $12 billion in annual revenue from Caribbean users and advertisers yet maintain little to no physical presence in the region. They employ few, if any, local workers; operate virtually no data centers; contribute minimally to new submarine cable investment; and generally pay neither corporate tax nor sector-specific regulatory fees in Caribbean. Despite this, their services drive significant demand for electricity, broadband capacity, and international connectivity, costs borne by connectivity providers which are expected to deliver on social obligations to consumers, tourism-dependent businesses and governments. In small island systems, where infrastructure shocks cannot be diluted across large populations, this imbalance is particularly acute and underscores why aligning cost responsibility with value extraction is essential for economic sustainability. See Strand Consult’s report Gigabit Caribbean: Closing the Investment Gap in Fixed and Mobile Networks.

These cost-causation challenges are most visible in the Universal Service Fund.

As Strand Consult has previously documented, the Universal Service Fund (USF) was designed for a communications market in which costs were recovered primarily from traditional voice providers and traffic patterns were modest and predictable. Today, broadband networks carry exponentially greater volumes of data driven by hyperscale platforms, cloud services, streaming, advertising, and AI workloads, yet USF contributions remain narrowly assessed on legacy services while the largest sources of traffic and value creation contribute little or nothing. The program supports broadband and communications connectivity for approximately 135 million Americans—connections that directly enable more than $200 billion in annual revenue for the largest technology platforms, even as those platforms make no direct contribution to sustaining the USF.

This mismatch mirrors the emerging problem in energy markets: infrastructure costs are rising in response to hyperscale demand, while recovery mechanisms lag behind reality. Modernizing the USF to reflect current usage and cost causation would restore neutrality by aligning contributions with network dependence, ensuring that the platforms most reliant on national connectivity help sustain the systems they increasingly dominate. Bipartisan, bicameral legislation such as the Lowering Broadband Costs for Consumers Act (S. 1651 / H.R. 4032) would take an important step in that direction by updating the USF framework, and there is an expectation for the USF Working Group to translate its findings into legislative action.

While broadband prices account for a relatively small share of household budgets, overall affordability pressures remain salient ahead of the November midterm elections. As with electricity, broadband is an essential service—and a politically ripe area for action.

Pay Your Way Is Market Capitalism

For telecom operators, regulators, and consumers alike, the electricity debate reinforces a lesson long evident in broadband: when cost causation is ignored, trust erodes, investment falters, and political intervention becomes inevitable. Requiring data centers to pay for the energy and network capacity they consume is not radical—it is market discipline. When firms bear the full costs they create, prices send accurate signals, investment improves, and innovation is rewarded rather than subsidized. This requires neither special taxes nor punitive regulation, only correct pricing at the front end. When hyperscalers do not pay, consumers do, and that distortion weakens markets and public trust.

To support informed policy development, Strand Consult launched its Global Project for Business Models for Broadband Cost Recovery, providing evidence-based analysis, policy research, and transparency tools to help operators and policymakers assess the problem and evaluate solutions at both local, national, and global levels.