Power, Compute and National Power

National power has never rested upon resources alone. Coal beneath the ground did not make Britain an industrial power until mining, finance, engineering, transportation and machinery converted that resource into productive capacity; petroleum acquired geopolitical significance because refining, pipelines, engines, logistics and industrial systems transformed crude oil into mobility and military reach; uranium became strategically consequential only after scientific institutions, enrichment infrastructure, reactor engineering and weapons programs learned how to unlock the energy contained within it. Resources become power when a society develops the ability to convert them into usable capability.

Artificial intelligence introduces another conversion mechanism into that long history. Electricity can now operate computational systems that perform increasingly sophisticated forms of analysis, prediction, optimization, simulation, generation and automated reasoning, allowing energy and industrial capital to produce something that previously depended almost entirely upon human cognition. The result does not make intelligence reducible to electricity, nor does it make computational capacity synonymous with national strength, but it does create a new pathway through which physical resources can be transformed into economic, scientific, institutional and military capability.

The chain developed throughout this series can therefore be followed from power through compute and intelligence into productivity, capability and ultimately national power. That progression does not imply technological determinism. A nation can possess enormous energy resources and govern them poorly, accumulate computers without using them productively, or build advanced models without possessing institutions capable of incorporating them into economic and strategic systems. National strength still depends upon human judgment, political institutions, entrepreneurship, capital allocation, scientific culture, education, military organization and countless other factors that cannot be compressed into a technological equation.

What artificial intelligence changes is the amount of usable cognitive capacity that can increasingly be produced through industrial systems. Once that becomes possible at scale, energy, compute and intelligence enter the traditional calculus of national capability in a more direct way than they did during earlier periods of the digital economy.

National Power Is Ultimately a Conversion Problem

National power is often discussed through inventories: population, gross domestic product, military spending, energy reserves, industrial output, territory, natural resources and technological assets. Those measurements are useful, but they describe stocks of resources more readily than they describe the ability to convert those resources into outcomes. A large population may contribute less than its potential when education and organization are inadequate, just as extensive mineral reserves may remain uneconomic without transportation networks, capital and refining capacity. It may spend heavily on defense while struggling to manufacture munitions at scale, or possess advanced laboratories whose discoveries rarely become commercial products.

The practical value of these assets depends upon the processes that turn them into usable capability. Scientific knowledge must become technology, technology must become production, production must become economic value, and economic value must be translated into the institutions, infrastructure and strategic systems through which a nation expands its range of choices. Artificial intelligence increasingly enters that sequence because machine cognition can enhance several of the conversion processes simultaneously. AI can assist scientific research, improve engineering, automate portions of manufacturing, optimize logistics, accelerate software development, analyze intelligence, support administrative systems and improve the coordination of increasingly complex organizations.

The strategic value of an advanced AI model depends upon whether its machine intelligence can be translated into broader national capability. That distinction is important because nations will not necessarily benefit from AI in proportion to the number of data centers they construct or benchmark records their companies achieve. The greater advantage will belong to systems capable of incorporating computational intelligence throughout productive institutions without allowing the physical requirements underneath that intelligence to become binding constraints.

Intelligence Is Becoming Partly Manufacturable

Human intelligence remains the origin of the AI system. Scientists develop the theories, engineers design the hardware, businesses determine applications, institutions establish objectives, and people ultimately decide what purposes technology should serve. Machine cognition should therefore be understood as an extension of human productive capability rather than an independent substitute for the civilization that created it.

Machine cognition changes the scale at which certain forms of analytical work can be reproduced, even though the development of human expertise remains essential. Producing a highly skilled physician, engineer, scientist, intelligence analyst or military planner requires years or decades of human development. A society can expand the number of such specialists through education, immigration and institutional investment, but the process remains constrained by biology, time and individual specialization.

Computational intelligence introduces another mechanism. Once the models, semiconductors and supporting systems exist, additional machine cognition can be produced by adding computational capacity, improving efficiency and supplying the electricity required to keep those machines operating. This means a portion of economically useful intelligence is becoming reproducible through capital formation. The implications become clearer when considered at national scale. A society capable of adding gigawatts of computational capacity is not merely increasing the number of servers inside its borders; it is increasing the quantity of machine analysis, inference, simulation and automated decision support that can potentially be applied across its economy.

The International Energy Agency projects global data-center electricity consumption rising from approximately 485 TWh in 2025 to roughly 950 TWh by 2030, with electricity consumption at AI-focused facilities growing substantially faster and roughly tripling over the same period. The United States and China together account for nearly 80 percent of the projected increase in global data-center electricity consumption through 2030, demonstrating how heavily the physical expansion of machine intelligence is concentrating within large national systems capable of financing substantial infrastructure. The significance of those facilities lies in their ability to convert electrical and computational inputs into cognition that can be applied throughout other forms of productive activity.

Economic Power Begins With Productivity

The most immediate connection between machine intelligence and national power runs through productivity. Economic strength ultimately depends upon how effectively labor, capital, resources and knowledge can be combined to produce valuable goods and services. Improvements in productivity allow a society to generate more output without requiring proportional increases in every input, which is why technological advances have historically had consequences far beyond the industries that invented them.

Electricity provides a useful analogy. The largest economic benefit of electrification was not the creation of an electricity industry by itself, but the reorganization of factories, offices, transportation, communications and households around electrical power. Artificial intelligence may produce a similar pattern if machine cognition becomes embedded throughout economic activity. The principal gains would then come not simply from companies selling AI products but from manufacturers using predictive systems, hospitals reorganizing administrative and clinical workflows, logistics networks optimizing continuously, engineers iterating designs more rapidly, financial institutions improving analysis and software companies producing more with smaller teams.

The scale of current investment suggests that businesses are already treating computational capacity as a significant productive asset. Global data-center investment approached half a trillion dollars in 2024 according to the IEA, while the agency's updated outlook describes infrastructure financing requirements that are becoming too large to remain confined to the balance sheets of hyperscale technology companies alone. If those investments generate sufficient returns, machine intelligence becomes another mechanism through which capital can increase the productivity of an economy. Countries that can supply the required compute cheaply and integrate it effectively may therefore gain not simply a technology industry but a productivity multiplier capable of reaching many other sectors.

This is where electricity returns to the national-power equation. The United States is already experiencing renewed electricity-demand growth after years of comparatively modest increases. EIA's September 2026 outlook projects electricity generation reaching a record 4,368 billion kWh in 2026 and rising again in 2027, with data centers and increased manufacturing activity among the principal sources of additional demand. When productive intelligence requires electrical supply, electricity capacity becomes indirectly connected to the amount of cognitive capital the economy can deploy.

Scientific Power Can Compound Faster

Science provides another pathway through which compute can become national capability because modern research is increasingly computational. Physics, chemistry, materials science, genomics, engineering, climate modeling, drug discovery and numerous other disciplines already depend heavily upon simulation and data analysis. Artificial intelligence can expand that process by helping researchers explore larger hypothesis spaces, interpret experimental data, design candidate materials and automate portions of scientific workflows. The significance lies not simply in performing existing research tasks more cheaply, but in increasing the rate at which knowledge itself can accumulate.

The U.S. Department of Energy's Genesis Mission illustrates how national scientific institutions are beginning to organize around that possibility. DOE is constructing an integrated AI-for-science platform connecting supercomputers, experimental facilities, scientific datasets and AI systems across its national-laboratory network, while projects announced during 2026 span advanced manufacturing, biotechnology, critical materials, nuclear energy, quantum science and other domains.

The program does not prove that AI will deliver every scientific breakthrough anticipated by its sponsors, and public-sector research initiatives should ultimately be evaluated by results rather than ambition. Its strategic relevance lies in the institutional assumption underneath it: computational intelligence is now considered sufficiently important to scientific productivity that governments are deliberately integrating AI with national research infrastructure.

Greater compute can potentially accelerate the discovery of materials, energy technologies, semiconductors and manufacturing methods that in turn expand computational capacity, creating a reinforcing relationship between scientific progress and the resources available for further research. This relationship also operates through the energy system discussed earlier in the series. Energy supports computation that produces machine intelligence, whose application can improve technologies used to produce and consume energy. At national scale, that cycle can become one mechanism through which scientific advantage compounds rather than remaining static.

Military Power Increasingly Depends Upon Cognitive Systems

The relationship becomes more direct in defense, where information, speed, logistics and decision-making have always influenced military effectiveness. Modern militaries operate enormous sensing networks that generate satellite imagery, radar data, communications intelligence, acoustic information, logistics data and operational reporting. The challenge is not simply collecting information but transforming it into useful decisions before the information loses relevance. AI is increasingly being applied to precisely that problem.

The U.S. military's 2026 AI strategy explicitly identifies warfighting, intelligence and enterprise operations as major areas of deployment, while programs announced during the year include AI-enabled battle-management and targeting tools intended to process operational information and present commanders with options more quickly. The department has also expanded secure access to frontier AI systems across both unclassified and classified environments.

These developments should not be interpreted as evidence that machine systems replace military judgment or that greater compute automatically produces superior military outcomes. Warfare remains an intensely human activity shaped by doctrine, training, morale, leadership, logistics, geography and political objectives. What changes is the computational layer supporting those activities. AI can potentially accelerate intelligence analysis, logistics planning, simulation, cyber operations, autonomous systems, maintenance and command support. The U.S. military's own current deployment efforts identify precisely these functions, demonstrating that defense institutions increasingly treat machine cognition as an operational capability rather than merely an experimental technology.

Once military systems depend upon machine intelligence, the physical resources beneath AI acquire a national-security dimension. Compute availability, semiconductors, secure networks and reliable electricity become parts of the chain connecting industrial capacity to military capability. The power plant and the processor remain far removed from the battlefield, but modern warfare increasingly connects them.

Energy Security Is Becoming Cognitive Security

Energy security traditionally concerns the ability of a nation to maintain adequate supplies of fuel and electricity despite price shocks, geopolitical disruption or physical failure. The expansion of AI adds another reason energy security matters because electricity increasingly supports systems performing economically and strategically important cognitive work. A prolonged energy disruption affecting large computational clusters could eventually impair more than consumer internet services. It could reduce access to systems supporting logistics, financial operations, scientific research, government administration, industrial control, cybersecurity or military analysis.

The relationship will not be perfectly direct because computing workloads can move geographically, backup generation can protect critical facilities and cloud providers can distribute services across multiple regions. Those redundancies are precisely why resilience remains possible. Nevertheless, the more cognitive activity becomes dependent upon computational infrastructure, the more the resilience of that infrastructure becomes part of economic and institutional resilience.

Energy, compute and cognitive security consequently become connected through the systems that keep machine intelligence available. Maintaining electricity supplies helps preserve computational services, which increasingly support cognitive work across important institutions. A country capable of maintaining abundant and resilient electrical systems preserves the physical foundations of both conventional industrial production and increasingly important forms of machine cognition.

China's continuing expansion of its electrical system demonstrates the scale at which this relationship can operate. The country's National Energy Administration reported total installed generating capacity of approximately 3.89 TW at the end of 2025, 16.1 percent above the previous year, while official statistics show substantial continued expansion in solar and wind generation alongside other parts of the power system. Those capacity figures cannot by themselves measure China's AI capability because installed generation includes resources with very different utilization profiles and serves an economy far larger than its data-center sector. Their relevance lies in demonstrating the scale of the electrical platform upon which computational expansion must ultimately operate.

Strategic Autonomy Is About Options, Not Autarky

The relationship between compute and national power must be understood alongside the benefits of international specialization, which allow countries to combine complementary strengths. The modern semiconductor and computing economy has achieved extraordinary performance precisely because production is internationally specialized. American firms lead important areas of chip design and cloud computing; Taiwan plays an enormous role in advanced semiconductor fabrication; South Korea is central to memory; Japan, Europe and the United States possess important semiconductor-equipment and materials capabilities; numerous additional countries contribute manufacturing, resources and specialized components. Attempting to reproduce every stage of those supply chains within every national economy would destroy many of the efficiencies that created their extraordinary technological sophistication.

Strategic autonomy is better understood as preserving options when conditions deteriorate. Countries can benefit from semiconductor trade while developing alternative suppliers, obtain minerals internationally while monitoring concentrated refining capacity, and import transformers while accounting for the possibility that multi-year equipment shortages will constrain grid expansion.

The same principle applies to compute. A nation can purchase computational services internationally while still recognizing that excessive dependence upon foreign-controlled infrastructure may become consequential when workloads involve defense, critical infrastructure, scientific research, sensitive government data or strategically important industries. The objective is therefore not computational isolation but resilience: enough domestic capability, allied capacity and diversified supply that external disruption does not eliminate access to essential systems.

This distinction matters because a productive intelligence economy will almost certainly remain global. The strategic question concerns how that interdependence is structured.

Sovereign Compute Will Become a Larger Policy Question

As machine intelligence becomes embedded in critical institutions, governments will increasingly confront decisions about which workloads can safely depend upon infrastructure located abroad. Some data already falls under residency requirements because of privacy, financial regulation or national-security considerations. AI introduces additional questions because models themselves can contain sensitive information, computational workloads may involve proprietary or classified data, and access to frontier systems can be limited through commercial or governmental controls.

This does not mean every government should construct a nationally isolated AI ecosystem. Smaller economies in particular can benefit enormously from accessing globally distributed cloud infrastructure rather than attempting to reproduce hyperscale systems independently. Yet the political significance of compute changes when access becomes a condition for performing important state functions. A country dependent upon foreign providers for ordinary office software is in a different strategic position from one dependent upon them for defense intelligence, national scientific computation, secure communications or the operation of critical industrial systems.

China's national infrastructure programs demonstrate one approach to treating compute itself as a coordinated strategic resource. Official data from the National Data Administration reported approximately 2.45 million PFLOPS of intelligent-computing capacity by the end of July 2026, with more than 85 percent of that capacity associated with eight national computing hubs and three coordinated compute-electricity development areas. The same agency describes its national computing network as a major infrastructure project intended to improve coordination between computing resources and electrical supply.

The policy model is distinctive to China's institutional system and is not directly transferable to market economies with different governance structures. What it demonstrates is that national governments increasingly recognize compute capacity itself as a strategic infrastructure category.

Capital Markets Become Part of the Strategic System

An intelligence economy also requires extraordinary quantities of capital. Data centers, semiconductor fabs, power plants, transmission systems and advanced manufacturing facilities demand investments measured in billions or tens of billions of dollars. Some projects may fail, some technologies will prove uneconomic and periods of overinvestment are entirely possible, which makes disciplined capital allocation essential.

The ability of a financial system to fund large, risky technological investments is nevertheless a national capability in its own right. Deep capital markets allow private actors to experiment at scales that government budgets alone would struggle to reproduce. Venture capital can fund speculative technologies, public equity can finance rapidly growing companies, debt markets can support infrastructure and institutional investors can provide long-duration capital to assets whose returns develop over decades.

The AI buildout is already testing those mechanisms. The IEA notes that data-center investment has become sufficiently large that continued expansion will depend increasingly upon external capital markets rather than technology-company balance sheets alone.

National power therefore emerges not simply from possessing resources but from possessing institutions capable of mobilizing them. Energy resources must find financing, semiconductor designs must reach fabrication, infrastructure projects must attract long-duration capital and scientific discoveries must find pathways into commercial production. Financial depth is part of the conversion mechanism.

Institutions Determine Whether Capacity Becomes Capability

This is also why national comparisons based purely upon physical scale can mislead. Electricity generation, compute capacity, semiconductor production and industrial output provide valuable information, but countries differ considerably in how effectively institutions translate those assets into useful economic and strategic outcomes.

Markets coordinate prices and capital allocation through decentralized decisions, while governments establish regulations, fund research and maintain national-security institutions. Universities educate the scientists and engineers working across corporations that organize production, utilities that coordinate electrical systems and financial institutions that assess which projects should receive capital. The performance of the overall system depends upon how these institutions interact.

Artificial intelligence can amplify strong institutions because capable organizations can use machine cognition to improve decisions and processes. It can also amplify dysfunction when organizations automate poorly designed procedures or deploy powerful systems without the information, incentives or expertise required to use them productively.

The relevant national capability therefore includes absorptive capacity: the ability of firms, governments, laboratories and other institutions to incorporate new technologies into actual operations. A country could build enormous amounts of compute while obtaining disappointing returns if businesses cannot reorganize around it. Another could possess fewer frontier-scale data centers yet extract substantial economic value through widespread adoption across existing industries. Electricity consumption and GPU counts measure inputs whose contribution to national power depends upon a more complex process of institutional adoption and productive use.

The United States and China Represent Different Configurations of Capability

The United States and China are particularly important to this discussion because they combine enormous economies with substantial technological, industrial and energy systems, while approaching the intelligence economy from different institutional and productive structures. The United States possesses leading frontier AI companies, major cloud platforms, globally important semiconductor designers, deep capital markets, extensive energy resources and a research system that includes universities, national laboratories and private industry. Its electricity demand is now accelerating as data centers and manufacturing expansion place new pressure upon a grid that spent much of the previous fifteen years operating in an environment of relatively modest load growth. The IEA expects data centers to account for roughly half of U.S. electricity-demand growth through 2030.

China combines a vast manufacturing base with an electrical system undergoing extraordinarily rapid expansion and a state-directed effort to increase domestic semiconductor and computing capacity. Its total electricity demand is expected by the IEA to grow by an amount through 2030 comparable to the European Union's current annual electricity consumption, while official Chinese data show simultaneous expansion in both power generation and intelligent-computing capacity.

The semiconductor production system complicates any simple bilateral comparison because the relevant supply chain extends far beyond either country. SEMI expects China, Taiwan and Korea to remain the largest destinations for semiconductor-equipment spending through 2028, while substantial investment continues in the Americas and other regions as AI demand and supply-chain regionalization drive capacity expansion.

The strategic issue is therefore not reducible to determining which country possesses the largest single input. The more revealing question concerns how each system combines research, energy, industry, capital, compute and institutions into sustained productive capability. That comparison will remain dynamic because strengths at one layer can coexist with dependencies at another.

National Power Becomes More Computational Without Becoming Less Human

There is a temptation when discussing AI and geopolitics to imagine a future in which national power becomes a simple function of computational scale, as though the country with the most processors inevitably possesses the greatest strategic strength. History offers little reason to expect national power to become that mechanically predictable.

Technology changes the instruments through which nations exercise power, but people still determine objectives, institutions still coordinate action and societies still decide whether resources are used competently. Military history contains numerous examples of technically sophisticated forces failing because strategy, leadership or logistics proved inadequate, while economic history contains countries rich in resources that remained poor because institutions prevented those resources from being used productively. Artificial intelligence operates within these human and institutional conditions, adding computational capabilities whose value depends upon how they are applied.

The most significant effect may therefore come from AI increasing the leverage of already capable systems. An effective scientific institution supplied with powerful machine cognition may discover more; a productive manufacturer can automate further; a competent military can process information faster; an efficient logistics system can optimize at greater scale. The same technology introduced into weak institutions may yield much less.

Machine intelligence therefore expands the range of possible capability without guaranteeing the quality of the decisions made with it. National power remains human in purpose even as portions of its cognitive machinery become computational.

Intelligence Dependency Could Become a New Form of Strategic Dependency

The inverse side of this argument deserves equal attention. If machine cognition becomes an important input into productive and governmental systems, countries unable to produce or reliably access sufficient compute may become increasingly dependent upon those that can. That dependency need not initially appear coercive. It may simply look like ordinary cloud consumption: companies and governments purchase access to sophisticated systems because foreign providers can deliver them more cheaply and effectively. There is nothing inherently problematic about such trade. International specialization often makes economies richer precisely because every country does not duplicate every technology.

The strategic concern emerges when substitution becomes difficult. The ability of a business to migrate from a foreign accounting platform offers a different degree of flexibility from that available to a state whose scientific laboratories, defense systems, industrial planning and critical infrastructure rely upon computational services unavailable elsewhere.

The distinction resembles other strategic dependencies. Energy-importing nations participate normally in international trade while still maintaining reserves and diversified suppliers because sudden interruption could become economically destabilizing. Semiconductor supply chains are similarly global, yet governments increasingly consider concentration risks because alternative fabrication capacity cannot be created quickly during a crisis. Compute may gradually acquire comparable characteristics. The more important machine intelligence becomes to economic and governmental operations, the more valuable assured access becomes.

The New Strategic Reserve May Include Compute

This raises a question that has received comparatively little attention: whether computational capacity eventually begins to function like other strategic reserves. Governments maintain petroleum reserves to sustain transportation and industry during energy disruptions, while militaries stockpile munitions because factories cannot instantly meet wartime consumption. Telecommunications networks address a related need through redundant capacity that preserves communication when a route fails.

Compute has traditionally been treated primarily as a commercial service whose capacity expands in response to demand. That assumption may become less sufficient if machine intelligence penetrates critical systems. A national emergency involving cyber conflict, military mobilization, a major scientific challenge or disruption of foreign cloud access could sharply increase demand for secure computation at the same moment private-sector demand remains high. Maintaining some amount of available computational headroom could therefore become strategically analogous to maintaining spare industrial or electrical capacity.

This does not necessarily require a literal government-owned warehouse of idle GPUs. Strategic compute could take many forms: reserved cloud capacity, distributed national-laboratory systems, defense-specific infrastructure, agreements with domestic providers or capacity that can be redirected under predefined conditions. The rationale follows the broader relationship between available capacity and resilience. Resilience frequently requires maintaining capacity beyond what appears maximally efficient during normal conditions. The intelligence economy may prove no exception.

Energy Abundance Becomes Strategic Optionality

The same reasoning returns us to energy. A country with an electricity system designed only around existing demand possesses fewer options when a new industry suddenly requires large amounts of power. A country with abundant generation, strong transmission and the ability to build additional capacity can accommodate opportunities whose exact form was impossible to predict when the infrastructure was originally constructed.

This is one reason energy abundance has consequences beyond lower electricity prices. That abundance leaves room to pursue new opportunities when circumstances change. Additional power can support new factories, population growth, semiconductor plants, transportation systems, military facilities, scientific laboratories or computational infrastructure according to whatever future opportunities prove most valuable. Artificial intelligence increases the value of that optionality because energy can now support an input—machine cognition—that is applicable across an unusually broad range of economic and strategic functions.

The relationship does not justify indiscriminate construction or imply that every proposed power project is economically sound. Infrastructure consumes capital that could be used elsewhere, while generation technologies differ substantially in reliability, environmental impact, development time and cost. The important principle is that an electrical system with room to grow gives an economy more choices than one operating continuously against scarcity. National power has always been partly about the number of choices available when circumstances change.

Compute Abundance Can Create Similar Optionality

The same logic applies one layer higher. When compute is scarce and expensive, organizations reserve it for only the applications offering the largest expected returns. As computational costs fall and capacity expands, more marginal applications become economically viable. More affordable computation allows hospitals to analyze more medical images and manufacturers to inspect more products, while scientists can run additional simulations, defense organizations can process more sensor data and financial institutions can examine a wider range of scenarios.

This expansion resembles previous technological transitions in which falling costs changed not only the price of existing activities but the number of activities society found worthwhile performing. If machine cognition becomes sufficiently inexpensive, organizations will not simply perform today's amount of analysis more cheaply; they will likely undertake forms of analysis that were previously impossible to justify. That is why national compute capacity can eventually matter beyond the technology industry itself.

Abundant compute creates more room for experimentation across these activities. The resulting opportunities to test ideas can increase the likelihood of discovering valuable applications. The resulting innovations can then increase demand for additional compute, producing another feedback loop between capacity and capability.

The Intelligence Economy Changes the Meaning of Infrastructure

At the beginning of the digital era, infrastructure policy and technology policy could still appear to occupy different worlds. Roads, ports, electrical grids and power plants belonged to the physical economy, while software and internet services belonged to a relatively weightless digital economy whose most important inputs seemed to be talent and capital.

Artificial intelligence breaks down that distinction. Transmission lines and nuclear reactors can supply computational systems performing scientific research, while a missing transformer can delay the deployment of billions of dollars of AI equipment. Semiconductor fabs acquire strategic relevance through the defense systems they support, and fiber routes help determine where inference services can operate economically. The infrastructure beneath intelligence therefore becomes part of technological capability itself.

This changes how national investments should be understood. Maintaining grid resilience or semiconductor supply is no longer simply a matter of supporting conventional industrial production; these systems increasingly support the cognitive layer that coordinates other forms of production. The physical and digital economies are converging into one system.

The Countries That Can Convert Power Into Intelligence Gain Another Source of Capability

This brings the series back to the proposition from which it began. Artificial intelligence is turning electrical capacity into an economic input for cognition. That statement does not imply that electricity alone produces intelligence, any more than oil alone produces transportation or uranium alone produces nuclear power. Every transformation requires capital, technology, institutions, human knowledge and complex industrial systems. What has changed is that usable cognition can increasingly be produced through those systems.

A country with abundant power, advanced semiconductors, computational infrastructure, capital, scientific institutions and industries capable of deploying AI can convert physical resources into cognitive capacity at a scale unavailable to previous generations. That cognitive capacity can then influence productivity, science, manufacturing, logistics, finance, defense and government administration. The pathway is therefore no longer speculative: Energy supports the computational systems that produce machine intelligence, whose application can expand productive and strategic capability.

Countries unable to supply or reliably access the systems underneath that chain will not necessarily become technologically irrelevant, and smaller nations can remain prosperous by specializing, trading and using computational infrastructure produced elsewhere. The strategic difference lies in dependency. A country that must import increasing portions of the cognitive infrastructure used throughout its economy possesses fewer options than one capable of producing substantial quantities itself or obtaining them through resilient alliances and diversified suppliers. Reliable access to cognitive infrastructure gives a nation another instrument of capability, although the quality of the decisions made with it still determines its value.