en flag
en flag

Why Efficiency Is the Most Dangerous Word in Energy Policy

When Optimization Removes the Capacity to Survive

8/23/202613 min read

windmills on green field under white sky during daytime
windmills on green field under white sky during daytime

When Optimization Removes the Capacity to Survive

Efficiency is one of the most powerful ideas in modern economic and engineering practice because, when used correctly, it allows societies to produce more value with fewer inputs. It lowers energy consumption, reduces waste, improves asset utilization, raises productivity, and can reduce both financial and environmental cost. The problem begins when efficiency is treated not as one design objective among several but as a universal measure of system quality. A system can become more efficient while simultaneously becoming more fragile, less reversible, more dependent on narrow assumptions, more difficult to repair, and more exposed to low-frequency failures. The governing distinction is therefore not between efficiency and inefficiency. It is between productive efficiency, which removes genuine waste while preserving the capacity to absorb error, and brittle efficiency, which improves measured performance by consuming the margins that made the system resilient. The strategic risk is not that societies pursue efficiency. It is that they increasingly optimize systems faster than governance can identify which apparently redundant resources are actually carrying resilience.

This distinction matters especially in energy because energy is not simply another commodity. Modern civilization converts energy into mobility, heat, cooling, food production, water treatment, telecommunications, healthcare, manufacturing, finance, computation, defense, logistics, and almost every other large-scale economic function. A failure inside a discretionary consumer market can remain relatively contained; a failure inside foundational energy infrastructure can propagate across multiple sectors simultaneously. Energy policy therefore cannot be judged solely through levelized cost, utilization, average output, nominal efficiency, or short-run consumer price. Those measures are useful, but they describe only part of the system. A complete evaluation must also account for redundancy, reserve capacity, maintenance, storage, transmission availability, fuel diversity, recovery time, black-start capability, supply-chain concentration, correlated failure, reversibility, and the economic value of avoiding catastrophic interruption. Once those variables are included, some assets that appear inefficient under normal conditions begin to resemble insurance, while some assets that appear highly efficient reveal a hidden dependence on unusually favorable operating assumptions.

The structural problem can be stated simply: efficiency measures performance inside a boundary, while resilience determines whether the boundary survives when conditions move outside the expected case. If the analytical boundary contains only normal operating conditions, unused capacity looks wasteful. If the boundary includes extreme demand, infrastructure failure, geopolitical disruption, prolonged weather events, cyberattack, fuel shortages, component shortages, or simultaneous failures across interconnected systems, the same unused capacity may become essential. This is why the phrase “we have never needed it” is one of the weakest arguments available for removing a resilience asset. Fire suppression systems are valuable precisely because they are rarely activated. Emergency inventories, reserve generators, spare transformers, backup communications, additional transmission paths, trained operators, water reserves, cybersecurity redundancy, and manual fallback procedures can remain dormant for years while still carrying substantial economic value. Their absence is discovered only when the system enters a state for which average-condition optimization was never designed.

The mistake is reinforced by accounting. Efficiency gains are usually visible immediately. Remove spare capacity and capital expenditure declines. Reduce staffing buffers and labor productivity rises. Compress inventories and working capital improves. Defer maintenance and current-period cost falls. Centralize a supply chain and unit economics improve. Raise asset utilization and return on invested capital increases. Resilience losses, by contrast, remain largely invisible until something goes wrong. This creates a systematic measurement asymmetry: the benefit of removing slack appears continuously, while the cost of removing slack appears discontinuously. Organizations therefore receive repeated short-term confirmation that the optimization was correct right up until the system reaches a condition in which the removed margin would have been valuable. The eventual failure can then appear exceptional or unpredictable even though the architecture had been systematically increasing exposure for years.

This is why utilization should not be confused with health. A system operating continuously near its theoretical capacity can look superior to one carrying significant unused capacity, but the unused capacity may be exactly what prevents small disruptions from becoming queues, bottlenecks, service failures, or cascading outages. Queueing theory has long demonstrated that waiting times and congestion rise sharply as utilization approaches capacity in systems subject to stochastic variation. Energy networks experience an analogous problem at a larger scale. Demand is not perfectly predictable. Generators fail. transmission assets require maintenance. weather conditions change. Fuel availability fluctuates. A network optimized only around expected conditions can therefore become increasingly unstable as spare capacity approaches zero. The apparent inefficiency of maintaining reserve capacity must be compared not with zero, but with the expected and tail cost of the failures that reserve capacity prevents.

The same logic applies to maintenance. Preventive maintenance is economically peculiar because success often produces no visible event. A component is inspected, serviced, and does not fail. The organization therefore sees the cost but not the counterfactual failure. Under aggressive optimization pressure, this makes maintenance an attractive target for deferral. The asset continues operating, confirming the belief that maintenance can be postponed again. Over time, however, degradation accumulates. The system can reach a state in which several components have reduced safety margins simultaneously, creating correlated vulnerability that historical failure rates no longer represent accurately. Deferred maintenance is therefore not simply a future operating expense. It can be a transfer of risk from visible current cost into less visible future system fragility.

Supply chains demonstrate the same pattern. Concentrating procurement among a smaller number of highly efficient suppliers can reduce transaction costs, simplify logistics, improve purchasing leverage, and increase consistency. Those benefits are real. But supplier concentration also reduces substitution capacity. If supposedly separate suppliers depend on the same geographic region, port, semiconductor foundry, raw material, cloud provider, electricity network, or transportation corridor, nominal diversification may conceal common-mode failure. The correct resilience question is therefore not “How many suppliers do we have?” but “How many genuinely independent failure paths do we have?” Counting redundant entities without mapping shared dependencies can produce the illusion of resilience while preserving the same underlying point of failure.

This problem extends directly into energy transition policy. New generation technologies can deliver substantial gains in cost, emissions, modularity, or deployment speed, but the system consequence depends on the architecture around them. A generation technology cannot be evaluated solely by its marginal operating cost if its integration requires additional transmission, storage, reserve generation, network reinforcement, balancing services, interconnection, or long-duration backup. Conversely, technologies that appear expensive at the asset level may provide system services whose value is not captured by a narrow unit-cost comparison. The relevant analytical unit is therefore increasingly the whole system required to deliver reliable energy, not the isolated generating asset. Energy economics becomes misleading whenever it compares visible production costs while leaving system-balancing, resilience, infrastructure, and failure-recovery requirements outside the boundary.

The deeper problem is that optimization often changes the system it is optimizing. Efficiency lowers cost, lower cost can increase demand, greater demand increases scale, and greater scale increases dependence on the optimized infrastructure. This feedback can transform a modest efficiency gain into a larger systemic commitment. The rebound effect in energy economics illustrates one part of this mechanism: when using an energy service becomes cheaper, consumers and businesses may consume more of it, partially offsetting the engineering savings. Rebound does not imply that efficiency is undesirable or that savings always disappear. It demonstrates something more important for policy: an engineering improvement at the unit level and a reduction in total system exposure are different claims. Governance must measure the system outcome rather than assume it from the component outcome.

Artificial intelligence and large-scale computation intensify the issue. Data centers can become more computationally efficient per operation while total electricity consumption rises because demand for computation expands faster than efficiency improves. More efficient chips can reduce the marginal cost of inference and thereby increase the number of inferences economically viable to run. Improvements in cooling, utilization, and workload scheduling can make compute cheaper while increasing the strategic dependence of companies and governments on continuous electricity supply. Efficiency therefore does not necessarily reduce dependence. In some cases it enables much greater dependence by making the service economical enough to embed everywhere. The policy implication is that efficiency gains should be evaluated alongside demand elasticity and dependency expansion, not only unit consumption.

Financial systems provide a useful analogy. Leverage can improve capital efficiency because less equity supports a larger asset base. During normal conditions, this can improve returns and appear economically superior. But increasing leverage reduces the distance between ordinary volatility and insolvency. The system becomes more efficient in its use of capital and less tolerant of error. Energy systems can experience an equivalent compression of margin when reserves, inventories, redundancy, maintenance capacity, spare parts, storage, or alternative pathways are removed. In both cases, efficiency improves as the system's ability to absorb deviations declines. The measured return increases partly because the architecture has transferred risk from everyday operation into tail events.

This distinction between normal-condition efficiency and tail-condition viability is central. Many models are calibrated primarily on historical observations. If historical conditions do not contain the relevant extreme event, the absence of failure can be misinterpreted as evidence that the architecture is robust. But infrastructure systems often possess nonlinear thresholds. A grid can absorb disturbances until reserve margins become insufficient. A supply chain can continue operating until inventories and alternative routes are exhausted. A financial institution can remain solvent until liquidity disappears faster than assets can be sold. An ecosystem can absorb pressure until regenerative capacity is exceeded. The system can therefore look stable immediately before the condition that reveals its structural weakness. Average historical performance is not proof of survivability under a different regime.

The governance challenge is made harder by organizational incentives. Executives are often rewarded for visible current performance. Asset utilization, return on capital, cost reduction, productivity, quarterly earnings, operating margin, and delivery speed are measurable continuously. Avoided catastrophe is not. A manager who eliminates expensive redundancy receives a visible financial benefit. A manager who preserves redundancy may appear less efficient for years without receiving observable evidence that the decision was valuable. The incentive structure therefore systematically favors consumption of resilience unless governance explicitly protects it. This is why safety-critical industries separate some resilience decisions from ordinary optimization. Aviation, nuclear power, financial clearing, critical infrastructure, and medicine use mandatory redundancy, inspection, reserve, or review requirements precisely because competitive pressure would otherwise create strong incentives to reduce those margins over time.

The critical distinction is therefore between waste and reserve capacity. Waste consumes resources without materially improving output, resilience, learning, safety, or option value. Reserve capacity produces little visible output under expected conditions but protects the system under defined deviations. Eliminating waste is genuine efficiency. Eliminating reserve capacity because it resembles waste is brittle optimization. The management challenge is determining which category an apparent inefficiency belongs to before removing it.

This requires a broader economic model. A narrow optimization might evaluate an asset through expected operating cost alone. A resilience-aware evaluation should incorporate at least operating cost, expected failure loss, recovery cost, duration of interruption, correlated-failure exposure, substitutability, irreversible harm, and option value. Even expected-value analysis can become inadequate when probabilities are uncertain or consequences are extreme. A one-in-a-century estimate can create false precision when the underlying system is changing rapidly enough that historical frequencies no longer represent future conditions. Under such circumstances, robustness across plausible scenarios may matter more than selecting the mathematically cheapest configuration under one estimated distribution.

Reversibility becomes particularly important. Some efficiency decisions are easy to undo. Others create path dependency. Removing inventory today can be reversed tomorrow if suppliers remain available. Closing domestic manufacturing capacity, losing specialized skills, dismantling infrastructure, concentrating strategic production abroad, or redesigning a grid around highly specialized dependencies may take years or decades to reverse. The same apparent cost saving therefore carries different strategic significance depending on the time required to restore the removed capacity. The correct question is not merely “How much money does this save?” It is “How quickly can we rebuild what we are removing if the assumptions supporting this optimization fail?”

This leads to a more rigorous concept of resilience-constrained efficiency. Rather than maximizing output per unit input and subsequently adding safety considerations, the system defines non-negotiable resilience requirements first and optimizes within them. Minimum reserve margins, recovery-time requirements, maintenance standards, maximum dependency concentrations, critical spare inventories, black-start capability, cyber isolation, manual overrides, water availability, fuel security, workforce capability, and emergency authority become constraints rather than variables available for continuous cost reduction. Once those constraints are defined, efficiency remains valuable. The objective becomes maximum useful performance subject to survivability requirements.

The difference is more than semantic. In a conventional optimization, resilience competes with efficiency and must continually justify its cost. In constrained optimization, resilience defines the feasible operating region, and efficiency selects the best solution inside that region. The first architecture naturally erodes margins whenever economic pressure rises. The second treats selected margins as load-bearing conditions that cannot be traded away casually.

The same approach changes how energy portfolios should be evaluated. Diversity is frequently discussed as a political compromise among technologies. Its more important function can be structural. Technologies with different fuel sources, operating profiles, geographic dependencies, supply chains, construction times, failure modes, and recovery characteristics can reduce correlated exposure. Diversity that merely multiplies assets sharing the same underlying vulnerability provides less protection. The relevant metric is therefore not technological variety for its own sake but failure-mode independence. A resilient portfolio is one in which the same shock does not simultaneously disable every major pathway.

Time is another hidden variable. Efficiency calculations often compare current costs while resilience benefits operate across much longer horizons. Infrastructure can last decades. Political cycles last years. Corporate performance incentives can operate quarterly. Maintenance decisions made today can influence reliability ten years later. Supply-chain concentration can remain harmless until geopolitical conditions change. Energy investments can lock societies into technical architectures for generations. Short-horizon optimization therefore systematically underweights long-horizon consequences unless governance deliberately corrects the mismatch.

This is where efficiency can become civilizational debt. Cheap and reliable energy enables expansion. Expansion increases economic complexity and dependence. Efficiency then removes apparent redundancy from the enlarged system. Reduced costs encourage further scale. The reinforcing loop becomes lower cost → greater scale → tighter optimization → lower slack → greater dependence → larger consequences of failure. Nothing in this sequence is individually irrational. Each decision can be defensible locally. The systemic danger emerges because the combined architecture becomes increasingly dependent on continued favorable conditions. The civilization is effectively borrowing resilience from the future to improve performance in the present.

Civilizational debt is therefore not identical to financial debt. It is the accumulation of obligations, dependencies, maintenance deficits, structural fragilities, and irreversible commitments created by systems whose current performance depends on assumptions future institutions must continue satisfying. A grid requiring uninterrupted access to specific components, a food system dependent on narrow energy inputs, an economy dependent on continuous digital infrastructure, or a logistics network dependent on highly concentrated transport corridors can all accumulate this form of debt. The cost becomes visible when the assumed condition fails and future society must absorb the repair burden.

A mature energy policy should therefore ask a different set of questions before removing any major margin. What failure was this apparent redundancy protecting against? Are the alternatives genuinely independent? What happens when several assumptions fail simultaneously? What is the recovery time? Can removed capacity be restored before downstream harm becomes irreversible? Who bears the loss if the optimization fails? Does the organization benefiting from the efficiency gain also bear the tail risk, or has that risk been externalized onto customers, workers, taxpayers, communities, or future governments? An efficiency claim that cannot answer these questions is incomplete.

The externalization question is especially important because narrow optimization frequently creates private benefit while transferring resilience cost elsewhere. A company can reduce inventories while expecting suppliers to absorb volatility. An electricity provider can reduce reserve investment while society bears outage consequences. A platform can centralize infrastructure while users bear concentration risk. A government can defer maintenance while future taxpayers finance reconstruction. The apparent efficiency exists partly because the accounting boundary excludes the party carrying the downside. Expanding the boundary can materially change whether the optimization remains economically attractive.

For this reason, the most sophisticated definition of efficiency is not “doing more with less.” It is doing more with less while preserving the capabilities required when the model is wrong. This introduces four conditions that narrow metrics routinely omit: detection, absorption, recovery, and adaptation. The system must detect deterioration before failure becomes irreversible, absorb disturbances without losing essential function, recover within an acceptable period, and adapt when the operating regime changes. Removing resources that perform these functions can improve measured efficiency while reducing total system performance across a sufficiently long horizon.

There is also an institutional dimension. Resilience depends not only on physical redundancy but on decision rights. Operators must sometimes possess the authority to stop systems, reduce output, reject unsafe optimization, or escalate anomalous conditions. An organization that maximizes efficiency while punishing interruptions can create a culture in which warning signals are suppressed because stopping the process carries immediate career or financial cost. The physical system may contain safety mechanisms while the institutional system neutralizes them. Resilience therefore requires that shutdown, refusal, escalation, and maintenance remain legitimate outcomes rather than being treated automatically as performance failures.

Artificial intelligence can improve this architecture if used carefully. AI can continuously monitor complex systems, identify emerging bottlenecks, estimate failure correlations, detect maintenance anomalies, simulate stress conditions, optimize reserve allocation, and identify where apparent redundancy provides little real protection because supposedly independent assets share common dependencies. But AI can also intensify brittle optimization if objective functions reward only throughput, cost, utilization, or efficiency. A sufficiently capable optimization system will become increasingly effective at finding and eliminating margins unless those margins are represented as constraints. The lesson is identical to the broader governance problem: optimization does not know what must not be optimized away unless the architecture tells it.

The strongest energy systems will therefore not be those that maximize every resource continuously. They will be those that distinguish idle waste from stored optionality. Redundancy can be insurance. Slack can be maneuverability. Maintenance can be future availability. Diversity can be protection against common-mode failure. Inventory can be time purchased against disruption. Skilled human operators can be recovery capacity. Spare infrastructure can be an option against conditions the forecast failed to anticipate. Reversibility can be protection against ignorance.

None of these principles argues for unlimited redundancy. Excess capacity has opportunity cost. Resources committed to unnecessary protection cannot be used elsewhere, and poorly designed resilience programs can become expensive institutional inertia. The relevant objective is therefore not maximal resilience any more than maximal efficiency. It is bounded optimization: the smallest resilience architecture capable of keeping the system inside acceptable failure and recovery conditions across plausible regimes.

This produces a substantially stronger governing law than the conventional instruction to “be efficient”:

Do not optimize away a capability whose replacement time exceeds the time available before its absence creates unacceptable harm.

The rule captures the relationship among redundancy, reversibility, time, and consequence. Removing a margin is rational when it is genuinely unnecessary or can be restored quickly. It becomes strategically dangerous when restoration takes longer than the system can survive without it.

The distinction is ultimately temporal. Efficiency realizes value immediately. Resilience often realizes value in the future, under conditions nobody can precisely predict. Institutions that discount the future aggressively will therefore consume resilience even while believing they are improving the system. Mature governance must deliberately resist that bias because foundational infrastructure is inherited. One generation's optimization becomes another generation's constraint.

The larger conclusion is not that efficiency is morally suspect or technically obsolete. Efficiency remains essential to prosperity, decarbonization, competitiveness, affordability, and resource stewardship. Waste is not resilience. Inefficient systems can consume the very resources needed to survive.

The danger begins when efficiency becomes a totalizing metric—when lower cost is assumed to mean better architecture, high utilization is assumed to mean stronger performance, and unused capacity is assumed to be economically irrational.

Those conclusions do not follow automatically.

The correct measure of an energy system is not simply how cheaply it performs when the world behaves as expected.

It is how efficiently it performs while retaining the capacity to remain functional when the world does not.

That is the distinction between optimization and resilience.

And it is why efficiency becomes dangerous precisely when a society becomes so successful at eliminating apparent waste that it can no longer distinguish waste from the capacity that was keeping the system alive.