Open Source Executive Briefing CW32: When the Bottleneck Moves, What Remains Valuable?

The most dangerous bottleneck is not always the one constraining growth. It may be the one protecting the current business model. The canonical Bottleneck Migration framework now develops this argument together with the AI Capacity Dividend: what becomes scarce next, and what should leaders do with the capacity released along the way?

Artificial intelligence is creating a resource paradox. It is increasing demand for electricity, accelerated computing, cooling, grid connections, specialist capability and capital. At the same time, it is lowering the cost of research, simulation, coordination and decision preparation in systems that produce those constraints. The first effect is visible immediately. The second depends on whether institutions can convert cognitive leverage into physical and operating capability through a bottleneck migration strategy.

This distinction matters because scarcity can produce exceptional economics. A company controlling a constrained resource, channel or capability can earn legitimate scarcity rent while the market waits for supply to expand. The strategic error is to confuse that temporary premium with permanent enterprise value.

AI may improve an incumbent’s current operating model. It may also help weaken the constraint that made the model profitable. The broader AI business model disruption begins when intelligence changes the constraint system itself. When that happens, the decisive question is no longer how efficiently the company can monetize the bottleneck. It is what remains valuable after the bottleneck moves.

Scarcity is an opportunity and a warning

The physical demand is real. The International Energy Agency projects that global data-centre electricity consumption could rise to around 945 terawatt-hours by 2030, approximately double its recent level. The IEA also expects data centres to account for about three per cent of global electricity demand by that date. These figures do not prove that AI will exhaust the energy system, but they establish that power, interconnection and infrastructure are becoming strategic constraints rather than background utilities.

The response cannot be limited to expanding the current architecture. More generation, facilities and cooling are necessary because current demand must be served. But investment should also examine whether the architecture producing the constraint can change. Better cooling solves today’s thermal problem. More efficient computing, workload-specific systems, improved interconnects and different energy arrangements may change the amount of heat and infrastructure required to produce useful computation.

This is the executive tension: a profitable constraint funds the organization while simultaneously creating an incentive to preserve the system that produced it. The moment of greatest scarcity rent may therefore be the moment when the successor model most needs capital.

Transferable value is the real strategic asset

Assets do not become transferable merely because they appear on the balance sheet. A data centre, distribution network, patent portfolio or customer base can lose strategic relevance when the customer problem changes. Transferable value exists when the organization can apply its relationships, data, infrastructure, expertise and judgment to the next constraint system.

A utility’s durable value may shift from exclusive electricity supply toward orchestration, resilience and trust across a more distributed grid. A semiconductor company’s value may shift from access to scarce chips toward architecture, software ecosystems and useful computation per unit of energy. A cloud provider’s value may move from raw capacity toward security, reliability and coordination across heterogeneous infrastructure.

The distinction becomes clearer through the idea of problem decay. Organizations usually plan for a competitor to deliver a better version of the existing solution. More disruptive transitions reduce the importance of the original problem. Streaming did not build a more efficient rental store. Digital photography did not improve film processing. LED lighting did not merely cool incandescent lamps more effectively. In each case, part of the incumbent operating system became unnecessary.

AI-era strategy therefore requires a more uncomfortable scenario: what if the constraint protecting the business weakens by ninety per cent? What would customers still pay for? Which capabilities would become more valuable? Which assets would become stranded? Which teams possess the judgment needed to transfer the organization rather than defend its current mechanism?

Capital allocation must precede certainty

Organizations often delay successor investment until the new system becomes commercially obvious. By that point, the capabilities required to compete may already have accumulated elsewhere. A disciplined response does not abandon the current business. It creates a portfolio in which today’s scarcity funds both immediate capacity and options for the world after the constraint.

The portfolio should distinguish three forms of investment. The first expands the constrained system because customers need capacity now. The second improves the productivity and sustainability of that system. The third explores architectures that could make the constraint materially less important. The risk is not that one horizon is inherently superior. It is that the first horizon consumes all available capital because its returns are easier to measure.

AI increases this pressure by expanding the number of plausible options. More analysis does not create more strategic capacity when governance cannot decide which options deserve funding. This is the allocation problem behind the AI Capacity Dividend: recovered capacity has strategic value only when leaders direct it deliberately. Human judgment remains the validation gate between technical possibility and durable customer outcomes, acceptable external costs and executable operating models.

Operating models determine whether abundance arrives

Research signals show the mechanism without proving the outcome. The US Department of Energy has reported machine-learning systems adjusting magnetic controls in real time to improve plasma stability in tokamaks. The National Renewable Energy Laboratory’s OptGrid work demonstrates distributed coordination of grid-edge resources. The Department of Energy’s materials programmes describe closed-loop systems connecting prediction, synthesis, characterization and analysis.

These developments do not establish commercial fusion, unrestricted electricity supply or a guaranteed materials breakthrough. They show that AI can lower the cost of search, control and coordination inside technically difficult systems. Whether that capability becomes productive capacity depends on manufacturing, regulation, financing, standards, infrastructure and institutional ownership.

This is why AI operating model redesign cannot stop at use-case identification. AI-native professional services must validate the value hypothesis across the full operating system: what capability becomes possible, which dependency remains binding, who owns the decision, what evidence justifies the next investment and where human judgment must intervene.

Leadership Question

If the bottleneck protecting your margins weakened by ninety per cent, which capabilities would customers still pay for—and are you funding them now?

The executive decision

Do not treat today’s scarcity as proof of permanent value. Serve the current constraint, but separate the rent it creates from the capabilities that can travel beyond it.

Fund the successor before the market forces the transition. That means placing capital behind options that may reduce the importance of the current solution, assigning leaders who can operate across horizons and tracking where scarcity is migrating rather than where it was historically profitable.

The companies most exposed to AI business model disruption may not be those with the weakest current position. They may be those whose strongest margins depend on a bottleneck that intelligence can help redesign.

The durable strategic objective is not to own the constraint forever. It is to build the capabilities that remain valuable when the constraint no longer matters.