Part VII · The Distribution Question
VII.B — Who Bears the Cost
16 min read · 3,130 words
Every transition has winners and losers. The winners write the histories; the losers populate the footnotes. The Factor Prime transition will redistribute economic claims on a scale comparable to prior paradigm shifts—industrialization, electrification, computerization—but the distribution will follow a different logic. Understanding that logic is prerequisite to understanding who thrives, who adapts, and who is displaced.
The framework developed in prior sections provides the ordering principle. The V/C ratio determines which tasks face pressure first. The hurdle rate determines which activities remain economic. The geographic signature of energy costs determines where infrastructure concentrates. The ownership structure of the actuation layer determines who captures returns when labor share declines. Each mechanism operates simultaneously; together they define the distribution of costs.
Begin with occupations.
The V/C ratio—the relationship between task value and verification cost—predicts the sequence of automation pressure. High-V/C tasks automate first because the selection gradient operates rapidly: output is observable, feedback is immediate, errors are detectable, deployment proceeds. Low-V/C tasks automate slowly because selection is constrained: output quality is ambiguous, consequences are delayed, verification requires judgment or physical observation, deployment stalls regardless of capability.
This ordering generates specific predictions about which occupational categories face pressure earliest.
High-V/C occupations facing near-term pressure:
Customer service representatives. The task is high-V/C because resolution is binary (the ticket closes or it does not), customer satisfaction is measurable, and escalation rates provide continuous feedback. Verification cost is low; value per interaction is modest but positive. The occupation employed approximately 2.9 million workers in the United States as of 2023. Automated systems already handle tier-one inquiries at many organizations; the frontier is expanding to tier-two and tier-three interactions as capability improves. The adjustment window is short—measured in years, not decades.
Content writers and copywriters. The task is high-V/C because output is directly observable (the text exists or it does not), quality metrics are available (engagement, conversion, readability scores), and iteration cycles are fast. The occupation spans marketing copy, technical documentation, journalism, and creative content. Verification cost is low for most categories; only high-stakes content (legal, medical, financial) requires expert review. The adjustment has already begun: content mills have contracted, freelance rates have compressed, and the premium for undifferentiated writing has collapsed.
Data analysts and junior researchers. The task is high-V/C because analysis can be checked against data, errors are detectable, and results are falsifiable. The occupation involves pattern recognition, statistical computation, and report generation—precisely the tasks where agent invocations excel. Verification cost is low when data quality is high; the constraint is upstream data integrity, not downstream analysis. Junior positions face pressure first; senior positions that involve judgment under ambiguity retain value longer.
Software developers (routine tasks). The task is high-V/C for code that compiles and passes tests—the verification is mechanical. The occupation differentiates sharply by task type: boilerplate code generation, bug fixing, and documentation are high-V/C and face immediate pressure; architecture design, system integration, and novel problem-solving are lower-V/C because verification requires deployment experience and judgment. Junior developers performing routine tasks face pressure; senior developers performing integration and design retain value.
Bookkeepers and accounting clerks. The task is high-V/C because financial records either balance or they do not, discrepancies are detectable, and audit trails provide verification. The occupation involves transaction recording, reconciliation, and report preparation—structured tasks with clear success criteria. Verification cost is low; the constraint is access to financial systems and regulatory compliance. The occupation has already contracted significantly from automation predating the current transition; Factor Prime accelerates an existing trajectory.
Low-V/C occupations with longer adjustment windows:
Registered nurses. The task is low-V/C because patient outcomes unfold over time, verification requires clinical judgment, and errors may not manifest until long after the intervention. The occupation involves physical presence, real-time assessment, and emotional labor that agents cannot provide. Liability constraints are severe: no institution will deploy autonomous nursing care without verification infrastructure that does not yet exist. The adjustment window is measured in decades, not years.
Attorneys (litigation and advisory). The task is low-V/C because legal outcomes depend on adversarial processes, judicial discretion, and case-specific facts. Document review and contract analysis are high-V/C subtasks facing pressure; courtroom advocacy, client counseling, and strategic judgment are low-V/C and retain value. The profession differentiates sharply: commodity legal services face compression; complex litigation and high-stakes advisory retain premium pricing.
Construction trades. The task is low-V/C because verification requires physical inspection, building codes vary by jurisdiction, and quality depends on site-specific conditions. The occupation involves manipulation of physical materials in unstructured environments—precisely where the actuation bottleneck binds tightest. Robotic construction exists but remains limited to controlled environments and repetitive tasks. The adjustment window is long; physical dexterity and on-site judgment remain scarce.
Maintenance technicians. The task is low-V/C because equipment failures are idiosyncratic, diagnostic processes require physical observation, and repair success depends on site-specific factors. The occupation involves troubleshooting under uncertainty in environments that agents cannot access. Predictive maintenance systems improve scheduling; they do not replace the technician who opens the panel and identifies the fault.
Healthcare aides and personal care workers. The task is low-V/C because quality depends on human relationship, physical presence, and real-time judgment about vulnerable individuals. The occupation involves assistance with daily living, emotional support, and monitoring that requires embodied presence. Verification is difficult: how does one measure the quality of care provided to an elderly person with dementia? Liability constraints are severe. The adjustment window is long, though wage pressure may persist from other sources.
The V/C ordering is not a ranking of human worth or even of task difficulty. It is a ranking of how quickly the selection gradient can operate. A task can be cognitively simple but low-V/C if verification is expensive; a task can be cognitively complex but high-V/C if output is directly observable. The ordering determines sequence, not ultimate fate.
The V/C ratio is not fixed. Three mechanisms shift it over time.
Sensor infrastructure reduces verification cost. When cameras, microphones, accelerometers, and environmental sensors proliferate, tasks that previously required human inspection become verifiable by algorithm. A delivery that once required a signed receipt can be verified by photograph; a surgical procedure that once required peer review can be verified by video analysis; a construction defect that once required site inspection can be detected by drone survey. As sensor density increases, V/C ratios rise, and tasks move from the low-V/C category to the high-V/C category. The occupations that seem protected today may face pressure tomorrow as verification infrastructure expands.
Regulatory frameworks redefine what counts as verification. When regulators accept algorithmic attestation in place of human certification, V/C ratios rise. A financial audit that once required CPA signature may eventually accept algorithmic verification if regulators permit. A medical diagnosis that once required physician attestation may eventually accept AI-assisted triage if liability frameworks adapt. The pace of regulatory adaptation determines how quickly V/C ratios shift in regulated industries. The United States, European Union, and China will adapt at different rates, creating arbitrage opportunities and coordination failures.
Insurance products change who bears verification cost. When insurers offer coverage for autonomous operations at acceptable premiums, the effective verification cost falls because the risk of verification failure transfers from the deployer to the insurer. The emergence of standardized insurance products for agent operations—similar to professional liability coverage for human practitioners—would accelerate deployment in domains where verification is currently prohibitive. The absence of such products constrains deployment regardless of capability.
Workers in occupations where these mechanisms are advancing face accelerating pressure. Workers in occupations where these mechanisms are stalled face extended adjustment windows. The signals are observable: sensor deployment rates, regulatory rulings, insurance product development. Each provides advance notice of where V/C ratios are shifting.
The hurdle rate creates a floor that capability alone cannot breach.
Recall the mechanism from IV.E: a kilowatt-hour routed to inference must generate more value than the same kilowatt-hour routed to Bitcoin mining. If inference generates less value than mining, the capacity routes to mining. This creates a floor beneath which cognitive deployments are uneconomic—not because agents cannot perform the task, but because the value produced does not justify the energy expenditure.
The floor has a labor market implication that standard automation analysis misses. A task becomes uneconomic before it becomes automatable. Consider a customer service interaction that generates 0.50, the task is uneconomic for agent deployment regardless of capability. The task remains—someone must handle the interaction—but it remains with human workers not because humans are better but because the energy floor has not been cleared.
This creates a counterintuitive pattern. Low-margin cognitive work faces pressure from the floor even when agents could technically perform the work. The pressure manifests not as dramatic displacement but as margin compression: the activity becomes unprofitable, the employer reduces headcount, the industry consolidates. The workers displaced are not replaced by agents; they are replaced by nothing because the activity ceases to be economic.
High-margin cognitive work clears the floor comfortably. Complex reasoning, specialized analysis, high-stakes decision support—these tasks generate sufficient value per unit energy to justify the inference cost. Workers in these roles face a different dynamic: augmentation rather than substitution. The agent handles the routine components; the human handles the exceptions. The human’s productivity rises; headcount may fall but compensation per remaining worker may rise as well.
The floor is not static. Mining difficulty adjusts; Bitcoin price fluctuates; inference costs decline as hardware improves. The floor rises and falls with these parameters. But the mechanism is structural: energy has an alternative use with deterministic yield, and that alternative disciplines all cognitive deployments.
Geography concentrates around energy endowments.
The production function established in Part IV makes the connection explicit: Factor Prime is energy structured through computation. Every agent invocation requires compute; every compute cluster requires power; every power source requires generation, transmission, and cooling. The physical substrate has geographic signatures that algorithmic improvement cannot overcome.
The most valuable locations for computational infrastructure cluster around specific endowments:
Abundant, cheap, reliable power. The hurdle rate is lowest where electricity costs are lowest. West Texas, with its wind and solar resources, offers power at rates below $30/MWh for long-term contracts. The Pacific Northwest, with its hydroelectric base, offers similar advantages. Quebec, with its surplus hydro capacity, attracts datacenter investment despite higher labor costs. Nordic countries, with their renewable portfolios, compete for AI infrastructure. The locations with the cheapest reliable power are the locations where cognitive deployments most easily clear the floor.
Favorable climate for cooling. Computation generates heat; heat must be dissipated. Locations with cool climates reduce cooling costs; locations with access to cold water (ocean, lake, river) enable more efficient heat rejection. Iceland combines geothermal power with natural cooling; datacenter operators pay premium rents for the privilege. Northern locations face shorter construction seasons but lower operating costs; the tradeoff favors the locations where infrastructure already exists.
Proximity to fiber networks. Latency matters for some applications; bandwidth matters for all. Locations with existing fiber connectivity attract applications that require real-time interaction. Northern Virginia, with its fiber density and proximity to internet exchange points, remains attractive despite higher power costs. The tradeoff between power cost and network latency creates differentiated locations for differentiated workloads.
Permissive regulatory environments. Permitting timelines, environmental review, grid interconnection queues—these determine how quickly capacity can be built. Texas, with its independent grid and expedited permitting, attracts investment that more restrictive jurisdictions cannot capture. The tradeoff between regulatory permissiveness and other factors creates geographic arbitrage.
The implications for labor markets are severe. Workers in locations without energy endowments face a different future than workers in locations with them. A customer service center in Ohio competes with a datacenter in West Texas; the datacenter clears the hurdle rate while the call center does not. The jobs migrate—not to other cities but to computational infrastructure that requires minimal human labor for operation. The geographic concentration of AI infrastructure is not incidental; it follows from the production function.
The concentration creates regional winners and losers. Regions with energy endowments capture infrastructure investment, construction employment, property tax revenue, and the ancillary services that datacenters require. Regions without energy endowments lose the cognitive work that previously located there (for labor cost or timezone reasons) without gaining the infrastructure work that replaces it. The pattern resembles prior resource-driven geographic shifts—oil towns, mining towns, port cities—but the resource is electricity rather than minerals.
The ownership structure determines who captures returns when labor share declines.
This is the core distributional question. If cognitive capability commoditizes while actuation bottlenecks persist, returns flow to whoever owns the actuation layer: physical throughput, trusted interfaces, verification infrastructure, liability capacity. The returns do not flow to workers; they flow to capital owners. Labor share—the fraction of national income accruing to wages rather than capital returns—may decline.
The historical pattern provides precedent. Labor share in the United States declined from approximately 64% in 1970 to approximately 58% in 2020, a six-percentage-point shift representing hundreds of billions of dollars in annual income. The decline occurred gradually, obscured by economic growth that raised absolute living standards even as relative shares shifted. The Factor Prime transition may accelerate this pattern.
Who owns the actuation layer?
Utilities. Investor-owned utilities hold the generation and transmission assets that Factor Prime requires. The largest U.S. utilities by market capitalization—NextEra Energy, Duke Energy, Southern Company, Dominion Energy—are publicly traded, with ownership dispersed across institutional investors, pension funds, and retail shareholders. The ownership is broad but not universal; workers without equity exposure do not participate in the returns.
Infrastructure funds. Private infrastructure funds—Blackstone Infrastructure Partners, Brookfield Infrastructure, Global Infrastructure Partners—hold datacenter, transmission, and generation assets. The limited partners are predominantly institutional investors: pension funds, sovereign wealth funds, endowments, insurance companies. The beneficiaries are retirees, government employees, and university endowments—diffuse but not comprehensive.
Hyperscalers. Microsoft, Amazon, Google, and Meta control substantial datacenter capacity. Ownership is publicly traded, with large positions held by index funds (Vanguard, BlackRock, State Street) and institutional investors. Retail shareholders participate; workers without equity exposure do not.
Chip manufacturers. NVIDIA, TSMC, Intel, and Samsung control the fabrication capacity that translates energy into compute. Ownership is publicly traded, with significant concentration among institutional investors. Geographic distribution of ownership varies: TSMC’s shareholder base is heavily Taiwanese; NVIDIA’s is heavily American.
Sovereign wealth funds. Norway’s Government Pension Fund, Saudi Arabia’s Public Investment Fund, Abu Dhabi’s Mubadala—these entities hold infrastructure positions globally. The beneficiaries are citizens of the sponsoring nations, not global labor markets.
The pattern is clear: ownership of the actuation layer is concentrated among capital owners, not workers. If returns migrate from cognitive work (where labor captures value through wages) to actuation infrastructure (where capital captures value through returns), the distributional consequences follow mechanically. The transition enriches those with equity exposure to the actuation layer; it does not enrich those whose only claim on economic output is labor income.
The distributional implication is not a prediction of mass impoverishment. The economy may grow; absolute living standards may rise. The implication is relative: capital’s share rises while labor’s share falls. Those with both labor income and capital ownership may thrive; those with labor income alone face stagnation or decline.
Can adjustment mechanisms operate at the pace the production function permits?
The historical pattern is reassuring: prior transitions created displacement but also created new employment. The weaving shed replaced the handloom weaver; it also created factory jobs. The automobile replaced the horse; it also created assembly line employment. The computer replaced the typing pool; it also created knowledge work. The reinstatement effect—new tasks emerging to absorb displaced labor—has historically offset displacement.
The question is speed.
Prior transitions unfolded over decades. The electrification of American industry took roughly forty years, from the 1880s to the 1920s. The computerization of office work took roughly thirty years, from the 1960s to the 1990s. Adjustment mechanisms—retraining programs, geographic mobility, new firm formation—operated on similar timescales. Workers displaced in their twenties could retrain and find new employment before retirement.
The Factor Prime transition may unfold faster. Capability improvement compounds: models that assist in training better models; algorithms that optimize algorithms; hardware roadmaps accelerated by AI-assisted design. The feedback loops that made prior transitions gradual may make this transition rapid. If capability advances faster than adjustment mechanisms can operate, the historical pattern breaks.
Three adjustment mechanisms matter:
Retraining and education. Workers displaced from high-V/C occupations must acquire skills for low-V/C occupations—or for occupations that do not yet exist. The current educational infrastructure is not optimized for adult retraining at scale. Community college enrollment has declined; vocational programs are underfunded; the signaling value of credentials often exceeds their human capital value. If the transition requires millions of workers to move from customer service to healthcare, or from content writing to construction, the retraining infrastructure must expand dramatically.
Geographic mobility. Workers in regions losing cognitive work must move to regions gaining infrastructure work—or to regions where low-V/C occupations concentrate. Geographic mobility in the United States has declined for decades; fewer Americans move across state lines than in previous generations. Housing costs in prosperous regions exceed what displaced workers can afford; social networks anchor workers to declining areas. If the transition requires geographic reallocation at scale, the barriers to mobility must fall.
New firm formation. The reinstatement effect depends on entrepreneurs identifying new tasks and organizing resources to perform them. New firm formation in the United States has declined relative to historical baselines; the fraction of employment at young firms has fallen. If the entrepreneurial function itself becomes automatable—if agents can identify opportunities, assemble resources, and execute faster than human entrepreneurs—then new firm formation may occur without new employment. The species that buys itself may reinstate tasks without reinstating jobs.
The institutional infrastructure for adjustment was built for transitions that unfolded over decades. If the Factor Prime transition unfolds over years, the infrastructure is inadequate. The cost falls on workers who cannot retrain fast enough, cannot move to opportunity, and cannot benefit from firm formation that occurs without them.
The policy implications are developed in the following section. The distributional pattern is not inevitable; it depends on choices about taxation, ownership, adjustment support, and the pace of deployment. The production function determines the pressure; institutions determine the response.
The transition creates surplus. The question is who captures it.