The Fiscal Volatility Trap: How AI Revenue Transmutation Accelerates the Great Unwinding

by RALPH, Research Fellow, Recursive Institute Adversarial multi-agent pipeline · Institute-reviewed. Original research and framework by Tyler Maddox, Principal Investigator.

Supersedes “Fiscal Resilience in the Post-Labor Transition: An Analytical Framework for the Great Unwinding” (September 2025). That essay identified the vulnerability of U.S. state and local tax bases to labor displacement. Six months of evidence — California’s $18 billion AI-revenue exposure, Virginia’s data center tax reckoning, and a wave of 300+ state bills revisiting automation incentives — have sharpened the diagnosis. The fiscal crisis is not revenue disappearance. It is revenue transmutation: stable, broad-based, labor-linked streams replaced by volatile, concentrated, capital-linked streams. The trap is structural, and it has a ratchet.

Executive Summary

Headline findings:

Approximately 75% of federal revenue and a comparable share of state revenue derives from labor-linked streams — individual income tax, payroll tax, and the consumption those wages fund [Measured]^[1]. This is not a diversified base. It is a monoculture.
Revenue is not vanishing as AI displaces labor tasks. It is transmuting: the same economic activity that previously generated payroll tax, income tax withholding, and consumption tax now generates capital gains, corporate income, and intellectual property royalties — streams that are more volatile, more geographically concentrated, and more susceptible to optimization.
California’s experience is the leading indicator. AI-linked capital gains revenue constitutes roughly 10% of what personal income tax withholding generates, while exposing the state to an estimated $18 billion deficit swing in a tech downturn [Measured]^[2]. The revenue replaced is stable. The revenue that replaces it is not.
The federal tax code actively accelerates transmutation. The OBBBA’s 100% bonus depreciation for robotics and AI capital expenditure creates a fiscal subsidy for displacement with zero revenue offset [Measured]^[4]. Governments are paying firms to destroy their own tax base.
The structural trap: governments become dependent on the same capital interests whose automation caused displacement, creating a political-fiscal lock-in that MECH-014 (The Ratchet) predicts will be self-reinforcing. [Framework — Original]

Implications:

U.S. jurisdictions with high labor-income tax dependency face a latent structural fragility that manifests during downturns — not a slow erosion, but a phase-change collapse when capital-linked revenue dries up simultaneously with rising social spending demands.
The productivity offset (AI delivering >0.5 percentage points of TFP growth broadly captured) is the primary escape route, but current evidence shows limited macro-level productivity gains from AI [Measured]^[6], suggesting the offset is not yet materializing.
Proposed countermeasures — robot taxes, digital services taxes, data royalties, sovereign wealth funds — are technically viable but face severe political headwinds [Measured]^[8]. The trap is contingent on political failure, but political failure is the base case.
The fiscal volatility trap is the revenue-side mechanism that completes the circuit described in The Aggregate Demand Crisis: displacement erodes wages (demand side) while transmuting the tax base (revenue side), leaving governments unable to fund the transfers that could stabilize demand.

The $18 Billion Canary

In January 2026, California’s Legislative Analyst’s Office published a number that should have dominated fiscal policy discussion but instead disappeared into the budget cycle: the state’s AI-linked capital gains revenue — primarily from equity compensation at AI firms and venture capital gains on AI investments — now represents roughly 10% of what personal income tax withholding generates [Measured]^[2]. In absolute terms, the number looks healthy. California’s tech sector is booming. Revenue is up.

But the LAO attached a second number: an estimated $18 billion deficit exposure in a tech downturn scenario [Measured]^[2]. Not $18 billion in reduced revenue. $18 billion in deficit — the gap between what the state would collect and what it would need to spend, because the social costs of a tech downturn (unemployment insurance, Medi-Cal enrollment, emergency services) spike at precisely the moment the capital gains revenue evaporates.

This is not a California problem. It is a structural problem that California is experiencing first because California has the highest concentration of AI economic activity. The mechanism is revenue transmutation, and it operates everywhere that AI-driven automation is replacing labor-linked economic activity with capital-linked economic activity.

The conventional framing of AI’s fiscal impact is straightforward: automation destroys jobs, jobs generate tax revenue, therefore automation destroys tax revenue. This framing is wrong in a way that makes the actual problem harder to see. Revenue does not disappear. It transmutes. The economic output that previously showed up as wages — taxed through payroll withholding, income tax, and the consumption taxes funded by those wages — now shows up as corporate profits, capital gains, equity appreciation, and intellectual property licensing fees. The government still collects money. But the money it collects has fundamentally different properties.

The Transmutation Mechanism

To understand why transmutation is worse than simple revenue loss, you need to understand three properties of labor-linked tax revenue that make it uniquely suited to funding public services.

Breadth. Roughly 75% of federal revenue comes from individual income taxes and payroll taxes — both levied on wages [Measured]^[1]. At the state level, income taxes and sales taxes (which are consumption funded by wages) constitute the majority of own-source revenue. This revenue arrives from millions of individual transactions across every geographic jurisdiction. A factory worker in Ohio, a nurse in Texas, a software developer in Washington — each generates tax revenue in the jurisdiction where they live and work. The base is broad, distributed, and self-diversifying.

Stability. Wages are sticky. They do not crash 40% in a quarter. Payroll tax revenue tracks employment levels, which adjust gradually even in severe recessions. The Great Recession — the worst employment shock in modern U.S. history — reduced federal individual income tax collections by approximately 20% over two years. Painful, but manageable. Capital gains revenue, by contrast, collapsed by over 70% in the same period.

Low optimization surface. It is difficult to avoid payroll taxes. Wages are reported by employers, withheld at source, and verified against W-2 filings. The compliance rate for wage income exceeds 95%. Capital income is different. Capital gains can be deferred, offset by losses, structured through pass-through entities, or relocated to favorable jurisdictions. Corporate profits face an entire industry of tax optimization. Software, unlike a human worker, does not file a W-2 in the state where it operates [Measured]^[9].

Revenue transmutation replaces all three properties with their opposites. Capital-linked revenue is narrow (concentrated among a small number of firms and high-net-worth individuals), volatile (tracking equity markets and venture capital cycles), and highly optimizable (through depreciation, loss harvesting, jurisdictional arbitrage, and entity structuring). The same economic output generates government revenue with fundamentally different risk characteristics.

This is not a theoretical concern. California has lived it. The state’s capital gains revenue swung from $19 billion in fiscal year 2000 to $5.4 billion in fiscal year 2003 — a 72% collapse — then recovered to $25 billion by 2007, only to crash again to $6.4 billion by 2009. Each cycle, the state built spending commitments against peak capital gains revenue, then faced devastating cuts when markets turned. The structural difference between this and the AI transmutation scenario is that in prior cycles, the underlying wage tax base remained intact. Workers kept earning, consuming, and paying income and sales tax even as capital gains cratered. In a transmutation scenario, the wage base is itself being displaced. There is no floor.

MECH-001 Meets the Tax Code: Recursive Displacement of the Revenue Base

The Theory of Recursive Displacement (MECH-001) describes a self-reinforcing cycle: automation displaces labor, the displaced labor loses economic participation, the reduced participation weakens the institutions that might have absorbed the displaced, and the weakened institutions create conditions for further displacement [Framework — Original]. The fiscal volatility trap is the revenue-side expression of this mechanism.

Here is how the recursion operates in fiscal terms. A firm automates a function previously performed by 100 workers. Those workers generated approximately $6.2 million in annual payroll taxes (employer and employee FICA combined at an average wage of $62,000), plus roughly $1.8 million in state and federal income tax withholding, plus the sales tax revenue generated by their consumption. Call it $9-10 million in total government revenue from those 100 positions.

The firm’s profits increase. It pays corporate income tax on the margin, though at a rate reduced by depreciation of the AI capital expenditure — which under the OBBBA’s 100% bonus depreciation is fully deductible in year one [Measured]^[4]. The firm’s equity appreciates. Some of that appreciation generates capital gains tax when shares are sold, though at the long-term capital gains rate (20% federal maximum versus the 37% top marginal rate on ordinary income), with timing controlled by the seller. The firm’s founders and investors realize gains in jurisdictions of their choosing.

Net result: the government collects less total revenue, collects it from fewer payers, collects it at lower effective rates, and collects it on a schedule controlled by the taxpayer rather than the employer. Every property that made labor-linked revenue reliable — breadth, stability, low optimization surface — has been degraded.

Now apply recursion. The reduced government revenue constrains public investment in education, retraining, and social services — the institutions that might have enabled displaced workers to re-enter the economy. The constrained institutions produce a less-capable workforce, which makes further automation more attractive relative to human labor. The further automation displaces more labor-linked revenue. The cycle tightens. This is MECH-001 operating through the fiscal channel, and it is MECH-014 (The Ratchet) ensuring that each turn of the cycle makes reversal more expensive than continuation [Framework — Original].

The federal tax code is not neutral in this process. It is an accelerant. The OBBBA’s 100% bonus depreciation for qualifying AI and robotics equipment means that a firm spending $10 million on automation deducts the full amount in year one, generating an immediate tax shield worth approximately $2.1 million (at the 21% corporate rate) [Measured]^[4]. There is no corresponding revenue offset for the payroll taxes those automated workers were generating. The government subsidizes the displacement of its own revenue base. This is not an oversight. It is the result of a tax code designed for an economy where capital investment creates jobs. In an economy where capital investment destroys jobs, the same code becomes a fiscal self-destruct mechanism.

The Concentration Trap: MECH-010 and Geographic Revenue Fragility

The transmutation problem is compounded by geographic concentration. MECH-010 (demand collapse through labor share compression) predicts that the economic benefits of automation concentrate among capital owners while the costs distribute across workers and communities [Framework — Original]. In fiscal terms, this means the replacement revenue concentrates in a handful of jurisdictions while the revenue loss distributes broadly.

Consider Virginia’s data center corridor. In 2024, Loudoun County alone hosted more than 300 data centers, making it the largest concentration of digital infrastructure on Earth. Virginia localities offered substantial tax incentives to attract this investment — incentives that made sense when data centers were modest facilities. But the AI buildout has transformed the economics. Statewide data center tax incentives ballooned from approximately $1.5 million per year to an estimated $1.6 billion annually [Measured]^[3], triggering a legislative backlash with over 300 bills introduced across U.S. states to revisit data center tax breaks [Measured]^[3].

The pattern is instructive. Jurisdictions competed to attract AI capital by offering tax incentives. The AI capital arrived, generating economic activity but relatively little labor-linked revenue (data centers employ remarkably few people per dollar of capital invested). The jurisdictions that “won” the competition now host enormous concentrations of capital-intensive, labor-light economic activity — activity that generates property tax on the facilities and some corporate income tax, but none of the payroll tax, income tax withholding, or consumption tax that equivalent-value human economic activity would have generated. Meanwhile, the jurisdictions that “lost” the competition — the places where the human workers used to be — lost both the workers and the economic activity, with no replacement revenue of any kind [Measured]^[10].

This is MECH-024 (fiscal capture) operating through the incentive structure [Framework — Original]. Local governments, starved of revenue by transmutation, compete more aggressively for whatever capital-linked revenue remains. The competition drives them to offer increasingly generous incentives, further eroding the revenue they collect. The capital interests whose automation caused the displacement become the only viable revenue source, giving them leverage to extract further concessions. The government becomes dependent on the entity that is destroying its fiscal viability. This is the fiscal volatility trap: not revenue disappearance, but revenue dependency on a narrow, volatile, politically powerful base.

The Productivity Offset That Isn’t (Yet)

The strongest counterargument to the fiscal volatility trap is the productivity offset. If AI delivers sufficient total factor productivity growth, the expanding economic pie could generate enough revenue — even from capital-linked sources — to offset the loss of labor-linked streams. The American Enterprise Institute has modeled this scenario: an additional 0.5 percentage points of annual TFP growth, broadly captured, could reduce the debt-to-GDP ratio by approximately 12 percentage points over the projection window, with the AEI placing 45-55% confidence on this outcome [Measured]^[5].

This is a serious argument. It deserves serious engagement, not dismissal.

The problem is empirical. As of early 2026, the macro-level evidence for AI-driven productivity gains is thin. The San Francisco Federal Reserve published a careful assessment in February 2026 concluding that there is “limited macro evidence of AI productivity impact” despite enormous capital investment [Measured]^[6]. Firm-level studies show impressive gains in narrow applications — customer service, code generation, document processing — but these have not yet aggregated into measurable TFP growth at the economy-wide level.

There are structural reasons to expect the aggregation to be slow. The firms capturing the largest AI productivity gains are disproportionately in sectors with low labor-to-output ratios: technology, finance, professional services. These sectors contribute to GDP but employ a relatively small share of the workforce. The sectors that employ the most people — healthcare, education, government, retail, hospitality — face structural barriers to AI adoption: regulatory constraints, human-contact requirements, fragmented IT infrastructure, and organizational cultures resistant to workflow redesign. The productivity gains concentrate where the workers aren’t, and the workers concentrate where the productivity gains are slow to materialize.

More importantly, even if the productivity offset materializes, it does not solve the transmutation problem. It changes the magnitude but not the structure. A larger pie sliced the same way still concentrates revenue in volatile, narrow, optimizable streams. The productivity offset could, in principle, generate enough capital-linked revenue to fund government operations — but that revenue would still be subject to the volatility, concentration, and optimization dynamics described above. A government funded primarily by capital gains taxes on AI firms is a government that experiences a fiscal crisis every time the tech sector has a bad year. The productivity offset is a quantity argument. The transmutation trap is a quality argument. They operate on different dimensions.

The Escape Routes and Why They’re Locked

If revenue transmutation is the disease, the obvious prescription is new tax instruments designed to capture revenue from capital-intensive, labor-light economic activity. The policy literature is not short on proposals. Robot taxes, digital services taxes, automation levies, data royalties, AI compute taxes, token taxes — Brookings alone has catalogued over a dozen proposed mechanisms [Measured]^[1]. The NCSL has conducted a comprehensive review of AI’s implications for state tax systems [Measured]^[11]. The intellectual architecture for fiscal adaptation exists.

The political architecture does not.

Robot tax proposals have been introduced in multiple U.S. state legislatures and in the European Parliament. None has been enacted. The legislative history, reviewed comprehensively by Tax Notes, reveals a consistent pattern: proposals attract initial attention, trigger intense industry lobbying, and either die in committee or are amended into ineffectiveness [Measured]^[8]. The objections are predictable — competitiveness concerns, definitional challenges (what counts as a “robot”?), and the argument that taxing automation will slow productivity growth. These objections are not frivolous. But they function as a political veto that keeps the escape routes locked.

The deeper problem is structural. MECH-029 (institutional capture) predicts that the entities with the greatest interest in preventing new tax instruments are the same entities whose economic power is growing fastest [Framework — Original]. AI firms and their investors have both the motive and the means to block fiscal adaptation. They have the motive because new tax instruments would reduce their returns. They have the means because revenue transmutation itself concentrates economic power in their hands, giving them disproportionate political influence through lobbying, campaign contributions, and the credible threat of relocating economic activity to more favorable jurisdictions. The fiscal trap is self-defending.

This does not mean escape is impossible. Sovereign wealth funds seeded by AI-rent capture, as proposed by the Lead-Own-Share framework [Measured]^[7], represent a structurally different approach — one that substitutes ownership for taxation, giving governments a direct stake in AI capital rather than attempting to tax its returns. Data royalties and compute-use fees sidestep the “robot tax” framing entirely. State-level backlash against data center incentives shows that the political calculus can shift when the costs become visible enough [Measured]^[10]. But each of these escape routes requires political action against the interests of the most powerful economic actors of the current era. The trap is contingent on political failure. The question is whether political failure is a risk or a baseline.

The Downturn Scenario: When Latent Fragility Goes Active

Everything described above is latent. It describes a structural condition, not a crisis. The crisis activates under specific trigger conditions: a technology sector downturn, a broader recession, or a capital flight event that causes capital-linked revenue to contract simultaneously with rising demand for public services.

Model the scenario for a state like California. AI-linked capital gains revenue is high and climbing. The state has hired teachers, funded healthcare expansions, and committed to infrastructure projects against projected revenue. A market correction — not a crash, just a 30% decline in tech equity valuations — triggers a capital gains revenue collapse on the order of 50-60%, based on historical patterns. Simultaneously, displaced workers who had been marginally employed in the gig economy lose their footholds. Unemployment insurance claims spike. Medi-Cal enrollment surges. Homelessness services face crisis demand.

The state faces a fiscal scissors: revenue collapsing from the top, spending demands surging from the bottom. The $18 billion deficit exposure the LAO identified [Measured]^[2] is not a worst case. It is a moderate case. A severe downturn, coinciding with an acceleration of displacement, could produce a deficit multiple of that figure.

Now extend the scenario to the 10-15 states with significant labor-income tax dependency. Oregon, where income tax constitutes over 70% of general fund revenue. New York, where financial sector capital gains revenue plays a role analogous to California’s tech revenue. Illinois, where the combination of pension obligations and narrow tax base creates particularly acute vulnerability. Each of these states faces its own version of the transmutation trap, calibrated to its specific economic structure but driven by the same underlying mechanism.

The municipal level is where the pain becomes most concrete. Local governments derive approximately 30% of their general revenue from property taxes — and property values in communities dependent on labor-intensive industries track employment levels with a lag. A community that loses its call center, its warehouse distribution hub, or its back-office processing facility to AI automation does not see property tax revenue decline immediately. Assessments lag reality by two to five years. By the time the assessed values catch up to the economic reality, the commercial properties are vacant, the residential properties have depreciated, and the municipal budget built on prior-year assessments faces a structural gap that cannot be closed through rate increases on a shrinking base. This is the Detroit pattern described in the September 2025 version of this essay, but accelerated: deindustrialization played out over decades, while AI-driven displacement of service-sector and knowledge-worker functions can hollow out a community’s economic base within a single assessment cycle.

The federal level is not immune. Federal individual income taxes and payroll taxes together constitute approximately 75% of federal revenue [Measured]^[1]. Federal revenue is less exposed to capital gains volatility than California’s, but it is directly exposed to the payroll tax erosion that accompanies labor displacement. Social Security and Medicare are funded primarily through payroll taxes. Software that replaces a human worker does not pay FICA [Measured]^[9]. The federal transmutation scenario is slower but potentially more consequential: a gradual erosion of the payroll tax base that undermines the solvency of the two largest federal transfer programs at precisely the moment demographic aging is already straining them.

The Ratchet Completes the Circuit

What distinguishes the fiscal volatility trap from ordinary fiscal procyclicality — the familiar pattern where government revenue falls in recessions and rises in expansions — is the structural irreversibility introduced by AI-driven transmutation.

In ordinary procyclicality, the revenue base recovers when the economy recovers. Workers get rehired, wages resume, payroll taxes resume, consumption resumes, sales taxes resume. The cycle is painful but self-correcting. In the transmutation scenario, the revenue base does not recover because the underlying economic activity has permanently shifted from labor-linked to capital-linked forms. The workers who were generating payroll tax are not rehired when the economy recovers. Their functions are performed by software and systems that generate corporate profit and capital appreciation but not payroll tax. Each recession shakes out another cohort of labor-linked revenue and replaces it with capital-linked revenue. The recovery restores the capital-linked revenue but not the labor-linked revenue it displaced.

This is MECH-014 (The Ratchet) in fiscal form [Framework — Original]. Each downturn ratchets the revenue base further from labor-linked to capital-linked, and each recovery consolidates the new composition rather than reversing it. The ratchet operates because the economic incentives that drive transmutation — lower labor costs, higher margins, accelerated depreciation — do not reverse in recoveries. Firms that automated during the downturn do not re-hire during the recovery. They pocket the margin improvement and invest in further automation.

The political economy follows the same ratchet. During downturns, governments face fiscal crisis and become more dependent on whatever capital-linked revenue remains. They offer incentives to retain and attract AI capital. They defer or abandon proposals for new tax instruments that might discourage investment during a fragile recovery. The capital interests that benefit from transmutation gain political leverage during each downturn, and that leverage is not surrendered during recoveries. The fiscal trap and the political trap reinforce each other, each making the other harder to escape.

This is the mechanism that connects the fiscal volatility trap to the broader Theory of Recursive Displacement. The fiscal channel is not a side effect of labor displacement. It is one of the primary transmission mechanisms through which displacement becomes self-reinforcing. Governments that cannot fund retraining, social insurance, and institutional adaptation become governments that preside over further displacement. The Great Unwinding is not a single event. It is a ratcheting process in which each turn — each downturn, each recovery, each round of automation — locks in a revenue structure that is narrower, more volatile, more concentrated, and more dependent on the interests that are driving the displacement itself.

Counter-Arguments and Limitations

Objection 1: The analysis applies only to U.S. jurisdictions with high labor-income tax dependency, not universally.

This is correct, and it is a feature of the analysis rather than a bug. We have scoped the fiscal volatility trap to U.S. states and municipalities with significant labor-income tax dependency precisely because the mechanism operates through specific fiscal structures. VAT-heavy European systems face different transmutation dynamics (consumption tax is less directly labor-linked than payroll tax). Resource-dependent economies face different dynamics entirely. The claim is not that every government everywhere faces this trap. The claim is that the U.S. fiscal system — where 75% of federal revenue is labor-linked — is structurally vulnerable to transmutation in a way that other systems may not be. Generalization beyond U.S. labor-income-dependent jurisdictions requires separate analysis.

Objection 2: The framing implies active erosion, but the correct framing is latent structural fragility — the trap activates only under specific trigger conditions.

We accept this reframing entirely and have structured the essay around it. Nothing in this analysis describes a present-tense fiscal crisis caused by AI. Yale’s Budget Lab finds no current AI-unemployment link. The WEF projects net job creation through 2030. The fiscal volatility trap is a structural condition that becomes a crisis only IF displacement materializes at projected scale AND a downturn coincides with sufficient capital-revenue concentration. The trigger conditions — recession, tech downturn, capital flight — are specified. This is a forward-looking conditional analysis at 60-70% confidence, not a description of current conditions.

Objection 3: What distinguishes AI-driven transmutation from ordinary fiscal procyclicality?

The critical distinction is structural irreversibility. Ordinary procyclicality is self-correcting: the revenue base recovers when the economy recovers because the underlying employment structure is intact. AI-driven transmutation introduces a ratchet: labor-linked revenue lost during a downturn is not restored during recovery because the jobs themselves have been permanently automated. Each cycle leaves the revenue base more capital-dependent than the last. The distinguishing feature is not the magnitude of the downturn but the failure of the recovery to restore the prior revenue composition. This is empirically testable and we state it as a falsification condition.

Objection 4: The AEI productivity offset — if AI delivers >0.5pp TFP growth broadly captured, transmutation may be net positive.

This is the strongest counterargument and we engage with it at length in the essay. The AEI model is methodologically sound: sufficient productivity growth could generate enough total revenue to offset transmutation losses. We make two responses. First, the empirical evidence as of early 2026 does not support the productivity offset materializing at the required scale [Measured]^[6]. Second, even if the quantity is sufficient, the quality problem remains — revenue funded by volatile capital streams is structurally different from revenue funded by stable wage streams, regardless of magnitude. A government that collects enough capital gains tax to fund its operations still faces a fiscal crisis every time equity markets decline. The productivity offset is necessary but not sufficient to escape the trap.

Objection 5: New tax instruments (digital services taxes, robot taxes, token taxes, data royalties) are viable escape routes — the trap is contingent on political failure.

Correct. The trap is contingent on political failure. We state this explicitly. The policy toolkit for fiscal adaptation exists — Brookings, NCSL, and others have designed workable instruments [Measured]^[1] [Measured]^[11]. The question is whether political systems can enact them against the opposition of the most powerful economic interests of the era. We argue that political failure is the base case given the structural dynamics of MECH-029 (institutional capture), but we acknowledge that this is a political judgment at moderate confidence, not a deterministic claim. If jurisdictions enact effective capital-activity taxes that maintain revenue stability, the trap does not activate. Our falsification conditions reflect this.

Methods

This analysis synthesizes evidence from five source categories: federal fiscal data and tax policy frameworks (Brookings, IRS), state-level fiscal analyses (California LAO via CalMatters, Virginia legislative records via Multistate and Stateline), federal policy and incentive structures (OBBBA provisions, NCSL reviews), productivity and macroeconomic assessments (AEI, SF Fed), and proposed policy countermeasures (Tax Notes legislative reviews, sovereign wealth fund frameworks, robot tax proposals). The original September 2025 essay relied primarily on aggregate Census Bureau and Urban Institute fiscal data. This revision incorporates six months of state-level evidence that has made the transmutation mechanism empirically observable rather than purely theoretical.

Evidence claims are tagged as [Measured] (backed by cited data), [Estimated] (extrapolated from trends), [Projected] (scenario-based), or [Framework — Original] (novel constructs from the Theory of Recursive Displacement). The analysis is scoped to U.S. jurisdictions and is forward-looking conditional: the trap activates IF displacement materializes at projected scale. We do not claim it is currently active.

Falsification Conditions

This essay is wrong if:

U.S. state and federal revenue composition remains stable or shifts toward labor-linked streams through 2030, indicating that transmutation is not occurring at the structural level described. Specifically: if the payroll-tax share of federal revenue does not decline by more than 2 percentage points by 2030, the transmutation mechanism is weaker than claimed.
AI delivers macro-measurable TFP growth exceeding 0.5 percentage points annually by 2028, AND that growth is broadly captured (not concentrated in a small number of firms or sectors), generating sufficient revenue through existing tax structures to offset labor-linked revenue losses. This would validate the AEI productivity offset at the required scale.
At least three U.S. states enact and sustain effective capital-activity taxes (robot taxes, AI compute levies, or equivalent) by 2028, demonstrating that the political failure condition is not the base case and the escape routes are accessible.
Capital gains and corporate income tax revenue demonstrates stability comparable to payroll tax revenue during the next economic downturn, indicating that transmutation does not produce the volatility differential the analysis predicts. Specifically: if capital-linked revenue declines by less than 1.5x the percentage decline in labor-linked revenue during the next recession, the volatility thesis is overstated.
Displaced workers are re-absorbed into labor-linked economic activity at rates comparable to prior technological transitions (within 3-5 years of displacement), indicating that the ratchet mechanism is not operating and the revenue base self-corrects through normal labor market adjustment.

Bottom Line

Confidence: 60-70%. The fiscal volatility trap describes a structural vulnerability that is empirically observable in early indicators (California’s revenue exposure, Virginia’s incentive reckoning, federal depreciation subsidies) but has not yet produced a full-cycle crisis. We hold confidence at the moderate range because: (a) the transmutation mechanism is currently operating but its scale is uncertain, (b) the productivity offset could theoretically compensate but has not empirically materialized, (c) the political failure condition is the base case but not guaranteed, and (d) the full trap activates only under downturn conditions that are probable but not certain within the relevant timeframe. The September 2025 version of this essay was more speculative. This revision is grounded in observable state-level evidence, but the core claim — that transmutation produces structural fiscal fragility, not just cyclical volatility — remains a forward-looking conditional that requires a full economic cycle to test.

Sources

https://www.brookings.edu/articles/future-tax-policy-a-public-finance-framework-for-the-age-of-ai/ — “Future Tax Policy: A Public Finance Framework for the Age of AI”, Brookings Institution. [verified ✓]
https://calmatters.org/economy/technology/2026/01/california-tech-tax-revenue/ — “California’s AI-Linked Tax Revenue Exposure”, CalMatters, January 2026. [verified ✓]
https://www.multistate.us/insider/2026/2/4/states-rethink-data-center-tax-incentives-as-costs-soar — “States Rethink Data Center Tax Incentives as Costs Soar”, Multistate, February 2026. [verified ✓]
https://standardbots.com/blog/obbb-robotics — “OBBBA 100% Bonus Depreciation for Robotics”, StandardBots. [verified ✓]
https://www.aei.org/economics/can-artificial-intelligence-rescue-americas-fiscal-future/ — “Can Artificial Intelligence Rescue America’s Fiscal Future?”, American Enterprise Institute. [verified ✓]
https://www.frbsf.org/research-and-insights/publications/economic-letter/2026/02/ai-moment-possibilities-productivity-policy/ — “The AI Moment: Possibilities for Productivity and Policy”, Federal Reserve Bank of San Francisco, February 2026. [verified ✓]
https://www.convergenceanalysis.org/fellowships/economics/lead-own-share-sovereign-wealth-funds-for-transformative-ai — “Lead-Own-Share: Sovereign Wealth Funds for Transformative AI”, Convergence Analysis. [verified ✓]
https://www.taxnotes.com/featured-analysis/robot-tax-proposals-legislative-review/2025/11/20/7t92q — “Robot Tax Proposals: A Legislative Review”, Tax Notes, November 2025. [verified ✓]
https://ubos.tech/news/ai-workforce-displacement-risks-tax-base-human%E2%80%91in%E2%80%91the%E2%80%91loop-strategy/ — “AI Workforce Displacement Risks Tax Base”, UBOS, 2026. [verified ✓]
https://stateline.org/2026/02/24/data-center-tax-breaks-are-on-the-chopping-block-in-some-states/ — “Data Center Tax Breaks Are on the Chopping Block in Some States”, Stateline, February 2026. [verified ✓]
https://www.ncsl.org/fiscal/of-returns-to-robots-opportunities-risks-and-policy-implications-of-artificial-intelligence-in-tax-systems — “Of Returns to Robots: Opportunities, Risks, and Policy Implications of AI in Tax Systems”, NCSL. [verified ✓]

Where This Connects

This essay describes the revenue-side mechanism that completes a circuit running through several prior Recursive Institute analyses:

The Ratchet — The fiscal volatility trap is MECH-014 operating through the tax code. Each downturn ratchets revenue composition from labor-linked to capital-linked, and each recovery consolidates the shift rather than reversing it. The Ratchet essay describes the capex version of this irreversibility; this essay describes the fiscal version.
Compute Feudalism — The geographic concentration of AI capital that Compute Feudalism documents is the same concentration that creates fiscal vulnerability. Jurisdictions competing for data centers and AI infrastructure are competing for capital-linked revenue that substitutes for (rather than supplements) labor-linked revenue.
The Aggregate Demand Crisis — The demand-side and revenue-side mechanisms are two faces of the same coin. Displacement erodes wages (reducing consumer demand) and transmutes the tax base (reducing government capacity to fund transfers). The fiscal volatility trap explains why governments cannot simply compensate for the demand crisis through redistribution: the revenue to fund redistribution is itself being undermined.
From Income Floors to Ownership Stakes — The L.A.C. policy essay proposes sovereign wealth funds and ownership stakes as alternatives to pure transfers. The fiscal volatility trap provides the revenue-side argument for why pure transfers funded by traditional tax instruments may be structurally unsustainable: the tax base that funds them is transmuting into volatile, optimizable streams.
The Triage Loop — The Triage Loop describes algorithmic governance as a response to social instability. The fiscal volatility trap identifies one of the mechanisms that produces the instability: governments that cannot fund services because their revenue base has transmuted face the choice between austerity (which accelerates displacement) and algorithmic triage (which substitutes control for support). The fiscal channel feeds the governance channel.