I’ve spent an inordinate amount of time prognosticating on the economic and societal consequences of artificial intelligence. Much of that contemplation has taken place during long, solitary hours on the bicycle, where an uninterrupted stretch of road provides ample opportunity to disappear into a subject and pursue it to its more uncomfortable conclusions. Away from the bike, I’ve found myself reading everything from detailed histories of the Industrial Revolution to contemporary employment reports, occupational classifications, wage distributions, and economic forecasts, all in an attempt to understand how a society transitions when the technology responsible for its prosperity begins displacing the very people who helped create it.
The Industrial Revolution has been a particular fascination, largely because it offers an imperfect but extraordinarily instructive account of what happens when productivity advances faster than the institutions responsible for accommodating it. We tend to remember the eventual prosperity, the proliferation of new industries, and the remarkable improvement in living standards. It’s considerably easier to overlook the generations of people who endured the transition, the communities whose economic foundations deteriorated, and the rather protracted interval between technological advancement and broadly shared prosperity.
I’ve also long subscribed to the principle that all ships rise and fall with the tide. A broadly prosperous society benefits nearly everyone, including those whose individual circumstances might otherwise insulate them from economic hardship. Employment supports consumption, consumption sustains businesses, businesses generate tax revenue, and that revenue finances the institutions upon which much of society depends. The relationship is imperfect, naturally, and prosperity has never been distributed with anything approaching mathematical uniformity. Nevertheless, the underlying interdependence is difficult to dispute. When a sufficiently large portion of society begins struggling economically, the consequences eventually reach those who initially appear immune.
Artificial intelligence has the potential to disrupt that relationship in a particularly unusual way. Unlike the mechanization of agriculture or the automation of industrial manufacturing, AI is progressing rapidly into occupations that have traditionally benefited from higher education, specialized knowledge, and substantial professional compensation. Computing, finance, law, management, accounting, engineering, marketing, and countless other disciplines now find portions of their expertise increasingly reproducible through software. The economic implications extend considerably beyond the individuals employed in those professions, reaching into household consumption, higher education, municipal finances, government tax receipts, and the long-term sustainability of social benefits.
I consider myself relatively fortunate to be able to examine much of this from a position of some economic insulation, although that privilege hardly diminishes my concern. The relationship between individual prosperity and collective well-being is precisely why the broader implications warrant consideration, especially from those who might otherwise be inclined to dismiss them.
There is also something rather extraordinary about the technological opportunity itself. AI could substantially accelerate scientific discovery, improve medical care, democratize access to sophisticated intellectual capabilities, and allow individuals to accomplish things that previously required entire organizations. I have little interest in preserving inefficient processes merely because human beings have historically performed them. But enthusiasm for technological advancement should be accompanied by an equally serious examination of its externalities, particularly when the potential beneficiaries and those bearing the costs may be very different populations.
I firmly believe that the more knowledgeable and prepared we are for this transition, the better the outcome will be for both humanity and technological advancement. An informed society has considerably more opportunity to adapt than one that oscillates between uncritical enthusiasm and reactionary opposition, neither of which strikes me as a particularly useful approach to something this consequential.
A recent Bloomberg graphic, comparing occupational exposure to AI using research from Anthropic and OpenAI, finally provided a quantitative starting point for a question I’d been considering for some time. The occupations appearing most exposed weren’t confined to low-wage or traditionally repetitive work. They were concentrated in several of the professional and administrative categories that contribute substantially to American household income, tax receipts, and the economic justification for higher education.
I wanted to expand upon that analysis by incorporating employment figures, median salaries, the taxes those salaries generate, and the economic consequences of potential displacement. The recent midterm elections, increasingly contentious political rhetoric surrounding AI, and the emerging international competition for computing infrastructure introduce an additional and rather fascinating complication.
What follows is an analytical exercise developed with substantial research, calculations, and editorial assistance from OpenAI. The underlying employment and fiscal figures come from published sources, while the displacement scenarios and longer-range forecasts are explicitly hypothetical. My objective is to explore the mechanisms, their potential consequences, and the political decisions likely to influence the outcome, rather than pretend that anyone can accurately predict the labor market of 2040.
The professional class, and an uncomfortable inversion of technological risk
Before interpreting the Bloomberg graphic, an important distinction needs to be established. Occupational exposure to AI measures the extent to which a profession contains tasks susceptible to AI assistance or automation. It does not provide a direct probability that a worker will lose their job, nor does it establish that the entire occupation will eventually become redundant. Anthropic’s March 2026 research combines theoretical AI capability with observed workplace usage, placing additional weight on applications that automate rather than simply assist with work. The accompanying OpenAI-derived index examines a broader measure of technical task susceptibility. These methodologies differ, and their results should be understood accordingly.
That qualification is consequential because technological capability and commercial adoption are separated by a great many considerations. Reliability, regulation, liability, organizational inertia, customer expectations, and the cost of implementation all influence whether a technically feasible application becomes an economically practical substitute for human labor. An attorney may use AI to research precedent without relinquishing professional responsibility for the resulting advice. An engineer may automate substantial portions of a design process while remaining accountable for its physical consequences. A financial analyst may produce considerably more research without the employer necessarily requiring fewer analysts.
Nevertheless, the gap between observed and theoretical exposure is difficult to ignore. Bloomberg’s graphic places observed exposure in computer and mathematical occupations near 35%, while theoretical exposure exceeds 90%. Management shows an even more pronounced disparity, with observed exposure around 13% against theoretical exposure near 91%. The extent to which those gaps eventually narrow will depend on considerably more than the underlying models becoming capable of performing additional tasks. But the magnitude of the theoretical opportunity provides an indication of how extensively established workflows could change.
AI exposure by occupational group
Observed and theoretical share of tasks · 22 BLS groups · readings from the Bloomberg chart, 2026
Observed use trails theoretical exposure in every group. The widest gaps are architecture and engineering at 82 points, management at 78, and life and social science at 68; the narrowest is building and grounds maintenance at 4.5, and the comparison is directional.
Sources: Bloomberg’s Anthropic and OpenAI exposure comparison; BLS 2025 occupational employment matrix. Percentages are approximate readings of the Bloomberg chart. Hover or focus a row for detail.
To place those observations in an economic context, I matched the occupational categories in the graphic with the Bureau of Labor Statistics’ 2025 employment and median-wage estimates, published in August 2026. I also modeled a representative worker’s approximate federal labor-tax contribution at each occupational median, using 2026 income-tax provisions and both employer and employee Social Security and Medicare contributions.
| Occupational group | AI exposure observed / theoretical | Employment | Median annual salary | Illustrative annual federal labor taxes |
|---|---|---|---|---|
| Computer & mathematical | 35% / 93% | 5.49M | $109,280 | $31,931 |
| Office & administrative | 34% / 88% | 18.79M | $47,450 | $10,774 |
| Business & financial | 29% / 94% | 11.37M | $82,660 | $22,002 |
| Sales | 27% / 67% | 14.53M | $38,530 | $8,339 |
| Legal | 20% / 85% | 1.45M | $102,500 | $29,402 |
| Arts & media | 18% / 78% | 2.99M | $62,750 | $14,951 |
| Education & library | 18% / 62% | 9.91M | $60,570 | $14,356 |
| Management | 13% / 91% | 13.68M | $126,520 | $38,456 |
| Life & social science | 10% / 78% | 1.63M | $82,530 | $21,954 |
| Healthcare practitioners | 5% / 61% | 10.23M | $86,530 | $23,446 |
| Architecture & engineering | 4% / 86% | 2.69M | $99,520 | $28,291 |
| Community & social services | 4% / 54% | 3.27M | $58,300 | $13,736 |
| Protective services | 2.5% / 32% | 3.87M | $50,080 | $11,492 |
| Healthcare support | 2% / 31% | 8.42M | $38,340 | $8,287 |
| Personal care | 2% / 25% | 4.49M | $36,410 | $7,760 |
| Installation & repair | 1.5% / 25% | 6.49M | $59,620 | $14,096 |
| Agriculture & forestry | 1% / 20% | 1.04M | $36,630 | $7,820 |
| Construction | 1% / 22% | 7.49M | $59,540 | $14,074 |
| Food & serving | 0.8% / 22% | 13.83M | $35,050 | $7,389 |
| Production | 0.5% / 24% | 8.78M | $46,990 | $10,648 |
| Building & grounds maintenance | 0.5% / 5% | 5.70M | $37,680 | $8,107 |
| Transportation | 0% / 23% | 14.14M | $44,350 | $9,928 |
The concentration of employment is remarkable. Computer and mathematical occupations account for approximately 5.5 million positions, with median annual compensation exceeding $109,000. Business and financial operations account for another 11.4 million, management encompasses nearly 13.7 million, and office administration represents almost 18.8 million. Taken together, these four categories alone comprise approximately 49.3 million positions, or roughly 29% of the employment represented in the BLS matrix.
The distinction between an occupation and an industry is important here. A financial institution employs accountants, software developers, administrators, managers, analysts, and numerous other professionals whose exposure levels differ considerably. Likewise, an occupation’s median salary cannot be treated as the average compensation of everyone employed within it. The table is intended to illustrate the scale and composition of the exposed workforce, rather than produce a literal count of jobs destined to disappear.
Still, the pattern suggests a rather interesting inversion of the relationship between education and technological vulnerability. For generations, specialized knowledge has provided a degree of protection against mechanization. As software begins replicating portions of that knowledge, occupations demanding extensive formal education can exhibit greater theoretical exposure than jobs requiring physical dexterity, environmental adaptability, or direct interpersonal care. A plumber working beneath a century-old house may be considerably more difficult to replace with autonomous technology than an analyst reviewing standardized financial statements.
Of course, technological exposure doesn’t establish relative economic value. The analyst may continue generating substantially more revenue, and the plumber may face entirely different competitive pressures. The more pertinent consideration is whether certain capabilities that have historically commanded a professional wage premium become increasingly abundant through automation. Compensation reflects numerous factors, but scarcity remains one of the more consequential.
If the scarcity of specialized intellectual production diminishes, it would be surprising if the market value of some of that production remained entirely unaffected.
The professional career ladder may be the first casualty
One of the more insidious consequences of AI may emerge through a mechanism that produces relatively little immediate evidence of unemployment: the gradual disappearance of entry-level recruitment.
Most professional careers depend on a lengthy developmental process. Junior attorneys research cases and draft documents, analysts assemble financial models, accountants reconcile records, and beginning software engineers write and test code. These functions produce economic value for employers, but they also provide the experience through which junior employees acquire the judgment necessary to perform more sophisticated work. Much of the early-career labor that organizations regard as relatively routine is indispensable to the development of their eventual senior workforce.
AI happens to be particularly effective at accelerating many of those introductory tasks.
Consider an accounting firm employing one hundred people. As AI capabilities improve, the firm’s experienced professionals discover they can manage considerably more work without expanding their teams. The organization doesn’t necessarily dismiss existing employees, particularly those with established relationships and institutional knowledge. It simply reduces recruitment, allows natural attrition to diminish headcount, and becomes increasingly selective about filling vacancies.
Financial performance may actually improve. Clients continue receiving services, senior professionals retain their positions, and the business becomes more productive. Yet the number of younger accountants acquiring the experience necessary to replace those senior professionals declines. Extend that process across accounting, legal services, finance, consulting, software development, insurance, and other professional disciplines, and a relatively inconspicuous change in corporate hiring begins developing into a structural labor-market problem.
The early empirical evidence is becoming difficult to dismiss. A Stanford Digital Economy Lab study, revised in August 2026, examined payroll records covering millions of American workers. The researchers found no evidence of widespread economy-wide displacement, but employment among 22-to-25-year-olds in highly AI-exposed occupations was approximately 19% below the level implied by trends among their less-exposed peers. Experienced workers showed no comparable divergence.
That figure represents a relative employment gap, rather than evidence that 19% of existing positions had been eliminated. The researchers also emphasize that the study cannot definitively attribute the entire divergence to AI. Nevertheless, much of the difference appears to have developed through reduced hiring rather than increased termination.
A September 2026 Census Bureau working paper reported similarly concerning findings. Graduates from the most AI-exposed decile of college majors experienced a five-percentage-point reduction in their adjusted probability of initial employment and 13% lower initial earnings. Approximately half the earnings deterioration was associated with lower compensation within employing industries, while the remainder reflected graduates entering less remunerative sectors.
There are undoubtedly other economic forces influencing these outcomes, including interest rates, changes in business investment, and broader hiring cycles. Still, the possibility that AI is beginning to restrict the professional employment pipeline deserves considerable attention, particularly because the consequences could accumulate for years before becoming readily apparent in conventional unemployment statistics.
An economy that produces an abundance of highly educated graduates while progressively reducing their opportunities to acquire meaningful professional experience is creating a peculiar contradiction. Universities continue preparing students for occupations whose entry-level functions may be undergoing substantial technological substitution, while employers increasingly depend on experienced professionals whose eventual successors are struggling to enter the field.
That contradiction also raises uncomfortable questions about the economics of higher education. American families have spent generations accepting the proposition that obtaining a degree, often at considerable expense, provides a dependable path toward greater lifetime earnings. If the scarcity premium associated with certain credentials deteriorates, students and parents will reasonably begin reconsidering the return on that investment. The implications extend into tuition revenue, student borrowing, household formation, and the financial viability of institutions whose economic justification is closely tied to graduate employment outcomes.
I suspect this deterioration in the professional succession process is among the more plausible near-term consequences of AI, and potentially one of the least appreciated.
When productivity and employment begin moving in different directions
The American economy depends on an elaborate network of relationships between household income, consumption, business investment, and government revenue. For most working households, employment provides the principal means of participating in that network. Wages finance housing, food, transportation, healthcare, education, discretionary spending, and retirement savings, while also generating substantial federal and state tax receipts.
The federal government’s dependence on labor-related revenue is particularly relevant. According to the Congressional Budget Office’s February 2026 baseline, individual income taxes were projected to contribute approximately $2.8 trillion in fiscal 2026, with payroll taxes contributing another $1.8 trillion. Corporate income taxes, by comparison, were projected at $404 billion, customs duties at $418 billion, and other revenue sources at approximately $197 billion. Individual income and payroll taxes therefore account for roughly 82% of projected federal receipts.
Not every dollar of individual income tax originates from wages, and payroll-tax revenue includes income earned outside conventional salaried employment. Nevertheless, the importance of labor-derived income to federal finances is unmistakable.
Where federal revenue comes from
Per 100 dollars of projected receipts · fiscal 2026 · CBO February 2026 baseline
Individual income and payroll taxes supply 82 of every 100 dollars of projected receipts. Corporate income tax supplies 7.
Figures as cited in the article: $2.8 trillion individual income tax, $1.8 trillion payroll, $404 billion corporate, $418 billion customs, about $197 billion other. Ticks are rounded so that the ring sums to 100.
Consider the tax implications of a professional position earning $125,000 annually. The employee pays federal income tax and their portion of Social Security and Medicare contributions, while the employer generally contributes corresponding payroll taxes. The employee subsequently distributes much of the remaining income through expenditures that support other businesses, their employees, and additional tax receipts.
Now imagine software allows the employer to produce a comparable quantity of work without maintaining that position. The underlying economic value doesn’t necessarily disappear. It may emerge as higher corporate profitability, lower prices, additional output, or greater returns on invested capital. But the mechanism through which that value becomes taxable income has changed.
Artificial intelligence doesn’t independently contribute to Social Security, and an increase in corporate profitability isn’t automatically taxed at the same effective rate as the wages it replaces. Capital appreciation may remain unrealized for years, corporate profits can be distributed through various arrangements, and international structures complicate the allocation of taxable income between jurisdictions.
Consequently, an economy could become substantially more productive while experiencing pressure on certain tax receipts. Whether total government revenue increases or decreases depends on the magnitude of the productivity gains, how they are distributed, the creation of new employment, and the tax treatment of the resulting income.
This is an important qualification because AI adoption does not inherently reduce government revenue. A sufficiently large expansion of GDP could more than compensate for lost wage-related receipts. The fiscal concern arises when the distribution of economic value shifts away from the mechanisms upon which existing tax systems depend.
A hypothetical fiscal stress test
To better understand the potential magnitude, I constructed a simplified model combining the Bloomberg exposure figures with BLS employment, median salaries, and illustrative federal labor-tax contributions.
Applying each occupational category’s observed-exposure percentage to its employment count produces approximately 21.6 million exposure-weighted worker equivalents. This is a synthetic index of exposed employment, rather than an estimate of jobs likely to disappear. To explore possible outcomes, I then assumed that varying proportions of this exposure eventually translated into reduced demand for human labor through layoffs, attrition, declining recruitment, or reductions in working hours.
Where the 21.6 million sits
Exposure-weighted worker equivalents · observed exposure × BLS 2025 employment · directional
Office and administrative, sales, and business and financial operations together account for 13.6 million of the 21.6 million equivalents, or 63%. The index is synthetic, as the article says, and carries no forecast of jobs lost.
Derived from the table above: employment multiplied by observed exposure, summed to 21.55 million.
| Illustrative 2035 scenario | Exposure converted into reduced labor demand | Full-time job equivalents | Annual wages displaced | Gross annual federal labor-tax shortfall |
|---|---|---|---|---|
| Limited disruption | 10% | 2.2 million | $144 billion | $37 billion |
| Substantial disruption | 25% | 5.4 million | $360 billion | $93 billion |
| Severe disruption | 50% | 10.8 million | $721 billion | $187 billion |
Tax lost against tax regained
Illustrative 2035 scenarios · annual · 2026 dollars · federal only
If every displaced wage dollar reappeared as taxable domestic profit at the 21% statutory rate, the corporate-tax gain would recover about four-fifths of the labor-tax shortfall in each scenario. That transfer is deliberately clean, and the article treats it as an upper bound.
Derived from the scenario table above. The corporate column applies 21% to wages displaced; it excludes reemployment, new occupations and secondary effects. Hypothetical, not a forecast.
The substantial-disruption scenario provides a useful illustration. If the equivalent of approximately 5.4 million positions were displaced, the model implies roughly $360 billion in annual lost wage income and a gross federal labor-tax shortfall of $93 billion. That represents approximately 1.7% of the government’s projected fiscal 2026 revenue, a meaningful amount but hardly sufficient, in isolation, to establish an impending fiscal catastrophe.
The eventual net loss could be much smaller. Displaced employees may find other work, launch businesses, or develop capabilities that complement AI. Greater productivity could lower prices, increase consumption, expand profits, and create new taxable economic activity. Conversely, additional unemployment benefits, reduced local spending, and secondary employment losses could deepen the fiscal consequences.
The model is particularly sensitive to those adjustments, which is why treating gross displacement calculations as predictions would be intellectually indefensible.
For comparison, suppose the entire $360 billion in displaced compensation became additional taxable domestic corporate profit. Applying the statutory 21% federal corporate-tax rate would generate approximately $76 billion, somewhat less than the modeled $93 billion reduction in labor-related receipts. That illustration assumes a rather implausibly clean transfer of economic value from wages to taxable profits, but it demonstrates that replacing employment with corporate earnings doesn’t necessarily preserve the associated federal revenue.
RAND examined this issue directly in August 2026, concluding that large-scale labor displacement could materially undermine federal finances, especially when highly compensated workers are affected. The researchers’ scenarios depend heavily on reemployment and the distribution of AI-generated profits, with corporate income taxes potentially insufficient to offset lost labor revenue.
There is, fortunately, a credible counterargument. Yale’s Budget Lab estimated that a rapid AI-driven growth scenario could increase annual federal receipts by as much as $246 billion by 2030 relative to its baseline. Interestingly, the researchers found that the gains could be nearly twice as large if labor retained its previous share of national income, illustrating how the distribution of productivity influences the government’s ability to collect revenue.
Both studies deserve attention because they examine different aspects of the same economic transformation. AI could generate sufficient economic growth to increase aggregate tax receipts while simultaneously reducing the proportion of that growth received by workers.
The distinction becomes especially consequential when examining the middle and upper-middle classes.
A household earning $200,000 annually might appear relatively insulated from economic insecurity, but that income often supports a substantial collection of fixed obligations. Mortgage payments, childcare, education expenses, health insurance, debt service, and retirement contributions can absorb a considerable portion of the household budget. If one professional loses employment or accepts a substantially lower salary, discretionary expenditures can deteriorate rapidly.
The resulting consequences extend into the surrounding economy. Restaurants lose customers, contractors experience fewer projects, retailers face weaker demand, and local governments collect less revenue from transactions and potentially declining property values. Businesses providing services to comparatively affluent households may find themselves affected by technological disruption despite possessing little direct exposure to AI.
There is also a distributional complication involving asset ownership. Professionals receiving income primarily through salaries may capture a smaller share of AI’s economic benefits than households holding substantial equity in the companies, infrastructure, or intellectual property responsible for the resulting productivity gains.
That divergence could produce an economy whose aggregate wealth increases while the financial circumstances of substantial portions of its professional workforce deteriorate.
And because public services depend on tax receipts, the consequences eventually extend into Social Security, Medicare, education, municipal operations, and the social programs intended to assist economically vulnerable households.
Social Security and Medicare Part A rely heavily on payroll taxation. A material decline in covered wages would place additional pressure on dedicated financing sources, while existing retirement and healthcare obligations would remain largely intact. Unemployment insurance and means-tested programs could experience greater demand, depending on the duration of displacement and the eligibility of affected households.
These systems are already confronting substantial fiscal pressure associated with demographic change and federal indebtedness. The CBO’s February 2026 baseline projected a federal deficit of approximately $1.9 trillion for fiscal 2026, with publicly held debt approaching 120% of GDP by 2036.
AI could eventually improve that outlook through greater productivity and economic growth. It could also aggravate the existing imbalance if the productivity dividend becomes increasingly disconnected from taxable labor income.
The question is how effectively the fiscal system will capture the economic value being created, particularly if conventional employment becomes a progressively smaller component of production.
The increasingly contentious economics of data centers
An examination of AI’s economic consequences would be incomplete without addressing the infrastructure responsible for making it possible. Data centers have become an extraordinarily convenient political target, particularly as concerns about electricity prices, water consumption, and potential employment displacement have converged into a broader narrative about technology companies prospering at the public’s expense.
Some of those concerns are entirely legitimate. Others rely on distorted statistics, misleading comparisons, and a rather casual disregard for the economic realities of modern computing.
It is worth remembering that data centers support considerably more than generative AI. Cloud computing, financial transactions, streaming services, medical records, electronic commerce, communications, scientific research, and the ordinary operations of millions of businesses depend on the same broad category of infrastructure. AI is driving a substantial increase in demand, particularly for facilities equipped to accommodate high-performance accelerators, but the notion that data centers exist exclusively to support chatbots is plainly incorrect.
The electricity requirements, however, are substantial and growing.
According to Lawrence Berkeley National Laboratory’s June 2026 report, data centers could account for approximately 11.8% of American electricity consumption by 2030, with modeled outcomes ranging from 9.5% to 15.3%. That represents a significant increase in demand for generation, transmission, and supporting electrical infrastructure. It also makes clear why utilities and regulators need to scrutinize proposed projects with considerably greater sophistication than has sometimes characterized the public debate.
A large computing facility may require dedicated substations, transmission upgrades, additional generating capacity, and long-term commitments from electricity providers. When those investments are made primarily to serve a prospective industrial customer, the allocation of their costs becomes critically important. If a data-center developer abandons a project or consumes substantially less power than anticipated, a poorly constructed utility agreement could leave existing customers bearing expenses associated with infrastructure that was never fully utilized.
That is a genuine economic externality, and residents are entirely justified in objecting to arrangements that transfer private development costs onto public utility bills.
The appropriate response is to establish transparent pricing, meaningful financial guarantees, and contractual obligations that allocate infrastructure expenses to the parties responsible for creating them. Oregon, for example, has already developed rate structures specifically intended to protect existing utility customers from cost shifting associated with very large electricity consumers. The Federal Energy Regulatory Commission has similarly initiated proceedings concerning the integration of large loads and the protection of ratepayers.
There is no compelling reason that encouraging AI infrastructure development should require residents to subsidize speculative utility investments.
Water consumption presents another legitimate but frequently misrepresented concern. Computing equipment generates heat, and large facilities require sophisticated cooling systems to maintain operational reliability. Some cooling systems consume water through evaporation, while others rely more heavily on air cooling, closed-loop liquid systems, or hybrid architectures. The environmental consequences depend on facility design, climate, local water availability, and electricity generation.
Broad assertions assigning a fixed quantity of water to every AI query are consequently of limited analytical value. They frequently obscure differences between water withdrawals and actual consumption, while neglecting the considerable variation between facilities and operating conditions.
A data center constructed in an arid region facing chronic water scarcity deserves a different level of scrutiny from one operating in a location with abundant water resources and an appropriately designed cooling system. Local conditions matter, as does the distinction between direct consumption and water used elsewhere to generate electricity.
The objective should be rigorous environmental disclosure and infrastructure planning appropriate to the location. Blanket assurances that every facility is environmentally benign are no more credible than claims that every data center represents an ecological catastrophe.
There is also a curious misconception concerning employment. Data-center construction can support substantial demand for electricians, engineers, construction crews, equipment manufacturers, transportation companies, and other skilled trades. Permanent on-site staffing is considerably smaller once a facility becomes operational, although maintenance, security, engineering, logistics, and supporting services continue generating economic activity.
When developers announce that a project will support thousands of jobs, those figures often include temporary construction work, indirect employment, and modeled economic spillovers. Such estimates should be examined carefully rather than accepted as equivalent to permanent on-site positions.
Nevertheless, evaluating a data center exclusively according to the number of employees physically working inside it misses much of its economic purpose.
A semiconductor fabrication plant, electrical generating station, or highly automated industrial facility may create economic value far beyond its direct employment. Computing infrastructure similarly enables other businesses to provide services, develop products, conduct research, and increase productivity.
The economic value of a data center therefore includes the productive activity it supports elsewhere, although that broader value should never be used to excuse poor infrastructure agreements or unwarranted public subsidies.
And this brings us to a consideration largely absent from the domestic political argument: the companies financing these facilities are participating in a global economy, and other countries are competing aggressively for the same investment.
America can refuse the infrastructure. The rest of the world is under no obligation to follow.
The November 7 episode of the All-In podcast included a discussion that I found particularly pertinent to the present political environment. Jason Calacanis raised the possibility that increasingly restrictive American data-center policies could encourage companies to move future computing infrastructure into Canada, Mexico, Southeast Asia, or the Middle East, allowing the United States to consume AI services generated by facilities located elsewhere.
His broader contention was that America could eventually become a net importer of computing services rather than a net exporter, surrendering associated infrastructure investment and construction employment along the way.
There is an element of speculation in that forecast, particularly because international computing markets are considerably more complicated than an ordinary trade in physical commodities. But the underlying economic logic is compelling.
Capital is mobile. Companies making enormous infrastructure investments evaluate electricity availability, construction costs, regulatory certainty, tax treatment, political stability, telecommunications connectivity, and proximity to customers. If one jurisdiction becomes prohibitively expensive or administratively unpredictable, alternative locations become increasingly attractive.
Modern telecommunications networks also allow many computing workloads to be performed far from the individuals consuming their output. The physical location of a server is not necessarily the location of the customer purchasing the service.
That geographical flexibility has limits. Latency-sensitive applications benefit from infrastructure near their users, some regulated industries require domestic data residency, and national-security considerations restrict where certain models, data, and advanced chips can be deployed. Training enormous models also requires specialized infrastructure and access to substantial concentrations of high-performance hardware.
Still, a meaningful portion of computational work can be performed internationally.
A company developing AI services for American customers may therefore have considerable flexibility in deciding where some of the underlying computation takes place.
This introduces an uncomfortable possibility for policymakers proposing broad restrictions on domestic data-center construction. Even if new American facilities become difficult to permit, the global demand for AI doesn’t simply disappear. Investment can migrate toward countries with more accommodating regulatory environments, sufficient electricity, reliable connectivity, and governments eager to attract it.
The United States might then consume some of the same AI services while forgoing a portion of the domestic construction activity, electrical infrastructure investment, local tax receipts, and supporting industrial development associated with producing them.
The distinction between producing and consuming computing capacity deserves consideration because the economic benefits of AI extend across an increasingly substantial supply chain. Facilities require land, construction, electrical equipment, cooling systems, networking infrastructure, specialized chips, maintenance, and reliable energy supplies. Over time, concentrations of such infrastructure can support additional technical expertise, related businesses, and regional investment.
Countries competing to accommodate those facilities understand that the opportunity extends beyond operating a warehouse filled with expensive computers.
And there is a particularly striking difference in how populations around the world perceive that opportunity.
A rather extraordinary disparity in public optimism
In September 2026, Gallup published results from an international survey examining public attitudes toward artificial intelligence.
Among respondents who were aware of AI, 93% in China believed the technology would mostly benefit their country. In the United States, that figure was just 36%.
Vietnam recorded 91%, Singapore 77%, and Nigeria 68%. American respondents were among the least optimistic of the 37 countries surveyed at that stage of Gallup’s international research.
Expected benefit of AI to the respondent’s own country
Share of AI-aware respondents saying AI will mostly benefit their country · countries as cited in the article · Gallup, September 2026
Of the five countries cited, the United States is the only one below 50%, and it sits 57 points beneath China.
Figures as cited in the article and checked against published coverage of the Gallup release. The full ranking also includes Israel at 69%, which the article does not cite and which is omitted here.
Gallup also found that approximately 93% of AI-aware Chinese respondents expected the technology to improve their daily lives, reinforcing the substantial difference in expectations.
Those findings deserve interpretation rather than indiscriminate extrapolation. The survey covered respondents familiar with AI, and international comparisons are inevitably affected by cultural differences, political environments, and the willingness of participants to express particular views. Public enthusiasm also shouldn’t be confused with evidence that an economy has successfully adopted the technology or that its government has established sound policies.
Nevertheless, the disparity is sufficiently pronounced to warrant attention.
The Stanford 2026 AI Index provides additional context, reporting particularly strong AI optimism and workplace adoption across several emerging economies, including China, India, Nigeria, and parts of Southeast Asia. Meanwhile, nearly two-thirds of Americans surveyed expected AI to result in fewer jobs over the next twenty years.
It would be simplistic to conclude that Americans are merely irrationally frightened of progress while other populations possess some superior understanding of technological economics. American skepticism is influenced by legitimate concerns about employment, institutional trust, affordability, and the distribution of economic gains. Other societies have different economic circumstances and expectations, and their optimism is hardly a guarantee that their transitions will be painless.
But the practical consequences of these attitudes could be substantial.
Governments whose populations generally regard AI as an instrument of national advancement may find it politically easier to facilitate infrastructure development, expand electricity generation, encourage adoption, and attract international investment. Governments responding to widespread skepticism may encounter considerably greater resistance, especially when the immediate local costs are visible and the potential long-term benefits are diffuse.
This creates an asymmetry in which domestic political opposition can constrain an American infrastructure project even as competing countries enthusiastically pursue the same investment.
And the competition is already materializing.
On November 4, 2026, Google announced plans to invest $15 billion in Malaysian digital infrastructure over 2027 and 2028, including data-center campuses, electrical infrastructure, water-resilience initiatives, and workforce-development programs.
The company’s announcement estimated that construction would support more than 37,000 Malaysian jobs annually across direct, indirect, and induced employment, including approximately 6,000 construction jobs. Google also projected more than 9,000 jobs supported annually after the facilities become operational, although those estimates include broader economic activity rather than exclusively permanent on-site personnel.
These are company-sponsored forecasts and should be evaluated accordingly. Nevertheless, the investment commitment itself is substantial, and it represents precisely the sort of opportunity that other jurisdictions are working to attract.
Malaysia’s expanding data-center industry is driven partly by growing regional demand, and Google’s announcement should not be interpreted as evidence that $15 billion has been diverted from American projects. International expansion can complement domestic investment, particularly when companies are serving customers in multiple regions.
The example does, however, establish that the alternative destinations contemplated in the All-In discussion are quite real.
For American policymakers, that should complicate the assumption that denying permission to build domestic infrastructure necessarily diminishes the underlying technological activity.
The economic consequences of importing compute
The All-In discussion used the prospect of America becoming a net importer of AI-generated tokens as a shorthand for the loss of domestic computing capacity.
A token is a unit of information processed by a language model, and importing tokens isn’t presently a conventional national trade statistic. Nevertheless, the phrase conveys a recognizable economic relationship: American businesses and consumers could purchase computing services that depend increasingly on facilities operating abroad.
The consequences would depend on who owns those facilities and the businesses providing the services.
An American company operating a data center in Malaysia could still retain substantial intellectual property, managerial expertise, and corporate profits within its American parent organization. Some of those earnings could remain subject to American taxation, depending on the relevant tax rules and structures.
It would therefore be incorrect to assume that every dollar invested overseas represents an equivalent reduction in American economic activity or that all resulting tax revenue would accrue exclusively to the host country.
The more defensible concern involves the geographical distribution of construction employment, electricity generation, infrastructure investment, operational expenditure, and local economic development. Those activities occur where the facilities are built, even when the owners or customers reside elsewhere.
There are also strategic considerations. Access to advanced computing is increasingly relevant to scientific research, industrial production, military capabilities, cybersecurity, and the development of sophisticated commercial technology.
A country substantially dependent on foreign computing infrastructure could become vulnerable to geopolitical disputes, export restrictions, infrastructure disruptions, or changes in the commercial terms under which that computing capacity is made available.
Those risks differ depending on the countries involved. A facility operating in Canada presents a different geopolitical and regulatory profile from one operating in China, and American semiconductor export controls also constrain where certain advanced hardware can be deployed.
The potential for international relocation therefore shouldn’t be exaggerated into an assertion that all computing infrastructure is geographically interchangeable.
Yet dismissing the possibility altogether would be equally imprudent.
Perhaps the most interesting secondary effect concerns competition within the AI industry itself. The All-In panel argued that limiting new infrastructure could disproportionately benefit companies already possessing substantial computing capacity, particularly large cloud providers and technology conglomerates.
The reasoning is straightforward. If new facilities become difficult to develop, existing capacity becomes more valuable. Companies with extensive infrastructure and access to capital can maintain or expand their competitive advantages, while smaller AI developers and emerging cloud providers face higher costs and more constrained access.
A regulatory response ostensibly intended to restrain powerful technology companies could therefore strengthen their position by restricting the supply of a resource essential to competing with them.
That outcome would be especially unfortunate if it diminished the availability of AI tools to smaller businesses and individual entrepreneurs, who might otherwise benefit disproportionately from inexpensive and widely accessible computing capabilities.
I find this one of the more persuasive arguments against indiscriminate restrictions on infrastructure development. A healthy AI ecosystem requires more than excellent models. It also requires sufficient computing capacity, viable competition, and the ability of new entrants to access the resources necessary to challenge established firms.
Protecting households from inappropriate utility costs, environmental degradation, and dubious public subsidies is entirely compatible with those objectives. Broad prohibitions, however, risk imposing consequences that extend far beyond the communities in which they originate.
The political paradox following the 2026 midterms
The November 3 midterm elections produced a decisive Democratic victory, returning both chambers of Congress to Democratic control and delivering a substantial repudiation of Republican governance under President Trump.
The result reflected a complex collection of economic and political circumstances, particularly concerns surrounding affordability, inflation, and the broader direction of the administration. It would be intellectually careless to interpret the election as an unambiguous public referendum on artificial intelligence, although growing opposition to data-center development and skepticism toward large technology companies formed part of the surrounding political environment.
As someone who considers himself politically agnostic, perhaps more accurately a political refugee, I find the implications considerably more interesting than the partisan outcome itself.
Democrats have historically placed substantial emphasis on expanding access to healthcare, supporting public education, protecting employment, and maintaining programs intended to assist economically vulnerable households. Financing those priorities requires substantial and sustainable tax revenue, much of which presently depends on individual income and employment-related contributions.
If AI progressively diminishes demand for highly compensated professional labor, government could encounter pressure on those revenue sources while simultaneously confronting greater demand for assistance.
The resulting dilemma is rather uncomfortable.
Restricting AI development could preserve certain forms of employment temporarily, particularly in industries where adoption remains discretionary. But sufficiently aggressive restrictions could also constrain productivity growth, discourage domestic investment, diminish technological competitiveness, and surrender portions of the resulting economic activity to other countries.
Unrestrained development presents its own difficulties. AI could generate enormous productivity gains while concentrating the financial returns among a comparatively small population of capital owners. Government revenue might grow, but the distribution of prosperity could become increasingly uneven.
Neither outcome offers a particularly attractive foundation for long-term social stability.
The political challenge will therefore involve accommodating technological advancement while ensuring that the economic gains remain sufficiently broad to sustain household purchasing power and public finances.
That becomes especially difficult when political rhetoric treats technological development and worker prosperity as inherently opposing objectives.
The November All-In discussion exposed some of these tensions, particularly the suggestion that opposition to data centers may reflect deeper dissatisfaction with economic conditions rather than exclusively environmental or technological objections.
I think that interpretation has merit. When households are struggling with housing, healthcare, electricity, and everyday expenditures, a proposed multibillion-dollar computing facility can become a highly visible symbol of an economy whose benefits appear increasingly concentrated elsewhere.
Residents may reasonably wonder why a company capable of financing enormous infrastructure investments requires tax incentives, or why their utility bills might increase to accommodate the development.
Dismissing those concerns as ignorance is unlikely to produce much political progress.
Yet the proposed remedies can become economically counterproductive if their broader consequences receive insufficient consideration.
A policy intended to prevent a technology company from exploiting local resources might instead discourage investment, reduce construction employment, and increase America’s dependence on computing services supplied from abroad.
It might also strengthen existing technology conglomerates by reducing the capacity available to their competitors.
Those possibilities deserve serious consideration before political opposition hardens into indiscriminate prohibition.
The emerging Democratic coalition is far from uniform on AI policy, and its constituencies include technology workers, entrepreneurs, labor organizations, environmental advocates, and consumers with substantially different interests. Some lawmakers favor restrictions on infrastructure development, while others are pursuing regulations concerning utility costs, liability, consumer protection, and worker assistance.
There is an enormous difference between requiring responsible development and preventing development altogether.
My expectation is that the distinction will become one of the more consequential political arguments of the coming decade.
The Industrial Revolution offers a warning, with an important qualification
Historical analogies have their limitations, particularly when applied to technologies whose eventual capabilities remain uncertain. Nevertheless, the Industrial Revolution offers useful insight into how societies accommodate transformative changes in productivity.
Mechanization disrupted established trades, altered the relative value of different skills, and reorganized the relationship between workers and owners of productive capital. Some occupations declined while others expanded, and the resulting economic benefits accumulated unevenly across regions and generations.
The eventual improvement in living standards is undeniable, but the timing deserves attention.
Technological advancement can precede broadly distributed prosperity by a considerable interval. During that period, workers may experience deteriorating bargaining power, declining demand for established skills, and substantial difficulty transitioning into newly emerging occupations.
The economic gains of industrialization were accompanied by institutional changes that took generations to develop. Education expanded, labor protections evolved, infrastructure improved, and governments gradually established mechanisms intended to mitigate some of the consequences of economic insecurity.
The distinction between immediate displacement and eventual prosperity is important because history’s reassuring conclusions are often drawn from a considerably longer perspective than the one available to the people experiencing the transition.
A worker whose profession becomes economically obsolete at forty may derive little comfort from the possibility that a new industry will provide excellent opportunities to workers born twenty years later.
AI could compress portions of this adjustment into a much shorter period.
Software can be distributed rapidly across organizations, often without the physical construction, machinery installation, or geographical relocation historically associated with industrial transformation. Once a capability becomes sufficiently reliable and economical, it may be adopted across thousands of businesses in comparatively little time.
The rate of technological diffusion could therefore exceed the rate at which educational institutions, labor markets, and fiscal systems can adapt.
There is, however, a substantial countervailing force.
Technological progress frequently creates new tasks, products, and industries that were difficult to anticipate beforehand. Economists Daron Acemoglu and Pascual Restrepo have explored this relationship between automation’s displacement effects and the creation of new tasks that restore demand for human labor.
Greater productivity can also reduce prices, increase real purchasing power, and make previously uneconomical services commercially viable.
AI could dramatically reduce the capital required to establish and operate sophisticated businesses. An individual might eventually build a company offering services that historically required a considerable professional workforce, allowing entrepreneurship to expand even as traditional employment structures contract.
The possibility is particularly compelling for those with specialized expertise, strong commercial relationships, and the ability to combine human judgment with increasingly capable technological tools.
But the existence of new opportunities doesn’t establish that every displaced worker will be positioned to exploit them, or that the resulting occupations will provide comparable compensation.
The speed, distribution, and accessibility of those opportunities will determine how much economic hardship accompanies the transition.
A forecast for 2027 through 2040
I don’t consider a prolonged nationwide employment collapse the most likely outcome on the evidence currently available. There are simply too many variables involving demand, productivity, entrepreneurship, and technological capabilities to justify such a deterministic conclusion.
I do, however, consider a period of increasing professional-class polarization plausible, particularly if AI reduces demand for certain occupations faster than the economy creates sufficiently remunerative alternatives.
The following is necessarily speculative, but I think the economic mechanisms are sufficiently clear to suggest how the transition might develop.
2027–2030: Accelerating adoption and increasingly contentious infrastructure
Over the next several years, I expect AI adoption to continue expanding across professional services, administrative operations, software development, finance, and other knowledge-intensive industries. Much of the initial benefit will probably appear through enhanced productivity rather than wholesale elimination of existing workforces.
Organizations will discover that fewer employees can manage greater workloads, and hiring practices will adjust accordingly. Some companies will expand output without materially expanding headcount, while others will gradually consolidate departments through attrition. The effects on junior recruitment may become increasingly visible, particularly in occupations where introductory tasks are readily automated.
Meanwhile, public debate surrounding data centers will intensify. Governments will face competing demands to encourage investment, protect ratepayers, preserve environmental resources, and prevent the appearance of preferential treatment for large technology companies.
I suspect political resistance will impose additional conditions on development, although broad efforts to suppress AI capabilities are unlikely to achieve their intended objectives without substantial economic consequences.
International competition will become increasingly relevant. Countries offering reliable electricity, reasonable permitting processes, and favorable investment conditions will attract a growing share of new infrastructure development. American firms will continue investing both domestically and abroad, but the geographic allocation of that investment could become more sensitive to regulatory uncertainty.
The danger is that American policymakers begin treating domestic infrastructure development as expendable while underestimating the economic and strategic value of retaining substantial computing capacity.
2030–2035: The professional succession problem becomes difficult to ignore
By the early 2030s, the cumulative consequences of declining recruitment could become more apparent.
Occupations that previously absorbed large numbers of university graduates may require materially fewer junior employees. Competition for remaining positions could intensify, placing downward pressure on starting salaries and weakening the economic return associated with certain educational credentials.
Experienced professionals might remain comparatively secure, particularly those possessing substantial judgment, institutional knowledge, customer relationships, and supervisory responsibilities. Younger workers could face a considerably more difficult path toward acquiring those advantages.
The implications would extend into higher education, household formation, housing markets, and consumer spending. Universities may confront pressure to restructure programs and justify tuition costs, while graduates entering an increasingly competitive professional labor market become more cautious about major financial commitments.
Some industries may eventually discover that reduced recruitment has compromised their ability to develop experienced successors. Organizations will then need to reconsider how professional expertise is cultivated when the routine tasks historically used for training have largely been automated.
At the same time, AI-enabled entrepreneurship could expand substantially, producing new commercial opportunities that compensate for a portion of the decline in conventional employment.
I suspect the distribution of those opportunities will be uneven.
Individuals with capital, established relationships, technical fluency, and commercial expertise may prosper enormously. Those dependent on traditional employment pathways could encounter greater difficulty translating education into financial security.
The economy might continue growing even as the circumstances of portions of its professional workforce deteriorate.
2035–2040: The distribution of productivity becomes a central political question
If labor’s share of national income contracts materially, the political discussion will eventually move beyond whether AI should be regulated and toward how the resulting economic value should be distributed.
The present American tax system relies heavily on employment-related income, while the country’s largest social obligations are already confronting significant demographic and fiscal pressure.
An economy capable of generating extraordinary productive output with comparatively less human labor may require a different approach to financing public institutions.
Some policymakers will advocate adjustments to corporate and capital-income taxation. Others will favor wage insurance, portable benefits, worker profit-sharing, expanded tax credits, or direct income transfers.
There will also be pressure to ensure that the economic gains from AI accrue to a sufficiently broad population to support the consumer demand upon which businesses continue to depend.
The challenge will be to accomplish that without undermining productive investment, discouraging entrepreneurship, or driving valuable economic activity into more accommodating jurisdictions.
I would expect considerable political disagreement over where those boundaries should be drawn.
An indiscriminate tax on computing could penalize the very productivity improvements capable of strengthening public finances. A more effective approach to taxing certain capital returns or economic rents might produce different incentives, although no policy eliminates the underlying trade-offs.
The question of ownership will become increasingly significant. If AI-generated productivity is broadly accessible through competitive markets, inexpensive tools, and widespread entrepreneurship, a much larger portion of society may participate directly in the economic benefits.
If the relevant infrastructure and intellectual property become highly concentrated, the distribution could look considerably different.
Those outcomes will depend partly on market competition and partly on political decisions concerning access, infrastructure, taxation, and the treatment of capital.
The future of prosperity
I remain enthusiastic about artificial intelligence and the possibilities it creates. The prospect of individuals gaining access to capabilities that once required large institutions is remarkable, as is the potential for substantial improvements in medicine, scientific research, education, and economic productivity.
It would be an extraordinary mistake to diminish those opportunities through poorly conceived restrictions motivated primarily by fear or political expediency.
It would be equally imprudent to assume that technological advancement inevitably distributes its benefits broadly enough, or quickly enough, to preserve the institutions and economic relationships upon which society depends.
The history of industrialization provides ample evidence that productivity and prosperity can evolve on different timelines. The present research suggests that AI may be beginning to alter professional employment through reduced recruitment, even before widespread displacement becomes apparent in aggregate labor statistics.
At the same time, the geographic distribution of computing infrastructure is becoming an increasingly consequential economic decision. Countries competing for investment understand that the benefits extend into energy development, construction, engineering, employment, technical expertise, and the longer-term opportunities created by access to advanced computing.
Americans are entitled to demand that infrastructure development respect local resources and avoid transferring private costs onto the public. But policymakers should recognize that international competitors are under no obligation to adopt comparable restrictions, particularly when their populations regard AI as an instrument of national advancement.
The possibility that America could restrict its own computing infrastructure while continuing to consume AI services generated elsewhere is worth taking seriously. It would be a curious form of economic self-sabotage to surrender productive capacity and associated investment without meaningfully preventing the technological displacement that prompted the restrictions in the first place.
I suspect the most consequential economic outcome of AI will involve a growing separation between aggregate productive capacity and the market value of certain forms of human labor.
Such a transition could produce extraordinary prosperity among some populations while leaving others confronting diminished employment opportunities, lower wage growth, and increasingly uncertain returns on education.
Government revenues could rise in absolute terms while becoming less closely connected to the income of ordinary households. Corporate profitability could expand even as some communities experience deteriorating purchasing power. A country might report excellent productivity growth while a substantial portion of its population becomes increasingly skeptical that economic progress is improving their circumstances.
That combination would create considerable pressure on political institutions, particularly those responsible for financing public benefits and maintaining social stability.
There is also a far more constructive possibility. AI could dramatically reduce barriers to entrepreneurship, expand access to sophisticated capabilities, and allow a much broader population to participate directly in the ownership and production of economic value. Greater productivity could increase real incomes, lower costs, and support industries that barely exist today.
Which outcome predominates will depend on the speed of adoption, the emergence of new occupations, competition, capital ownership, public policy, and the ability of educational and fiscal institutions to evolve.
I don’t pretend to know precisely how those variables will interact, and I am skeptical of anyone expressing extraordinary confidence in their predictions.
What seems increasingly clear, however, is that the economic implications extend far beyond the number of jobs any particular AI model might eliminate.
They concern the structure of professional advancement, the value of education, the distribution of capital, the financing of public institutions, and America’s position within an increasingly competitive international economy.
My hope is that we approach those questions with enough intellectual curiosity to understand the risks without allowing them to overwhelm the opportunity.
Artificial intelligence may become one of the most consequential instruments of economic progress in human history. Whether that progress becomes broadly shared will depend on decisions being made long before its ultimate consequences are visible.
I keep returning to the same principle that has informed much of this inquiry: all ships rise and fall with the tide. An economy capable of producing extraordinary wealth is a remarkable achievement, but its long-term stability remains inseparable from the prosperity of the society surrounding it.
AI may generate an economic tide unlike anything we’ve previously experienced. Preparing for it requires understanding who stands to benefit, who may be displaced, and how the resulting prosperity can be sustained.
The opportunity is extraordinary. So, too, is the responsibility to understand what we’re building.
Research and methodology
This article incorporates publicly available research published through November 8, 2026, including Anthropic’s Labor Market Impacts of AI, the Bureau of Labor Statistics’ 2025 occupational employment matrix, Stanford’s Canaries in the Coal Mine, the Census Bureau’s Graduating into Disruption, the Congressional Budget Office’s February 2026 fiscal projections, RAND’s Federal Revenue When AI Replaces Labor, and Yale’s Budget Lab analysis of AI-driven growth and federal taxation.
The infrastructure and international-adoption discussion draws on Lawrence Berkeley National Laboratory’s 2026 data-center energy research, Gallup’s September 2026 international AI survey, Stanford’s 2026 AI Index, Federal Energy Regulatory Commission proceedings, Google’s November 2026 Malaysian infrastructure announcement, and the November 7, 2026 episode of the All-In podcast.
The occupational salary and employment data reflect published BLS estimates. Federal labor-tax contributions are standardized illustrations rather than measurements of actual tax revenue by occupation. The fiscal scenarios use hypothetical displacement assumptions and exclude secondary economic effects; they should not be interpreted as predictions of future unemployment or government receipts.
The longer-range economic and political forecasts represent the author’s interpretations of available evidence, developed with substantial analytical and editorial assistance from OpenAI.