The collapse of marginal costs is enabling one-person companies to compete at scale AI and platforms are reshaping firm boundaries, productivity, and market structure Policy must adapt to support solo firms while limiting new platform bottlenecks
Longevity inequality turns years of life into a form of inherited economic advantage Wealth buys time through better prevention, treatment, and protection from medical ruin Closing the life-expectancy gap is structural economic policy, not just healthcare reform
AI capital is being miscounted, shrinking the economy on paper Hidden AI assets distort policy, funding, and skills planning Fixing measurement is now a growth and governance priority
Currently, most official economic reports treat the money companies spend on building AI models, gathering data, and integrating AI into their workflows as regular expenses rather than investments that will create value later. This is a problem because these investments are valuable.
Some studies suggest that AI could add trillions of dollars to the global economy each year. If policymakers and educators rely on economic data that does not accurately reflect these investments, they will misallocate resources for training, infrastructure, and education. This could lead to underfunded training programs, poorly planned public investments, and education systems that do not prepare people for jobs that require AI skills. Accurately measuring what we are building with AI will help us direct resources to the right places. If we continue to ignore AI's strengths, the public sector will focus on minor, obvious things rather than the fundamental drivers of economic growth.
Intangible assets are now essential to how companies create value. Companies are buying AI models and cloud computing time. Also, they are paying engineers to develop prompt libraries and rating systems. They are also creating ways for AI to function smoothly within their businesses. Much of this spending is recorded as current expenses under standard accounting practices. This makes it appear that these expenses reduce current profits without increasing the company's value. This accounting method affects decisions about treasury forecasts, education budgets, and public investments. Because the way we measure things changes slowly, an economy that is shifting toward scalable digital services that cost little to expand will seem weaker than it is on paper.
Three reasons exist why this miscounting continues. First, many AI-related expenses are diverse and change quickly, making them difficult to categorize neatly. Second, it is hard to measure the true cost because the prices of software-like services do not keep pace with technological changes. Therefore, improvements are hidden and hard to find in economic data. Third, many of AI's products are not sold directly or are bundled with other products. It shows up as increased benefits to consumers or as advantages that spread across companies, rather than as clear payments. Together, these things create an invisible capital problem. Fixing this issue is not just a matter of accounting. It changes which skills schools focus on, how administrators allocate funds for labs and cloud resources, and how regulators design incentives for public infrastructure such as power and data centers.
Recent research and reports show that AI's value is visible enough to take action. Studies that include broader categories of intangible assets (such as research and development, software, brands, and marketing—things you can't physically touch but that provide value) in national accounts show significant improvements in measured capital and worker output. The Bureau of Economic Analysis (a U.S. government agency that measures economic statistics) and related research have found that including recognized intangible assets significantly changes how we understand economic growth, especially in service and technology-heavy industries. These updated measurements indicate that intangible assets, including digital investments, have become major factors in production in many advanced economies.
To show how big this is, we can use a simple estimation. If we take the middle estimate of AI's potential impact (about $3.5 trillion per year globally) and assign 30–40% of that to company capital (model building, data sets, organization, cloud infrastructure), we get a capital contribution of about $1.05–$1.4 trillion per year. Also, a separate industry prediction suggests that about $100–200 billion in AI infrastructure spending (data centers, networking, high-performance chips) has been added to global investment through 2024–25. A portion of this spending increases physical capital and should be counted as investment. When national accounts fail to properly record these flows by treating them as immediate expenses or by ignoring changes in quality, measured GDP can underestimate real production by billions or trillions of dollars in large economies. The exact number depends on reasonable assumptions, but the key point is that the amount missed is large enough to change policy decisions.
If capital is not being properly accounted for, education is where these issues must be addressed. Traditional education that focuses solely on specific software skills or outdated IT training categories will miss the most valuable skills that create value when AI is used as capital. These skills include data management, prompt engineering as a form of system design, evaluation skills, and the ability to change work processes using AI outputs. This means three things must change. First, schools must teach students to design AI as a company asset, rather than just using individual tools. Second, administrative budgets should treat some AI-related expenses as investments in the business. This includes lab access, curated datasets, and faculty time for building reusable educational resources, rather than treating these as one-time operating costs. Third, policymakers should change funding formulas and accreditation standards to reward creations that can be reused and have lasting worth, such as data sets, modular courses, and validated evaluation methods. These changes ensure that both public and private groups create assets that provide long-term services.
Administrators must also change how they purchase and budget for things. Typical line-item purchasing for software or consulting fails to indicate whether a purchase creates a lasting asset. Treating certain purchases as capital would change depreciation accounting, free up cash, and yield clearer measures of return on investment for training programs. It also affects how institutions negotiate with cloud providers and chip sellers. Longer-term contracts that promote portability and reuse can be seen as investments that support the mission of education. Finally, policymakers should invest in measurement projects. For example, an AI intensity index that combines provider data, procurement records, and survey data would give education leaders the information they need to prioritize investments effectively.
Some may argue that counting AI expenses as capital could inflate balance sheets with things that disappear quickly. This is a fair warning. The solution is to have clear definitions and standards. Not every software purchase is capital. Standards must define AI capital as expenses that create redeployable service flows, such as model data, labeled datasets, internal evaluation tools, and systems that enable repeatable processes. National statistical agencies and standards organizations can make these distinctions clear. NIST and its partners can help create these standards. Once agreed upon, they allow auditors to distinguish between long-term assets and short-term consumption. The result is more accurate, not less.
Measurement is important for governance. When national accounts undercount AI-related capital, policymakers underinvest in public resources that enable fair access, such as electricity upgrades near data hubs, shared public datasets with privacy protections, and training programs for underserved communities. Industry investment in large data centers and specialized chips can increase overall capacity, but without public frameworks and shared infrastructure, benefits concentrate in companies and regions that already have advantages. Several recent studies point to a noticeable, but uneven, improvement in GDP from data center and chip investment. These improvements are meaningful, but they increase regional inequality if left unaddressed.
This suggests two policy priorities. First, create public-private funding mechanisms that support shared AI investments, such as curated datasets for public use, subsidized computing for universities and small companies, and regional improvements to electricity and networking that lower costs for newcomers. Second, change the way taxes are handled so that investments in reusable educational AI assets qualify for public support and favorable depreciation, encouraging the development of common resources rather than just supporting existing vendors. Both steps help make the benefits of AI more accessible while ensuring that national accounts better reflect social worth.
If the way we measure the economy is flawed, so will our policy choices. By treating many AI expenses as regular costs, official statistics make the economy seem smaller and less wealthy than it is. This affects everything from workforce planning to infrastructure investment. The solution is simple: use a strategy that combines a short-term AI intensity index with longer-term changes to national accounts that clearly identify AI's intangible assets. For educators and administrators, the message is clear: shift from teaching basic tool use to building lasting skills, and treat the creation of reusable data sets, models, and evaluation systems as valuable investments. Policymakers must fund public resources and set accounting rules that recognize long-term AI capital. When we fully account for what we build, our decisions will align with our ambitions.
The views expressed in this article are those of the author(s) and do not necessarily reflect the official position of the Swiss Institute of Artificial Intelligence (SIAI) or its affiliates.
Bureau of Economic Analysis. (2023). Marketing, Other Intangibles, and Output Growth in 61 Industries (Working Paper). U.S. Department of Commerce.
Brookings Institution. (2026). Counting AI: A blueprint to integrate AI investment and use data into US national statistics. Brookings Economic Studies.
Business Insider. (2025). AI's economic boost isn't showing up in GDP, and Goldman says that's a $115 billion blind spot. Business Insider.
McKinsey & Company. (2023). The economic potential of generative AI: The next productivity frontier. McKinsey Global Institute.
OECD. (2024). Digital Economy Outlook 2024 (Volume 1): Embracing the Technology Frontier. Organisation for Economic Co-operation and Development.
Reuters. (2025). J.P. Morgan forecasts spending on data centers could boost US GDP by 20 basis points in 2025-26. Reuters.
LLMs replace Wikipedia by absorbing its knowledge while diverting human attention AI convenience erodes verification and shared correction Public knowledge now needs active policy protection
Large language models (LLMs
World models let AI build an internal map of reality to simulate and plan They move systems beyond prediction toward reasoning and transfer Grasping world models is now essential for future AI education and policy
Minimum wages insure routine workers inside firms Shocks tend to push adjustment onto high-skill jobs Policy must pair the firm-level minimum wage with portable support for talent
The increase in South Korea's statutory minimum wage during the late 2010s and early 2020s provides an opportunity to examine how minimum wage laws affect businesses. According to the International Labour Organization, the minimum monthly wage in South Korea in 2017 was 1,352,230 KRW, rather than the previously stated figure. Instead of viewing the minimum wage as just another expense for companies, it can be seen as a form of internal insurance. It sets a standard for routine jobs, allowing for more flexibility in higher-skilled positions when the business faces challenges. This shift has implications for developing talent, fostering local innovation, and ensuring that government investments in training programs yield the expected benefits. This perspective shifts our thinking about minimum wage policy from a broad issue of income distribution to a matter of organizational structure, offering practical solutions to protect both the minimum wage and the ability to perform complex work within the country.
A minimum wage acts like an insurance agreement written into a company's payroll. When sales decline, managers decide where to cut costs. If the base pay for many routine positions is protected by law, adjustments are more likely to affect bonuses, working hours for mid-level employees, or pay cuts and layoffs for those in top-skilled positions. This situation creates an imbalance where lower-paid workers keep their wages but face a greater risk of job loss, while higher-paid workers experience wage reductions and job cuts. Studies have confirmed this pattern, showing that minimum wage laws reduce wage losses at the lower end but increase negative adjustments for higher-paid workers within the same business.
There are three direct consequences to this imbalance. First, companies with few highly skilled workers struggle to innovate, maintaining routine operations but struggling with risky projects. Second, while routine workers benefit short term, the company’s long-term product strategy declines as investment in skilled personnel drops. Third, advanced skill development through apprenticeships, internal training, and mentoring is hurt when top talent is used as a buffer. These effects reach beyond single businesses, influencing public training programs, university courses, and innovation plans, all reliant on returns from advanced skills.
Policy debates often gauge overall winners and losers, but this is too broad. For decisive action, measure how many workers at a company earn the minimum wage, and how sensitive higher wages are to company sales. Monitoring these metrics by industry and region would help benchmark risk. Businesses and policymakers could then see when skilled staff are being used as a buffer, and target support more effectively.
Another tool is to provide temporary wage support for displaced high-skilled workers. This support is not just a standard unemployment check but rather a wage-linked payment that helps cover part of a displaced specialist's lost income while they retrain or seek new employment. This support reduces the cost of retaining or rehiring skilled staff and minimizes skill loss during unemployment. Paired with employer-recognized certifications, it preserves the value of advanced skills within the economy. Public funding can be structured in stages, with shorter, full payments that decrease as the worker finds a new job, along with employer contributions tied to rehiring. The goal is to keep skilled workers active in the job market rather than unemployed. A report by Kim Kyeong-pil notes that South Korea’s unemployment benefits can exceed the net income of full-time minimum-wage workers, highlighting a flaw in the country's unemployment insurance structure.
A further step would require companies to report data on minimum-wage employment. A simple disclosure, such as the percentage of employees earning at or near the minimum wage and how their wages change in response to sales fluctuations, can be easily collected through payroll systems. This information helps to better target retraining programs and support for employers. It also influences management decisions, as boards and investors can see when a company is using top-level pay as a shock absorber and can push for better workforce strategies or invest in programs that maintain the company's ability to innovate.
The ease with which skilled workers can move across borders changes the situation. When domestic productivity declines, highly skilled workers may choose to move abroad rather than remain underemployed. While this can benefit individuals, it represents a loss for the country, as it loses skilled workers and the return on public training investments decreases. Data indicate high levels of worker mobility into developed countries in recent years, with permanent migration increasing significantly around 2021–2023. This is important for smaller economies that struggle to replace their skilled workforce. Therefore, policies must balance the ability to move with incentives to stay.
When mobility is limited, minimum wage laws can keep skilled workers in the country but push them into lower-quality jobs, slowing productivity and lowering the long-term value of education. This is concerning for markets with less mobility, as public investment in education becomes less effective and public support weakens. The best approach combines portable wage support, rapid retraining, and temporary public funding to support job transitions. For high-emigration countries, focus on retention through tax credits for rehiring, short public projects for displaced specialists, and international exchanges that maintain connections while offering mobility.
Educators should update certifications to be more transferable and prioritize short, industry-recognized options like micro-credentials and fast-track training. Career services must actively support workers in their job transitions. Universities should partner with industry for short fellowships that rapidly reintegrate workers into the workforce. Such programs should directly address how minimum wage increases shift economic pressures, ensuring that training supports mobility and reduces rehiring costs while maintaining the domestic skill base.
According to the International Labour Organization, administrators should create a national system to track minimum-wage employment and introduce short-term, industry-specific retraining grants, available only if employers invest in retraining their displaced workers. According to a report from Aju Business Daily, policymakers should consider launching wage-support programs targeted at specialized groups such as R&D staff or senior engineers and closely track their re-employment and retention rates against control groups. This approach could help identify which policies retain skilled workers most cost-effectively, particularly given the connection between low domestic pay and the rising overseas employment of these professionals.
It is essential to address minimum wage laws with a commitment to both social equity and economic resilience. Policymakers, industry leaders, and educators must unite to develop strategies that safeguard low-wage workers while sustaining innovation and advanced skills. Demand transparency: publish concrete data on policy impacts, and communicate how portable wage support can maintain the value of domestic training. Push for temporary employer incentives that prioritize rehiring skilled workers. Only by taking decisive, coordinated action can we ensure that minimum wage protections lead to sustainable growth and readiness for future challenges.
Examining minimum wage laws at the company level reveals a trade-off: while a minimum wage protects lower-income workers, it puts pressure on skilled workers. This is not an argument against minimum wages but rather a call to combine them with well-aimed tools that keep talent engaged, mobile within the country, and productive. Start by measuring the impact, creating portable wage support and rapid retraining programs, and funding short-term public projects that employ displaced specialists. These actions maintain the protective intent of minimum wage laws while preserving the benefits of training and innovation. Failing to act within companies will protect lower-wage workers while weakening the base of skilled labor, leading to slower growth and reduced opportunities. The solution lies in policy design; the cost of inaction is predictable and significant.
The views expressed in this article are those of the author(s) and do not necessarily reflect the official position of the Swiss Institute of Artificial Intelligence (SIAI) or its affiliates.
Adamopoulou, E., et al. (2024). Minimum wages and insurance within the firm. SSRN working paper.
CEPR / VoxEU. (2024). Minimum wages and insurance within the firm. VoxEU column summarizing firm-level evidence.
CountryEconomy. (2025). South Korea National Minimum Wage (USD equivalents). CountryEconomy data series.
International Migration Outlook 2023. OECD. (2023). International Migration Outlook 2023. OECD Publishing.
MacroTrends. (2023). South Korea inflation rate (CPI) 2022. MacroTrends economic data.
Reciprocal tariffs raise costs at home, shrink global trade, and rarely deliver lasting protection When two countries retaliate, third-party exporters often gain while consumers and firms lose Measuring the true cost of protection shows tariffs and counter-tariffs are equally damaging policies
Inflation is a mix of shocks and trends, not a single number Inflation decomposition clarifies causes and improves policy decisions It should be central to both forecasting and economic education
Inflation isn't simple.
The Donroe Doctrine replaces global leadership with blunt self-interest Tariffs now function as leverage, not policy tools Education and governance must adjust to a less predictable order
The Donroe Doctrine isn’t a sweeping histori
