Edward Conard

Top Ten New York Times Bestselling Author

  • “Unintended Consequences is full of substance, it is one of the must-read books of the year, and once I finish it I will be giving it a second read through right away.” - Tyler Cowen, Professor, George Mason University
  • “…reminds us that inequality sends a signal of what society lacks most, in America’s case, entrepreneurship and risk taking.” - Lawrence Lindsey, CEO, The Lindsey Group, former Director of the National Economic Council
  • “Unintended Consequences offers deep and well-argued analyses on almost every issue.” - The New York Times
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “Unintended Consequences provides a provocative interpretation of the causes of the global financial crisis and the policies needed to return to rapid growth. Whether you agree or not, this analysis is well worth reading.” - Nouriel Roubini, New York University; Chairman, Roubini Global Economics
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Monday, February 23, 2026

AI Summary. The US has a relative surplus of low-paid hourly workers and a shortage of high-skill workers, so many less productive US manufacturers could buy a robot but could not attract the talent to program it. China faces the opposite conditions, with underemployed STEM graduates, residency rules complicating coastal factory hiring, and robot programmers earning a fraction of US wages.

Austin Vernon Austin Vernon Blog
Date Posted:
Is Database:
Database
Is Important:
Important

High US wages for skilled workers constrain robot adoption in manufacturing. “Many less productive US manufacturing firms could buy a robot, but couldn’t attract the talent to program it.” The US can’t “emulate” the low-wage, high-skilled China model.

Does a shortage of high-skill workers limit US factory automation?

Core argument: The bottom 10% of US workers earn roughly as much as their counterparts in other developed countries, while US workers in the upper deciles perform much better than their international peers, making a skilled-for-hourly substitution a poor trade in many cases.

The US has a relative surplus of low-paid hourly workers and a shortage of high-skill workers, making a skilled-for-hourly replacement a poor trade in many cases. Our bottom 10% of workers earn roughly as much as those in other developed countries, even though US workers in the upper deciles perform much better than their international peers. Many less productive US manufacturing firms could buy a robot, but couldn't attract the talent to program it. China has the opposite problem, making robots more attractive. It has a surplus of STEM graduates that are often underemployed, while "Hukou" residency restrictions complicate hiring hourly labor in coastal factories. Low-level robot programmers earn a fraction of the US wage, even at a much higher supply in absolute and per capita terms. These jobs are boring and require long stints away from home in relative backwaters to support factory retoolings. The alternative offers for those with programming skills in the US are much better. Two key concepts come from this. First, high-end labor costs and availability are a challenge for US manufacturing. Second, technologies and market structure are contextual, where the US can't simply emulate China's manufacturing.

Takeaways by Macro Roundup® AI

  1. The bottom 10% of US workers earn roughly as much as their counterparts in other developed countries, while US workers in the upper deciles perform much better than their international peers, making a skilled-for-hourly substitution a poor trade in many cases.
  2. Many less productive US manufacturing firms could buy a robot but could not attract the talent to program it, since alternative offers for those with programming skills in the US are much better than long stints away from home supporting factory retoolings.
  3. China faces the opposite problem, with a surplus of often underemployed STEM graduates, low-level robot programmers earning a fraction of the US wage, and "Hukou" residency restrictions complicating hourly hiring in coastal factories, making robots more attractive there.

Related Articles:

  • Boeing Battles Brain Drain As Engineers Chase the Allure of Space — The average tenure of a Boeing engineer has fallen from 16.4 years to 12.6 years, according to data from the union representing 12,000 Boeing engineers, with tenure shortening in almost every age bracket. Analysts, recruiters and union officials describe this "brain drain" as a drag on operations that could make launching a new plane harder.
  • What Would It Take to Bring Back US Manufacturing? Part 2: Making American Manufacturing More Productive — The authors argue the United States is a decade away from the scale of automated "lights-out" manufacturing needed to meet its own end-to-end demand at comparable cost. With 98% of US manufacturers being small or medium enterprises, low deployment of mature productivity technologies reflects limited commercial adoption and investment rather than technical readiness.
  • America Is Missing The New Labor Economy – Robotics Part 1 — Building an identical robotic arm in the US costs about 2.2x more than in China, and US-labeled components still depend heavily on Chinese parts and materials with no scalable alternative. The author argues China is the only country positioned to capture full-scale intelligent robotic automation, which would let it outcompete the US across production.

AI Summary. The Budget Lab estimates the average effective US tariff rate faced by consumers reached 16%, the highest since 1936, before court-struck tariffs cut it to 9.1%, with new 15% Section 122 tariffs lifting it to 13.7%. If those tariffs expire after 150 days, the authors project $1.3tn in revenue over 2026-35, or $1.1tn after accounting for slower growth.

John Iselin Yale University
Date Posted:
Is Database:
Database

The new 15% tariffs under Section 122 bring the mean effective tariff rate to 13.7%, down from 16.9% prior to the SCOTUS ruling that the IEEPA tariffs were illegal. Section 122 tariffs will expire in 150 days unless extended by Congress.

How have court rulings and new tariffs changed effective tariff rates?

Core argument: The Budget Lab estimates the US average effective tariff rate facing consumers at 13.7%, down from 16% before the IEEPA-based tariffs were struck down — the highest since 1936 — and up from the 9.1% reached immediately after that ruling.

The Budget Lab (TBL) estimates the effects of all US tariffs and foreign retaliation implemented through February 20, 2026, including the new 15% Section 122 tariffs and the elimination of IEEPA-based tariffs. Under our baseline case, the Section 122 tariffs expire after 150 days. Before the IEEPA tariffs were struck down, consumers faced an overall average effective tariff rate of 16%, the highest since 1936. Immediately following the IEEPA ruling, the rate fell to 9.1%. After the Section 122 tariffs were imposed, the rate rose to its current level of 13.7%. If those tariffs expire in 150 days, the rate will fall again to 9.1%. Assuming Section 122 tariffs expire in 150 days, the administration’s tariffs will raise about $1.3 trillion over 2026-35, though slower economic growth reduces revenues and brings the net dynamic revenue to $1.1 trillion. (If they are instead made permanent, these figures would be $2.2 trillion and $1.9 trillion.)

Takeaways by Macro Roundup® AI

  1. The Budget Lab estimates the US average effective tariff rate facing consumers at 13.7%, down from 16% before the IEEPA-based tariffs were struck down — the highest since 1936 — and up from the 9.1% reached immediately after that ruling.
  2. The Budget Lab projects the administration's tariffs will raise about $1.3 trillion over 2026–35 if the new 15% Section 122 tariffs expire after 150 days, with net dynamic revenue of $1.1 trillion after slower economic growth reduces receipts.
  3. Should the Section 122 tariffs be made permanent, The Budget Lab estimates revenue of $2.2 trillion over 2026–35 and $1.9 trillion on a net dynamic basis; if they expire after 150 days, the effective tariff rate falls to 9.1%.

Related Articles:

  • US Notches One of Its Biggest Annual Trade Gaps Since 1960 — The US goods and services trade deficit reached $901.5bn for the year, one of the largest in data going back to 1960. US companies increased imports of computers and accessories by nearly $145bn from the prior year, an acceleration in demand that reflects large-scale artificial intelligence investment.
  • Help for the Heartland? The Employment and Electoral Effects of the Trump Tariffs in the United States — The authors find that the combined effect of U.S.-China trade war import tariffs, retaliatory tariffs, and farm subsidies on employment in exposed U.S. locations was at best a wash and possibly mildly negative, while residents of tariff-protected areas became less likely to identify as Democrats and more likely to vote Republican.
  • Did Tariffs Make American Manufacturing Great? New Evidence from the Gilded Age — Using digitized industry-level US tariff data from 1870–1900 matched to state-level manufacturing census data for over 80 industries, the authors estimate that tariffs reduced labor productivity. They argue the results are consistent with weaker international competition allowing less efficient firms to survive, reduced incentives to innovate, and lobbying for protection misallocating resources.

AI Summary. Unemployment in occupations exposed to artificial intelligence rose to 3.8% from 3%, a larger increase than the 0.3 percentage point rise in the national rate, while unemployment in non-exposed occupations edged lower. The rise is concentrated among workers aged 34 and under, whose jobless rate jumped to 5.3% from 3%.

Bernard Yaros and Matthew Martin Oxford Economics
Date Posted:
Is Database:
Database

Unemployment is rising more rapidly in AI-exposed occupations relative to non-exposed occupations, especially among younger workers. One driver of this is a rising supply of workers due to an uptick in participation.

Is unemployment rising faster in AI-exposed occupations, especially for younger workers?

Core argument: Unemployment in AI-exposed occupations rose to 3.8% in December 2025 from 3% in January, a sharper increase than the 0.3ppt rise in the national rate, while unemployment in non-exposed occupations edged slightly lower, based on Bureau of Labor Statistics microdata.

Using microdata from the Bureau of Labor Statistics, we split employment into two groups – AI-exposed and non-exposed occupations – and calculate separate unemployment rates (Chart 8). The jobless rate for AI-exposed occupations is lower than the overall unemployment rate but has risen to 3.8% in December 2025 from 3% in January, a sharper increase than the 0.3ppt rise in the national rate over the same period. Meanwhile, unemployment in non-exposed occupations has edged slightly lower. For more seasoned professionals in AI-exposed occupations – those 35 years and over – the rise in unemployment has been modest and consistent with the gradual softening in broader labor-market conditions over the past three years, rather than indicative of an AI-driven shock (Chart 9). The increase has been more pronounced among workers aged 34 and lower, where the unemployment rate has jumped to 5.3% in December from 3% at the beginning of 2025. The data are noisy, and younger workers are usually more sensitive to weakening labor market conditions. But we tested this relationship and the recent increase in unemployment for younger workers is larger than we'd expect given the rise in the national unemployment rate.

Takeaways by Macro Roundup® AI

  1. Unemployment in AI-exposed occupations rose to 3.8% in December 2025 from 3% in January, a sharper increase than the 0.3ppt rise in the national rate, while unemployment in non-exposed occupations edged slightly lower, based on Bureau of Labor Statistics microdata.
  2. Among AI-exposed workers aged 34 and under, the unemployment rate jumped to 5.3% in December from 3% at the beginning of 2025, larger than would be expected given the rise in the national unemployment rate.
  3. For AI-exposed workers aged 35 and over, the rise in unemployment has been modest and consistent with the gradual softening in broader labor-market conditions over the past three years, rather than indicative of an AI-driven shock.

Related Articles:

  • Looking for the Ladder — The article argues that the hiring downturn in occupations most exposed to generative AI reflects interest-rate sensitivity rather than AI itself, since roughly 38% of workers in the most AI-exposed quintile work in information, finance and insurance, and professional and technical services, versus under 2% in the least-exposed quintile.
  • Unprecedented ‘Jobless Boom’ Tests Limits of US Economic Expansion — Forecasters expect US gross domestic product to have expanded 2.7% over the year while employment barely grew, a divergence drawing comparisons to the "jobless recovery" that followed the early-2000s technology stock collapse. Unlike that episode, which began with a recession, this "jobless boom" is occurring without one — a first in the postwar era.
  • What Sweeping Revisions and a Blowout Month Tell Us About the U.S. Job Market — Labor Department revisions lowered previously estimated job gains for the past two calendar years, cutting the combined net change in construction, manufacturing, and retail by 115,000 jobs. Healthcare added about 437,000 jobs and social assistance about 321,000 over 12 months, while all other private-sector jobs combined shrank.

AI Summary. Capital spending by the largest cloud computing companies is projected to reach about $646bn, roughly 2% of US GDP. For comparison, annual US consumer spending growth is around $1tn, total US bank loan growth was about $700bn, and US defense spending was $917bn.

Torsten Sløk , Rajvi Shah, and Shruti Galwankar Apollo
Date Posted:
Is Database:
Database

Apollo forecasts that the hyperscalers will spend ~$656B or ~2% of US GDP on capex in 2026, outpacing the telecom companies’ spend during the dot-com bubble.

How large is cloud computing capital spending relative to the economy?

Core argument: Hyperscaler capital expenditure is expected to reach roughly $646 billion in 2026, equal to about 2% of US GDP.

The amount of money being spent on capex by hyperscalers is enormous at around $646 billion in 2026. Hyperscaler capex is expected in 2026 to be at approximately $646 billion, or about 2% of US GDP. Annual growth in consumer spending is currently around $1 trillion. For comparison, total US bank loan growth in 2025 was around $700 billion. [American] defense spending in 2025 was at $917 billion.

Takeaways by Macro Roundup® AI

  1. Hyperscaler capital expenditure is expected to reach roughly $646 billion in 2026, equal to about 2% of US GDP.
  2. Projected 2026 hyperscaler capex of about $646 billion compares with total US bank loan growth of roughly $700 billion in 2025 and US defense spending of $917 billion in 2025.
  3. Annual growth in consumer spending is around $1 trillion, exceeding the approximately $646 billion of hyperscaler capex projected for 2026.

Related Articles:

  • Rough Notes, Feb 20, 2026: AI Capex in Q4 GDP, Anthropic Revenues, Agentic Domains, etc. — An early estimate puts the floor contribution of artificial intelligence capital spending to fourth-quarter US GDP growth at about 64%, and likely closer to 80% including externalities and multipliers, though the calculation is not adjusted for imports. The author also notes that token price deflation persists and that US-China token price differences are immense and growing.
  • The Macro Implications of the AI Capex Boom — Artificial intelligence capital spending is estimated to add roughly 1.4 percentage points to US growth in one year and about 1.5 points the next, comparable to business investment's contribution during the tech bubble. A material part of that estimate reflects profits at Nvidia and a few other chipmakers that are unlikely to be recycled back into the economy.
  • Everyone’s Looking For A Bubble. No One Sees The Stampede — A proprietary model estimates monthly artificial intelligence revenue grew roughly 18-fold in two years, from $772m to $13.8bn, with AI accounting for an estimated 23% of Google Cloud, 10% of Azure and 5% of Amazon Web Services. The model's ratio of investment to revenue fell from 6.1x to 4.7x, still historically high.

AI Summary. U.S. power plant developers plan to add 86 gigawatts of new utility-scale generating capacity, which would be a record if realized, with solar at 51% of planned additions, battery storage at 28% and wind at 14%. The prior year saw 53 gigawatts added, the largest single-year installation since 2002.

EIA Staff Energy Information Administration
Date Posted:
Is Database:
Database

The US will add 86GW of new utility-scale electrical generation capacity to the power grid in 2026, a record high. Solar will make up 51% of the planned 2026 capacity additions, followed by battery storage at 28% and wind at 14%.

Is U.S. new generating capacity headed for a record high?

Core argument: U.S. power plant developers and operators plan to add 86 GW of new utility-scale generating capacity in 2026, a record if realized, according to the Preliminary Monthly Electric Generator Inventory report.

U.S. power plant developers and operators plan to add 86 gigawatts (GW) of new utility-scale electric generating capacity to the U.S. power grid in 2026 in our latest Preliminary Monthly Electric Generator Inventory report, a record if realized. Solar power makes up 51% of the planned 2026 capacity additions, followed by battery storage at 28% and wind at 14%. In 2025, 53 GW of new capacity was added to the grid, the largest capacity installation in a single year since 2002.

Takeaways by Macro Roundup® AI

  1. U.S. power plant developers and operators plan to add 86 GW of new utility-scale generating capacity in 2026, a record if realized, according to the Preliminary Monthly Electric Generator Inventory report.
  2. Solar power accounts for 51% of planned 2026 U.S. capacity additions, followed by battery storage at 28% and wind at 14%.
  3. The U.S. grid added 53 GW of new capacity in 2025, the largest single-year installation since 2002.

Related Articles:

  • Electricity 2026 — Global electricity demand is projected to grow faster than the world economy throughout the forecast period, driven by electricity-intensive industries, air conditioning, data centres and continued electrification. US electricity demand is forecast to rise about 2% a year over 2026-2030, more than twice the past decade's rate, with data centres accounting for almost half that growth.
  • Power Politics: Energy Self-Sufficiency in a Modern Mercantilist World — China leads in adding power to the grid, with more than 30,000 miles of ultra-high-voltage transmission lines versus zero in the US and lead times for large transformers of 48 weeks against a US average of 143 weeks. This positions China to accommodate new demand sources such as artificial intelligence.
  • China’s AI Power Play: Cheap Electricity From World’s Biggest Grid — China generates more than twice as much electricity as the US, and Goldman Sachs projects it will have roughly 400 gigawatts of spare capacity by 2030 — about 3x expected global data-center power demand. Morgan Stanley forecasts US data centers could face a 44-gigawatt electricity shortfall within three years.

AI Summary. Correcting for a mismatch in survey years and for comparing workers' place of work to natives' place of residence cuts the estimated H-1B wage gap to roughly one-third of the original claim. The remaining average masks that H-1B workers aged 21–22 earn about a 20% premium over comparable native-born workers, while a shortfall appears only after age 34.

Jiaxin He and Adam Ozimek Economic Innovation Group
Date Posted:
Is Database:
Database

He/Ozimek dispute Borjas’ finding that H1-B workers are underpaid. A “temporal mismatch” between the native sample and the H-1B sample means “effectively comparing 2020 wages with 2023 prices.” The ⅔ of H-1B winners aged <34 actually earn a wage premium.

Are H-1B workers underpaid relative to comparable native-born workers?

Core argument: Correcting for errors leaves an average H-1B wage gap only one-third as large as the Borjas analysis claimed, with a consistent geographic definition for both groups reducing the gap to 5.3%.

[Whether] H-1Bs are underpaid depends on which workers we examine. [Older] H-1B workers earn less than otherwise similar native-born workers, while [younger] H-1B workers earn more. [Overall, there is a] small average wage gap, [but it is] only one-third as large as the [Borjas] analysis claimed. [The above table] provides a step-by-step breakdown of how these errors manufactured the appearance of a large wage gap. The [most significant mistake] is a temporal mismatch between his H-1B and native-born samples: the H-1B applications span 2020–2023, while the ACS data [covers] just 2023. [Given] 15.1 percent inflation and [an] 18.7% wage increase for software occupations [from 2020 to 2023], comparing [wages of] H-1B workers [2020 to 2023 to] native-born [wages from 2023] overstates the wage gap, effectively comparing 2020 wages with 2023 prices. [Second], Borjas compares H-1B workers’ place of work to natives’ place of residence. [Using] a consistent definition of geography for both groups further reduces the gap to 5.3%. [Even after corrections], the aggregate average hides that younger H-1B workers are paid [more than Americans], while older H-1B workers are paid [less]. [H-1B workers ages 21–22] earn a 20% wage premium [over natives]. [After age 34], a negative wage gap emerges. Two-thirds of H-1B lottery winners are 33 or younger…the very group that earns a significant wage premium over natives.

Takeaways by Macro Roundup® AI

  1. Correcting for errors leaves an average H-1B wage gap only one-third as large as the Borjas analysis claimed, with a consistent geographic definition for both groups reducing the gap to 5.3%.
  2. The most significant error identified is a temporal mismatch: Borjas compared H-1B applications spanning 2020–2023 with 2023 ACS data, despite 15.1% inflation and an 18.7% wage increase in software occupations from 2020 to 2023.
  3. H-1B workers ages 21–22 earn a 20% wage premium over natives and a negative wage gap emerges after age 34, while two-thirds of H-1B lottery winners are 33 or younger.

Related Articles:

  • The H-1B Wage Gap, Visa Fees, and Employer Demand — An analysis merging Labor Department, Homeland Security, and American Community Survey data estimates the average H-1B visa worker earns about 16% less than a comparable US-born worker, implying payroll savings near $100,000 over a six-year visa term. Simulations suggest a $150,000–$200,000 visa fee would barely reduce hiring while raising $10bn–$20bn annually.
  • The Extreme Shortage of High IQ Workers — Semiconductor manufacturing requires workers of above-average intelligence, with the most demanding roles needing IQs in the top 1 in 1,000 — a pool of about 164,000 people in the US, of whom perhaps 100,000 can be placed in such professions. The author argues this scarcity explains reliance on high-IQ workers abroad and on smarter tools and non-human intelligence.
  • Third-Country Effects of U.S. Immigration Policy — U.S. H-1B visa denial rates rose from about 6% to 16%, and the authors estimate a 10 percentage point rise in denials increases Canadian applications by 30%, roughly one Canadian application per four forgone visas. More exposed Canadian firms increased sales and hired about 0.5 native workers per immigrant.

Tuesday, February 24, 2026

AI Summary. The author argues that capital rotates between the financial economy's asset-light booms and the physical economy, with commodity capacity constraints and the use of critical materials as geopolitical leverage now making returns on physical investment more attractive. In this view, higher interest rates reflect the better returns and capital spending needed to expand production capacity.

Jeff Currie Carlyle
Date Posted:

Since October 2025, equities tied to energy and infrastructure are up 30–50%, while technology related stocks are down 5–10%. Jeff Currie argues we are in a “decade-long physical capex super-cycle” as demand has run into physical capacity constraints.

Are commodity constraints shifting capital from financial to physical investment?

Core argument: Commodity capacity constraints, worsened by the "weaponization" of the periodic table, raise prices and shift demand toward physical over financial capital, which the author argues pushes interest rates higher and creates an inflation-duration trade-off.

Given the capacity constraints many commodity markets face today, which are made worse by the “weaponization” of the periodic table, as the global economy grinds against physical constraints and prices rise, the need for physical capital over financial capital leads to higher interest rates, creating an inflation-duration trade-off. As the cost of capital falls and investors expand their horizons, they become more focused on duration and longer-term growth opportunities. As longer duration financial returns become more attractive than physical ones, capital is redirected into the financial economy, i.e. the Nifty Fifty, Dot-com Boom and FAANG/Mag Seven Boom. Note that these were all ‘asset-light’ booms. The Nifty Fifty was about brands like Coca-Cola or McDonalds that were asset-light scalable franchises in the same way that Microsoft software was in the 1990s or Google was in the 2010s. Eventually, however, demand catches up to physical capacity constraints, creating better returns in the physical economy than the financial economy, motivating the redirection of capital back into the physical economy, i.e. 1968-1980 and 2002-2014. The higher cost of capital simply reflects the better returns in the physical economy and the need to attract capex to expand production capacity, which is where we are today.

Takeaways by Macro Roundup® AI

  1. Commodity capacity constraints, worsened by the "weaponization" of the periodic table, raise prices and shift demand toward physical over financial capital, which the author argues pushes interest rates higher and creates an inflation-duration trade-off.
  2. The Nifty Fifty, dot-com and FAANG/Magnificent Seven booms were all asset-light episodes in which a falling cost of capital drew investors toward longer-duration financial returns rather than physical ones.
  3. When demand catches up to physical capacity constraints, returns in the physical economy exceed those in the financial economy and capital is redirected toward expanding production capacity, as in 1968–1980 and 2002–2014, which the author identifies as today's condition.

Related Articles:

  • Capital Is Making a Comeback — Tangible investment in property, factories and equipment fell from 12.5% to 8.5% of private gross domestic product, while intangible investment in intellectual property, software and process knowledge rose from about 11.5% to 16.75%. The author argues demand for tangible capital investment will now have to rise.
  • The Macro Implications of the AI Capex Boom — Artificial intelligence capital spending is estimated to add roughly 1.4 percentage points to US growth in one year and about 1.5 points the next, comparable to business investment's contribution during the tech bubble. A material part of that estimate reflects profits at Nvidia and a few other chipmakers that are unlikely to be recycled back into the economy.
  • Power Politics: Energy Self-Sufficiency in a Modern Mercantilist World — China leads in adding power to the grid, with more than 30,000 miles of ultra-high-voltage transmission lines versus zero in the US and lead times for large transformers of 48 weeks against a US average of 143 weeks. This positions China to accommodate new demand sources such as artificial intelligence.

AI Summary. Wage data suggest greater artificial intelligence exposure is associated with lower wage growth in occupations where experienced workers earn little premium over entry-level workers, and with higher wage growth where that premium is large. The authors interpret this as AI substituting for workers in jobs requiring little tacit knowledge while complementing experienced workers elsewhere.

J. Scott Davis Federal Reserve Bank of Dallas
Date Posted:
Is Database:
Database

For occupations with a very low experience premium, AI exposure lowered wage growth, likely due to substitution of AI for costly experienced workers. At the top of the experience distribution, AI raised wage growth, as it complements experienced workers.

How does AI exposure affect wage growth across experience premiums?

Core argument: For occupations with a 0% experience premium, a 1 SD increase in AI exposure is associated with a 0.28 percentage point reduction in post-2022 wage growth, with a clearly negative confidence interval.

Chart 5 presents the estimated effect of a one-standard-deviation (1 SD) increase in an occupation’s AI exposure on an occupation’s post-2022 wage growth. For occupations with the median experience premium (40%), a 1 SD increase in AI exposure is associated with a 0.05ppt decline in wage growth. The 90% confidence interval for this estimate clearly includes zero, indicating that for an occupation with the median experience premium, AI exposure doesn’t much affect wage growth. However, for an occupation with a 0% experience premium, increased AI exposure is associated with a 0.28 percentage point reduction in wage growth. The confidence interval for this estimate is clearly negative. For occupations with a very low experience premium, AI exposure has a more negative effect on wage growth since AI substitutes for both entry and experienced workers. The low experience premium suggests there is not much tacit knowledge required for the occupation, so experienced workers are easily substituted by AI. Conversely, for occupations in the 90th percentile of the experience premium distribution, increased AI exposure is associated with a 0.2 percentage point increase in wage growth. For these occupations, AI exposure has less of a negative effect, and likely even a positive effect on wages, as AI substitutes for entry level workers but complements experienced workers.

Takeaways by Macro Roundup® AI

  1. For occupations with a 0% experience premium, a 1 SD increase in AI exposure is associated with a 0.28 percentage point reduction in post-2022 wage growth, with a clearly negative confidence interval.
  2. For occupations in the 90th percentile of the experience premium distribution, a 1 SD increase in AI exposure is associated with a 0.2 percentage point increase in wage growth, consistent with AI substituting for entry-level workers while complementing experienced ones.
  3. At the median experience premium of 40%, a 1 SD increase in AI exposure is associated with a 0.05 percentage point decline in wage growth, and the 90% confidence interval includes zero.

Related Articles:

  • Young Workers’ Employment Drops In Occupations With High AI Exposure — Among young labor market entrants, the rate of finding jobs has held steady only in occupations with low exposure to artificial intelligence, declining by more than 3 percentage points for the most AI-exposed group from its recent peak. The decline reflects fewer people entering these jobs rather than exits, with small aggregate effects.
  • Looking for the Ladder — The article argues that the hiring downturn in occupations most exposed to generative AI reflects interest-rate sensitivity rather than AI itself, since roughly 38% of workers in the most AI-exposed quintile work in information, finance and insurance, and professional and technical services, versus under 2% in the least-exposed quintile.
  • Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — Using payroll records covering millions of US workers, the authors find employment for workers aged 22–25 in occupations most exposed to artificial intelligence, such as software developers and customer service representatives, declined 6%, while employment for older workers in those occupations rose 6–9%. Employment in less-exposed occupations grew comparably across age groups.

AI Summary. Anthropic says three Chinese artificial intelligence developers — DeepSeek, MiniMax Group, and Moonshot — violated its terms of service by using thousands of fraudulent accounts to generate more than 16m exchanges with its Claude models. The company says this practice, known as distillation, lets rival labs improve their own models by training on outputs from more powerful systems.

Margi Murphy and Shirin Ghaffary Bloomberg
Date Posted:

Anthropic reported that three leading Chinese AI developers “illicitly” extracted results from Claude by training their own models on the outputs from the more powerful system, a tactic known as “distillation.”

Are Chinese AI labs using fraudulent accounts to distill rival models?

Core argument: Anthropic said DeepSeek, MiniMax Group, and Moonshot violated its terms of service by generating more than 16 million exchanges in total with its Claude models using thousands of fraudulent accounts to "illicitly extract" results.

Anthropic said three leading artificial intelligence developers in China worked to "illicitly extract" results from its AI models to bolster the capabilities of rival products, adding to growing concerns in the US about Chinese firms improperly gaining an edge. The San Francisco-based company said DeepSeek, MiniMax Group, and Moonshot violated its terms of service by generating more than 16 million exchanges in total with its Claude models using thousands of fraudulent accounts. With this tactic, known as distillation, Chinese AI labs can rapidly improve their models by training them on the outputs from more powerful systems, Anthropic said. DeepSeek generated more than 150,000 exchanges with Claude, and MiniMax generated more than 13 million, Anthropic said, with a focus on reconstructing Claude’s more advanced capabilities. The Claude maker said it was able to pinpoint the three companies with “high confidence” based on information on internet protocol addresses, metadata and “corroboration from industry partners who observed the same actors and behaviors on their platforms.”

Takeaways by Macro Roundup® AI

  1. Anthropic said DeepSeek, MiniMax Group, and Moonshot violated its terms of service by generating more than 16 million exchanges in total with its Claude models using thousands of fraudulent accounts to "illicitly extract" results.
  2. MiniMax generated more than 13 million exchanges with Claude and DeepSeek more than 150,000, with a focus on reconstructing Claude's more advanced capabilities, according to Anthropic.
  3. Anthropic said it pinpointed the three companies with "high confidence" based on internet protocol addresses, metadata, and corroboration from industry partners that observed the same actors and behaviors on their platforms.

Related Articles:

  • Rough Notes, Feb 20, 2026: AI Capex in Q4 GDP, Anthropic Revenues, Agentic Domains, etc. — An early estimate puts the floor contribution of artificial intelligence capital spending to fourth-quarter US GDP growth at about 64%, and likely closer to 80% including externalities and multipliers, though the calculation is not adjusted for imports. The author also notes that token price deflation persists and that US-China token price differences are immense and growing.
  • The Geopolitics of AI: Decoding the New Global Operating System — The performance gap between the top American and Chinese AI models narrowed from 9.3% to 1.7%, with Chinese firms relying on cheaper open-source models—DeepSeek prices usage 20-40x below OpenAI's ChatGPT. This price advantage could win adoption in developing countries and build dependency on Chinese AI ecosystems.
  • China’s DeepSeek Uses Banned Nvidia Chips for AI Model, Report Says — Chinese AI developer DeepSeek used Nvidia Blackwell chips that were smuggled into China through countries where their sale was permitted, according to a report citing unnamed sources. The chips were installed in data centers abroad, then dismantled and shipped to China after clearing inspection by server-equipment companies.

AI Summary. The U.S. goods trade deficit reached a record $1.24tn as goods imports rose 4.3%, while major exporters including Germany, Japan, South Korea and Taiwan launched spending programs largely tilted toward manufacturers dependent on overseas markets. The Brookings Institution argues reining in large U.S. fiscal deficits would reduce the shortfall more effectively than pressing other countries to buy more American goods.

Tom Fairless and Gavin Bade Wall Street Journal
Date Posted:
Is Database:
Database

American tariffs haven’t had a significant impact on the chronic surplus countries’ current accounts, in part due to policy responses aimed at competitiveness and continued high saving, while US consumption remains strong amidst its large fiscal deficit.

Would cutting fiscal deficits shrink the trade gap more than tariffs?

Core argument: The U.S. goods trade deficit rose to a record $1.24 trillion in 2025, driven by a 4.3% increase in goods imports.

The U.S. trade deficit in goods rose to a record high of $1.24 trillion in 2025, driven by a 4.3% increase in goods imports. Big exporting countries—Germany, Japan, South Korea, Taiwan and others—have launched government spending programs that are largely tilted toward supporting manufacturers dependent on overseas markets, directly providing credit support to exporters and small and midsize businesses. Germany’s surplus appears to be stabilizing at around 5% of GDP, higher than its fair value—a measure based on the country’s demographics and economic development—which the European Commission estimates at around 1% to 2% of GDP. China’s current-account surplus is expected to reach 4.3% of GDP this year amid surging exports and weak domestic demand, according to Goldman Sachs. Meanwhile, the U.S. current-account deficit is running at about 4% [of GDP] for the latest 12-month period, roughly double its level in 2019. Its scale partly reflects America’s much faster recovery from the pandemic than other countries, said Gian Maria Milesi-Ferretti, a senior fellow at Brookings. If the U.S. still wants to reduce its deficit, he argued, reining in its large fiscal deficits would be more effective than coercing other countries into importing more U.S. goods.

Takeaways by Macro Roundup® AI

  1. The U.S. goods trade deficit rose to a record $1.24 trillion in 2025, driven by a 4.3% increase in goods imports.
  2. Germany's surplus appears to be stabilizing at about 5% of GDP, above the 1% to 2% fair value estimated by the European Commission, while Goldman Sachs expects China's current-account surplus to reach 4.3% of GDP this year.
  3. The U.S. current-account deficit is running at about 4% of GDP over the latest 12 months, roughly double its 2019 level, and Brookings senior fellow Gian Maria Milesi-Ferretti argues that reining in fiscal deficits would be more effective than pressing other countries to import more U.S. goods.

Related Articles:

  • US Notches One of Its Biggest Annual Trade Gaps Since 1960 — The US goods and services trade deficit reached $901.5bn for the year, one of the largest in data going back to 1960. US companies increased imports of computers and accessories by nearly $145bn from the prior year, an acceleration in demand that reflects large-scale artificial intelligence investment.
  • The Dangerous Triumph Of Neo-Mercantilism — Neo-mercantilism is thriving in both China, which pairs infant-industry promotion with large trade surpluses, and in a United States focused on external deficits and protecting domestic markets. The author argues this will likely fracture the world economy, and that a new multilateral treaty could include a mechanism to combat large structural trade surpluses.
  • Sticking with What Works: Raising GDP Forecast on China’s Manufacturing Push — Stronger Chinese growth driven by export-oriented manufacturing may not produce net positive spillovers: for every 1pp export-driven boost to China's GDP, other economies may see a 0.1–0.3pp drag, with high-tech producers such as Europe and Japan facing greater pressure. Cheaper Chinese goods continue to exert a disinflationary impulse.

AI Summary. China's state-owned financial institutions added $100bn in foreign assets in one month and $70bn the next, absorbing dollars from the country's trade surplus as exporters sought to sell dollars for renminbi. The author argues dollar accumulation has shifted from China's foreign exchange agency to state banks, which either buy Treasuries or lend to investors who do.

Brad Setser Financial Times
Date Posted:

Talk of “dollar diversification” to the contrary, China bought a record $170B in US dollars and foreign securities in December and January. The lower visibility reflects a shift in the action from SAFE to the state owned banks as they sell renminbi.

Is China still accumulating dollar assets through its state banks?

Core argument: China's big state-owned financial institutions added an unprecedented $100bn to foreign assets in December and another $70bn in January, warehousing the dollars China Inc bought in the market.

The formal foreign exchange reserves of China’s central bank have not been rising rapidly, but do not be fooled. The dollars that China Inc bought in the market in the past few months are being warehoused by the country’s big state-owned financial institutions, which added an unprecedented $100bn to foreign assets in December and another $70bn in January. These huge, market-moving numbers showed that China’s massive trade surplus once again flows through the balance sheet of its core state financial institutions. Exporters now want to sell their dollars for renminbi to capture the renminbi’s expected appreciation. To offset that demand for renminbi, China’s state banks had to buy a ton of foreign exchange. [Though] the number of Treasuries that Chinese investors hold in US custodians has fallen sharply over time, that does not mean that China’s state — broadly defined — stopped accumulating dollars. Rather, dollar asset accumulation has shifted from China’s State Administration of Foreign Exchange agency (SAFE) ($3.4tn in foreign exchange reserves,an estimated 50-55% [of which are] in dollars) to the state banks (about $3tn in assets, with maybe 70% in dollars, according to my estimates from China’s own data for the commercial banks). If Chinese state investors are not directly buying Treasuries, they are lending money to other global investors who are. That is the only way the global flow of funds can add up.

Takeaways by Macro Roundup® AI

  1. China's big state-owned financial institutions added an unprecedented $100bn to foreign assets in December and another $70bn in January, warehousing the dollars China Inc bought in the market.
  2. Dollar asset accumulation has shifted from SAFE, which holds $3.4tn in foreign exchange reserves with an estimated 50–55% in dollars, to the state banks, with about $3tn in assets and perhaps 70% in dollars.
  3. Chinese exporters want to sell dollars for renminbi to capture the renminbi's expected appreciation, and China's state banks had to buy large amounts of foreign exchange to offset that demand.

Related Articles:

  • China’s Currency is Now Facing Substantial Appreciation Pressure — China's yuan faces appreciation pressure while Beijing maintains its commitment to currency stability, testing trading partners' willingness to absorb China's growing export surplus. If nominal appreciation is limited to a few percentage points, the real effective exchange rate stays at its depreciated level and China is likely to keep gaining global market share.
  • China’s Economy: Rightsizing 2025, Looking Ahead to 2026 — The authors argue China's official growth figures are inconsistent with ten consecutive quarters of deflation, estimating that real GDP is around 11% below official levels, with most of the adjustment in investment. That would imply a drop of roughly 9 percentage points in China's share of global investment, helping explain its export surge.
  • Sticking with What Works: Raising GDP Forecast on China’s Manufacturing Push — Stronger Chinese growth driven by export-oriented manufacturing may not produce net positive spillovers: for every 1pp export-driven boost to China's GDP, other economies may see a 0.1–0.3pp drag, with high-tech producers such as Europe and Japan facing greater pressure. Cheaper Chinese goods continue to exert a disinflationary impulse.

AI Summary. China froze exports of "dual-use" materials with both civilian and military applications, including rare earth magnets, to 20 Japanese companies and placed 20 more on a new watch list. The move deepens a dispute with Tokyo and shows Beijing using its dominance in critical supply chains to exert diplomatic pressure.

Ryan McMorrow, Joe Leahy, Harry Dempsey and Leo Lewis Financial Times
Date Posted:

China is freezing exports of “dual-use” materials to Japan, in an effort to “curb Japan’s ‘remilitarisation’ and nuclear ambitions.”

Is China using its critical supply chain dominance as diplomatic leverage?

Core argument: China's commerce ministry said it would freeze exports of "dual-use" materials, which have both civilian and military applications, to 20 Japanese companies, while adding another 20 groups to a new "watch list".

China has restricted exports of rare earth magnets and other critical materials to dozens of leading Japanese companies in an escalation of a dispute with Tokyo. China’s commerce ministry on Tuesday said it would freeze the flow of “dual-use” materials, which have applications to both civilian and military industries, to 20 Japanese companies. Another 20 groups are being added to a new “watch list”. The restrictions deepen a row with Japan and illustrate how Beijing is wielding its dominance in critical supply chains to exert diplomatic pressure. The global automotive industry is heavily dependent on China for rare-earth magnets. The announcement hit Japanese share prices, with MHI falling as much as 3%, Kawasaki Heavy shedding 4%, IHI Group losing 5.7% and NEC dropping more than 6%. Subaru fell 3.5%.

Takeaways by Macro Roundup® AI

  1. China's commerce ministry said it would freeze exports of "dual-use" materials, which have both civilian and military applications, to 20 Japanese companies, while adding another 20 groups to a new "watch list".
  2. Japanese shares fell on the announcement, with NEC dropping more than 6%, IHI Group losing 5.7%, Kawasaki Heavy shedding 4%, Subaru falling 3.5% and MHI down as much as 3%.
  3. The restrictions illustrate how Beijing is wielding its dominance in critical supply chains to exert diplomatic pressure, with the global automotive industry heavily dependent on China for rare-earth magnets.

Related Articles:

  • How Japan Built a Rare-Earth Supply Chain Without China — Japan has reduced its reliance on Chinese rare earths, with industry estimates putting China's share of Japanese rare-earth imports at roughly 60% to 70%, down from 90% or more. Japan's alternative chain runs from an Australian open-pit mine through a separation plant in Malaysia to Japanese magnet makers supplying automakers.
  • Japan Builds Up ‘Missile Archipelago’ Near Taiwan to Counter China — Japan is installing missile batteries, radar towers, ammunition storage and other combat facilities across the 160-island Ryukyu chain, whose southern end lies about 110km east of Taiwan. It is also deploying F-35 fighter jets and long-range missiles on Kyushu and expanding its amphibious rapid deployment force.
  • Japan PM Takaichi Vows To Boost Defense Spending To 2% of GDP by March — Japan plans to raise defense-related spending to 2% of gross domestic product in the fiscal year ending March 2026, two years ahead of the existing target date of March 2028. The government also set economic goals it says require no tax increases and prioritized domestic energy sources including nuclear power and Japanese-developed perovskite solar cells.