Edward Conard

Top Ten New York Times Bestselling Author

  • “…a must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a comprehensive explanation of the modern economy.” - Julian Robertson, Founder, Tiger Management
  • “Unintended Consequences offers deep and well-argued analyses on almost every issue.” - The New York Times
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “Unintended Consequences is far smarter and more thought-provoking than most economics written for the general public” - Greg Mankiw, Harvard University, Former Chairman of the Council of Economic Advisors
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “…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 full-throated defense of economic dynamism.” - The Wall Street Journal
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Friday, August 14, 2026

AI Summary. OpenAI cut its GPT-5.6 Luna prices by 80% and Anthropic launched Opus 5 at half the price of its Fable 5 model while cancelling a planned Sonnet 5 price increase. Headline token prices are not directly comparable, since more capable models can finish tasks using fewer tokens or attempts.

Jamie John and Clara Murray Financial Times
Date Posted:

Token prices for the leading US AI labs are falling. OpenAI cut the price of its mid-tier GPT-5.6 Luna by 80%, while Anthropic reduced the price of its Opus 5 by 50% relative to its frontier Fable 5 model.

Are leading AI developers cutting model prices as Chinese rivals gain ground?

Core argument: OpenAI cut GPT-5.6 Luna pricing from $1 to $0.20 per mn input tokens and from $6 to $1.20 per mn output tokens.

OpenAI cut the price of GPT-5.6 Luna from $1 to $0.20 per mn input tokens and from $6 to $1.20 per mn output tokens. Anthropic launched Opus 5 at $5 per mn input tokens and $25 per mn output tokens — half the price of its Fable 5 model. This week, the company called off a planned rise in prices for its Sonnet 5 model, which had been due to take effect from September. Headline token prices do not provide a straightforward comparison between AI models, however. More capable models can sometimes complete a task using fewer tokens or with fewer attempts, meaning a model that appears more expensive based on the headline price of tokens can ultimately cost less.

Takeaways by Macro Roundup® AI

  1. OpenAI cut GPT-5.6 Luna pricing from $1 to $0.20 per mn input tokens and from $6 to $1.20 per mn output tokens.
  2. Anthropic launched Opus 5 at $5 per mn input tokens and $25 per mn output tokens, half the price of its Fable 5 model, and called off a planned Sonnet 5 price rise due to take effect from September.
  3. Headline token prices do not offer a straightforward comparison across AI models, since more capable models can sometimes complete a task using fewer tokens or fewer attempts and ultimately cost less.

Related Articles:

  • Chart of the Day: Small Models are Closing the Gap to Frontier AI — Achieving 60% on MMLU, a test of academic reasoning, once required a 540bn-parameter model but was reached by a 3.8bn-parameter model two years later, a 142x reduction in parameters. Models small enough to run on phones or sensors shift inference from data centers to devices, weakening the case for paying cloud providers per token.
  • Who’s Afraid of Chinese Models? — Demand for frontier artificial intelligence models exceeds supply because computing capacity is scarce, allowing chip suppliers and their customers to resell compute at high margins. The author argues alarm over Chinese models is overblown, expecting the inference market to grow faster than training costs so labs can thrive at lower prices.
  • The AI Trade Is Losing One of Its Key Signals — An index blending artificial intelligence token prices and usage is ambiguous when it softens, since token prices have fallen more than 90% since 2023 while total spending has roughly doubled over the past year. Allianz Research measures a nearly 46% growth gap between artificial intelligence investment and sales, wider than the 32% divergence during the 2001 telecom bust.

AI Summary. Cash funded 85% of capital spending at Alphabet, Amazon, CoreWeave, Meta, Microsoft, Nebius, and Oracle in 2024, falling to 70% in 2025 and two-thirds in 2026, with 2026 capital spending guidance of $863bn exceeding their combined operating cash flow for the first time. About $282bn is funded by debt, equity, and leases.

Hannah Petrovic, William Gildea and Marija Gavrilov Exponential View
Date Posted:

An Exponential View analysis of capex guidance suggests the 7 largest hyperscalers will spend ~$863B in 2026, exceeding their combined operating cash flow for the first time. Two-thirds of the AI buildout is still funded by cash, down from 85% in 2024.

Are the largest tech firms funding AI capital spending with more debt?

Core argument: The Big Seven — Alphabet, Amazon, CoreWeave, Meta, Microsoft, Nebius, and Oracle — guided to $863 billion of capex for 2026, exceeding their combined operating cash flow for the first time.

In 2024, cash funded 85% of Alphabet, Amazon, CoreWeave, Meta, Microsoft, Nebius, and Oracle's (the Big Seven's) capex; it dropped to 70% in 2025 and is at two-thirds in 2026. Capex guidance of $863 billion for 2026 exceeds the Big Seven’s combined operating cash flow for the first time. Some $282 billion – around a third – is funded by debt, equity, and leases. Current capex reading is great news for anyone buying compute. A third of the buildout is about to come online – and its owners will be looking to earn with it.

Takeaways by Macro Roundup® AI

  1. The Big Seven — Alphabet, Amazon, CoreWeave, Meta, Microsoft, Nebius, and Oracle — guided to $863 billion of capex for 2026, exceeding their combined operating cash flow for the first time.
  2. Cash funded 85% of the Big Seven's capex in 2024, falling to 70% in 2025 and two-thirds in 2026, with about $282 billion financed by debt, equity, and leases.
  3. A third of the AI buildout is about to come online with its owners looking to earn from it, which the article casts as good news for buyers of compute.

Related Articles:

  • The State of the AI Economy — An analysis estimates the AI ecosystem generated $110bn in revenue over 12 months excluding double-counting, with the most recent month annualizing to a $175bn run rate. After separating AI-specific capital spending from the roughly $120bn hyperscalers already spent annually and depreciating compute over 6 years, revenues attributable to hyperscalers roughly cover depreciation.
  • The Summer I Turned Pretty — A warning sign in a boom cycle is when companies closest to final demand weaken while capital spending beneficiaries keep thriving, as when telecom operators' shares rolled over while communications equipment stocks kept rising. Semiconductor stock prices and free cash flow are soaring while large cloud providers' shares stagnate and their free cash flow plummets.
  • The Market Is Asking Questions — Markets are weighing whether the trillions committed upfront to AI data centers, chips, and power will generate returns above the cost of capital before those assets depreciate. With the buildout increasingly funded by debt rather than internal cash flow and borrowing spreads widening, financing costs could rise until new data-center spending no longer clears its return hurdle.

AI Summary. Goldman Sachs analysts tallied $1.5tn of lease commitments in the footnotes of large cloud computing companies' regulatory filings, of which $1tn had not yet commenced and so did not appear as conventional liabilities. The analysts noted this treatment can understate leverage and future liquidity needs as those obligations are eventually recognized.

Robin Wigglesworth Financial Times
Date Posted:

As of Q2, the hyperscalers have ~$2.5T lease and purchase commitments that are off balance sheet. GS warns “this treatment can understate leverage and future liquidity needs as these obligations are recognised and contractual payments come due.”

Do cloud companies' off-balance-sheet lease commitments understate their leverage?

Core argument: Goldman Sachs analysts counted $1.5tn of lease commitments in the footnotes of hyperscalers' regulatory filings, of which $1tn had not yet started and therefore did not appear in their accounts as conventional liabilities.

Goldman Sachs analysts scoured through the footnotes of the hyperscalers’ regulatory filings and counted $1.5tn of lease commitments, of which $1tn hadn’t started yet and therefore didn’t appear in their financial accounts as conventional liabilities. As those analysts obliquely noted: "From a credit perspective, this treatment can understate leverage and future liquidity needs as these obligations are eventually recognised and contractual payments come due." [We noted] an interesting titbit from an earlier Morgan Stanley report which also totaled up the purchase commitments of Alphabet, Microsoft, Amazon, Nvidia and Oracle. These are typically contractual obligations to buy chips, compute, electricity to power data centres, and other equipment, and came to nearly $1.5T.

Takeaways by Macro Roundup® AI

  1. Goldman Sachs analysts counted $1.5tn of lease commitments in the footnotes of hyperscalers' regulatory filings, of which $1tn had not yet started and therefore did not appear in their accounts as conventional liabilities.
  2. Goldman Sachs analysts noted that this treatment "can understate leverage and future liquidity needs as these obligations are eventually recognised and contractual payments come due.".
  3. An earlier Morgan Stanley report totaled the purchase commitments of Alphabet, Microsoft, Amazon, Nvidia and Oracle at nearly $1.5tn, typically contractual obligations to buy chips, compute, electricity to power data centres and other equipment.

Related Articles:

  • Big Tech AI Spending Spree Tops $1tn — Google, Amazon, Microsoft and Meta have together spent $1.1tn on capital investment since the start of the artificial intelligence boom in 2023, according to the companies' earnings reports. Executives at several of the firms told analysts the spending would continue to reduce free cash flow in coming quarters.
  • The Market Is Asking Questions — Markets are weighing whether the trillions committed upfront to AI data centers, chips, and power will generate returns above the cost of capital before those assets depreciate. With the buildout increasingly funded by debt rather than internal cash flow and borrowing spreads widening, financing costs could rise until new data-center spending no longer clears its return hurdle.
  • Big Tech Credit Risks Rise Sharply As AI Spending Soars — The cost of insuring debt against default for Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom, and Nvidia has reached record highs, with Oracle's five-year credit protection quoted at 215 basis points versus 144 earlier in the year. S&P Global Ratings cut Oracle to one notch above junk, citing an uncertain path to profitability amid massive AI investments.

AI Summary. The US government sold $25bn of 30-year bonds at a yield of 5.216%, the highest at such a sale since 2001, reflecting investors' demand for compensation to finance a growing deficit. The yield came in slightly above the prevailing market level before the bidding deadline, indicating demand modestly lagged expectations.

Greg Ritchie Bloomberg
Date Posted:

Though a $25B auction of 30-Year Treasurys drew “decent” demand, the yield rose to 5.216% — the highest 30-year yield at auction since August 2001.

Are investors demanding higher yields to finance the growing deficit?

Core argument: The US Treasury sold $25 billion of 30-year bonds at a yield of 5.216%, the highest since 2001, reflecting investors' demand for compensation to finance the nation's growing deficit.

The US government sold 30-year bonds at the highest interest rate in a quarter of a century, a testament to investors’ demand for compensation to finance the nation’s growing deficit. The yield at the $25 billion sale Thursday came in at 5.216%, the most since 2001, even as a drop in oil prices supported US debt in secondary-market trading. The sale was met with decent demand. The yield at Thursday’s 30-year sale was a little above the prevailing level seen in the market before the 1 p.m. bidding deadline in New York—a sign that demand slightly lagged expectations.

Takeaways by Macro Roundup® AI

  1. The US Treasury sold $25 billion of 30-year bonds at a yield of 5.216%, the highest since 2001, reflecting investors' demand for compensation to finance the nation's growing deficit.
  2. The 30-year auction yield came in slightly above the level prevailing in the market before the 1 p.m. New York bidding deadline, a sign that demand slightly lagged expectations.
  3. A drop in oil prices supported US debt in secondary-market trading on the day of the 30-year sale, which was met with decent demand.

Related Articles:

  • If Anyone Needs an Intervention, It’s the BOJ — A Federal Reserve press conference that offered no forward guidance left the market's projected path for interest rates barely changed, since investors form their own view from the data. But the 30-year inflation-protected Treasury yield has only been higher for a few days at the worst of the 2008 crisis, suggesting possible damage to the Federal Reserve's credibility.
  • US 30-Year Yield Soars to Highest Since ‘07 After Fed Stands Pat — The 30-year US Treasury yield rose more than 10 basis points to a 19-year high after the Federal Reserve left rates unchanged, while two-year yields fell to 4.23%, 10-year yields rose to 4.66%, and the dollar dropped. Interest-rate swaps priced roughly 60% odds of a rate increase in September.
  • The Dangerous Brew That’s Rattling Bond Markets — U.S. inflation has remained above the Federal Reserve's 2% target since 2022 while government borrowing runs high and companies borrow heavily to fund AI build-out. Even so, the 30-year Treasury yield reached a 19-year high of 5.18% while the 10-year yield, at 4.67%, sat below its October 2023 level.

AI Summary. The gap between crude oil prices and the prices of refined products has widened by about $35/barrel, because much of the world's refining capacity is either trapped behind the Strait of Hormuz or knocked offline by drone strikes. That refining shortage holds crude prices down, since buyers won't pay high prices for oil they cannot refine.

Paul Krugman Krugman Wonks Out
Date Posted:

Krugman asks why crude is “only” ~$25 above its prewar level. Much of the world’s refining capacity is trapped behind Hormuz or offline, raising the “crack spread” – the premium on refined products like diesel – while depressing the crude oil price.

Does a shortage of refining capacity hold crude oil prices down?

Core argument: The crack spread — the gap between crude prices and the prices of refined products — has risen about $35/barrel since the eve of the war.

The difference between the price of a barrel of crude and the price of the products refined from that barrel — the “crack spread” — has exploded, rising about $35/barrel since the eve of the war. Why has the crack spread widened so much? The main answer is that a lot of the world’s refining capacity is either trapped inside the Strait or offline as a result of Ukraine’s drone campaign. It’s not all about Iran. The shortage of refining capacity has held crude prices down, as buyers won’t pay extremely high prices for crude they can’t refine. Or to put it differently but equivalently, the cutoff of oil shipments through the Strait of Hormuz, in effect, required a large rise in global prices [of petroleum products] to ration demand, but much of that rationing has taken place through a rise in the crack spread rather than a rise in crude oil prices.

Takeaways by Macro Roundup® AI

  1. The crack spread — the gap between crude prices and the prices of refined products — has risen about $35/barrel since the eve of the war.
  2. Much of the world's refining capacity is either trapped inside the Strait of Hormuz or offline because of Ukraine's drone campaign, a shortage that holds crude prices down because buyers won't pay extremely high prices for crude they can't refine.
  3. The cutoff of oil shipments through the Strait of Hormuz required a large rise in global petroleum product prices to ration demand, and much of that rationing has occurred through a wider crack spread rather than higher crude prices.

Related Articles:

  • For the Oil Market, the Strait of Hormuz Isn’t Closed — Tanker tracking data identify at least 5m barrels a day of oil still transiting the Strait of Hormuz, and allowing for vessels crossing undetected the true figure could reach 7m–9m. Growing oil spills visible on satellite imagery indicate attacks on tankers alongside efforts to keep the waterway open.
  • How China Became The World’s Great Oil Power — China cut crude oil imports by half, or 5.5m barrels a day, without a corresponding economic slowdown, a decline experts estimate shaved $30 or more off the global Brent benchmark price. This ability to switch oil demand on and off lets the world's largest oil importer move prices much as OPEC does through output control.
  • U.S. Economy Less Vulnerable To Geopolitical Oil Price Shocks Than In The Past — The United States became a net oil exporter in late 2019, and oil and oil product spending fell from near 8% of GDP in 1980 to 3% in 2024. The authors estimate a 15% global oil supply disruption would now cut annualized U.S. real GDP growth by 0.3pp, versus 1.7pp in the rest of the world.

AI Summary. Global fertility is likely below the replacement level needed to keep population constant in the long run, though population momentum from large cohorts of women now having children means world population will keep growing for roughly another 30 years. Unless trends change, it is projected to peak near 9bn around 2056 and then decline.

Jesús Fernández-Villaverde and Patrick Norrick University of Pennsylvania
Date Posted:
Is Database:
Database
Is Important:
Important

The UN appears to systematically overestimate births; e.g. 33 of 37 countries with high-quality statistics registered fewer births in 2024 than the UN had forecast. Fernández-Villaverde and Norrick infer humanity is below replacement fertility in 2026.

Will world population keep growing even as fertility falls below replacement?

Core argument: As of 2026, humanity is likely to be below replacement fertility, with fewer births than needed to keep population constant in the long run.

In Table A1 we compare the World Population Prospects (WPP) estimates of births in 2022-2023 with the official numbers reported by several countries. [A2 shows the full sample with the deviations.] As of 2026, humanity is likely to be below the replacement fertility level: we are having fewer births than we need to keep population constant in the long run. This astonishing fact does not mean that population has stopped growing. Because of momentum effects (the large cohorts of women born two or three decades ago are having their children now, and their own parents have not died yet), world population will keep growing for another 30 years or so. But unless trends change, it will peak at roughly 9 billion around 2056 and then start falling, first slowly, then fast.

Takeaways by Macro Roundup® AI

  1. As of 2026, humanity is likely to be below replacement fertility, with fewer births than needed to keep population constant in the long run.
  2. World population is set to keep growing for roughly another 30 years because of momentum effects, as large cohorts of women born two or three decades ago have their children while their own parents are still alive.
  3. Unless trends change, world population will peak at roughly 9 billion around 2056 and then start falling, first slowly, then fast.

Related Articles:

  • The Demographic Future of Humanity: Facts and Consequences — The authors estimate the world total fertility rate is around 2.17, below the global replacement rate of roughly 2.21, meaning humanity is already below replacement, with fertility and births falling faster than anticipated. They project middle-income countries' fertility may fall below that of rich countries outside East Asia, with emigration compounding population decline.
  • Depopulation Globally and in the Asia-Pacific: The Shape of Things to Come — Population aging and fertility below replacement levels are connected to a revolution in family structure in which families grow smaller, fewer people marry, and voluntary childlessness becomes common across countries. As a result, families become less able to bear the demands that may be placed on them.
  • The Wealth of Working Nations — Because of population aging, growth in GDP per capita and GDP per working-age adult have diverged across advanced economies; from 1998 to 2019 Japan's accumulated growth per working-age adult was 31.9% versus 29.5% in the US. The authors find a standard neoclassical growth model calibrated to each country's working-age population tracks output per working-age adult in most economies.

Monday, August 17, 2026

AI Summary. Employment for workers aged 22–25 in the occupations most exposed to artificial intelligence has declined roughly 18pp relative to less-exposed peers, with no comparable decline in other age groups. The authors find this operates mainly through reduced hiring, concentrated where AI substitutes for human tasks rather than complements them.

Erik Brynjolfsson, Bharat Chandar and Ruyu Chen Stanford University
Date Posted:

If employment of the most AI-exposed workers aged 22–25 had kept pace with that of their less-exposed peers since January 2018, their employment would be 15% higher than at the July 2025 data vintage observed in our previous study, and 19% higher as of June 2026.

Is AI reducing hiring for young workers in exposed occupations?

Core argument: Event study estimates for workers aged 22–25 show employment in the most AI-exposed occupation quintile declining roughly 18pp relative to the least-exposed quintile by mid-2026, with no controls and under four control sets.

Figure 4 plots occupation event study estimates for the most exposed quintile (relative to quintile 1) for workers aged 22–25, tracing the long-difference coefficient as the endpoint of the difference is rolled forward month by month, with no controls and under four control sets: occupational interest rate exposure, education (college share), a work-from-home measure, and all three jointly. The relative decline grows to roughly 18pp by mid-2026. Other age groups show no comparable decline. This divergence, documented in prior versions of this study, has continued to widen; the shortfall was 15% below where it would have been had it kept pace with that of their less-exposed peers at the July 2025 data vintage, and 19% as of June 2026. It operates primarily through reduced hiring rather than increased separations, and declines are concentrated in occupations where AI usage substitutes for human tasks; complementary usage is associated with flat or rising employment.

Takeaways by Macro Roundup® AI

  1. Event study estimates for workers aged 22–25 show employment in the most AI-exposed occupation quintile declining roughly 18pp relative to the least-exposed quintile by mid-2026, with no controls and under four control sets.
  2. The shortfall for workers aged 22–25 relative to their less-exposed peers widened from 15% at the July 2025 data vintage to 19% as of June 2026, while other age groups show no comparable decline.
  3. The decline operates primarily through reduced hiring rather than increased separations and is concentrated in occupations where AI usage substitutes for human tasks, while complementary AI usage is associated with flat or rising employment.

Related Articles:

  • 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.
  • The Impact of AI on the U.S. Labor Market — Using observed AI usage data across 321 U.S. occupations, the authors estimate that highly exposed occupations saw a 6.7% decline in real wage growth with no detectable employment effects. The estimated declines are concentrated among the lowest earners, with service workers down 24.3% and the bottom wage quartile down 10.7%, while top earners show no significant effect.
  • You’re (Not) Hired: Artificial Intelligence and Early Career Hiring In The Quarterly Workforce Indicators — Employment among workers aged 22–24 in the most artificial-intelligence-exposed quintile of industry-state employers fell 15.2%, while other exposure quintiles were flat or posted modest gains. The decline is not generally observed among older workers; the 25-34 group's drop was less than one-third as large in percentage terms.

AI Summary. Artificial intelligence models are converging in capability and price, making switching easier and margins harder to defend, with US labs cutting prices on lower- and mid-tier offerings while keeping frontier models expensive. An 80% annual price decline requires 400% unit growth just to hold revenue flat, straining firms financing enormous capital spending.

Paul Kedrosky Applied Complexity
Date Posted:

US labs have “conceded commoditization” for models behind the frontier, while still trying to “preserve scarcity rents” for their frontier models. Kedrosky warns “this market-ceding tactic has failed in most markets historically, especially when competing with China.”

Can AI firms defend margins as model prices converge?

Core argument: An 80% year-over-year decline in AI model prices requires 400% unit growth simply to hold revenue flat, and considerably more to generate growth.

Good-enough models are converging in capability and price. That makes model switching easier and durable margins harder to defend. US labs are cutting lower- and mid-tier offerings while keeping their best models expensive. They are conceding commoditization below the frontier while trying to preserve scarcity rents at the top. This market-ceding tactic has failed in most markets historically, especially when competing with China. This is irreconcilable with trillion-dollar valuations and high & growing debt loads. OpenAI and Anthropic must finance enormous capital spending while competition pushes the models likely to generate the most volume toward commodity pricing. An 80% year-over-year price decline requires 400% unit growth just to stand still, and much more to deliver overall growth.

Takeaways by Macro Roundup® AI

  1. An 80% year-over-year decline in AI model prices requires 400% unit growth simply to hold revenue flat, and considerably more to generate growth.
  2. US AI labs are cutting prices on lower- and mid-tier models while keeping frontier models expensive, conceding commoditization below the frontier while trying to preserve scarcity rents at the top.
  3. OpenAI and Anthropic must finance enormous capital spending as competition pushes the models likely to generate the most volume toward commodity pricing, a position irreconcilable with trillion-dollar valuations and high and growing debt loads.

Related Articles:

  • OpenAI and Anthropic In Price War as Chinese AI Rivals Gain Ground — OpenAI cut its GPT-5.6 Luna prices by 80% and Anthropic launched Opus 5 at half the price of its Fable 5 model while cancelling a planned Sonnet 5 price increase. Headline token prices are not directly comparable, since more capable models can finish tasks using fewer tokens or attempts.
  • Who’s Afraid of Chinese Models? — Demand for frontier artificial intelligence models exceeds supply because computing capacity is scarce, allowing chip suppliers and their customers to resell compute at high margins. The author argues alarm over Chinese models is overblown, expecting the inference market to grow faster than training costs so labs can thrive at lower prices.
  • The AI Trade Is Losing One of Its Key Signals — An index blending artificial intelligence token prices and usage is ambiguous when it softens, since token prices have fallen more than 90% since 2023 while total spending has roughly doubled over the past year. Allianz Research measures a nearly 46% growth gap between artificial intelligence investment and sales, wider than the 32% divergence during the 2001 telecom bust.

AI Summary. Nine large technology companies disclosed roughly $3tn of off-balance-sheet commitments, mostly tied to artificial intelligence, in the footnotes of their most recent securities filings. These obligations are growing faster than traditional capital spending, which totaled about $600bn over the companies' latest reported year, and were roughly 3x what they owe on outstanding leases and long-term borrowings.

Peter Rudegeair and Peter Santilli Wall Street Journal
Date Posted:
Is Database:
Database

A WSJ analysis finds 9 firms involved in the data center buildout have ~$3T in off-balance-sheet commitments largely tied to AI infrastructure. The growth in such obligations has outpaced the firms’ capex growth over the last year.

Do off-balance-sheet commitments understate how much Big Tech spends on AI?

Core argument: Nine top tech companies carried roughly $3 trillion of off-balance-sheet commitments mostly tied to AI, according to a Wall Street Journal analysis of footnotes in their most recent securities filings.

Nine top tech companies had some $3 trillion of off-balance-sheet commitments mostly related to AI, according to a Wall Street Journal analysis of footnotes in their most recent securities filings. Those obligations are growing faster than traditional “capex,” which totaled about $600 billion over the past year they reported, and were about triple what the companies owe under their outstanding leases and long-term borrowings.

Takeaways by Macro Roundup® AI

  1. Nine top tech companies carried roughly $3 trillion of off-balance-sheet commitments mostly tied to AI, according to a Wall Street Journal analysis of footnotes in their most recent securities filings.
  2. The off-balance-sheet obligations were about triple what the nine companies owe under their outstanding leases and long-term borrowings, according to the Wall Street Journal analysis.
  3. The commitments are growing faster than traditional capital expenditures, which totaled about $600 billion over the past year the nine companies reported.

Related Articles:

  • The Hyperscalers’ Exploding ‘Purchase Commitments’ Reach $1.5tn — Goldman Sachs analysts tallied $1.5tn of lease commitments in the footnotes of large cloud computing companies' regulatory filings, of which $1tn had not yet commenced and so did not appear as conventional liabilities. The analysts noted this treatment can understate leverage and future liquidity needs as those obligations are eventually recognized.
  • The Market Is Asking Questions — Markets are weighing whether the trillions committed upfront to AI data centers, chips, and power will generate returns above the cost of capital before those assets depreciate. With the buildout increasingly funded by debt rather than internal cash flow and borrowing spreads widening, financing costs could rise until new data-center spending no longer clears its return hurdle.
  • Big Tech Credit Risks Rise Sharply As AI Spending Soars — The cost of insuring debt against default for Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom, and Nvidia has reached record highs, with Oracle's five-year credit protection quoted at 215 basis points versus 144 earlier in the year. S&P Global Ratings cut Oracle to one notch above junk, citing an uncertain path to profitability amid massive AI investments.

AI Summary. Investment-grade companies have sold nearly $1.5tn of bonds, up 36% from a year earlier, and some investors have sold US Treasuries to buy higher-yielding corporate debt. Bank of America economists estimate the surge in corporate-debt sales, along with more mortgage-backed securities issuance, pushed 10-year rates up by about 0.3 percentage point.

Davide Barbuscia, Ye Xie, and Michael MacKenzie Bloomberg
Date Posted:
Is Database:
Database

US investment-grade debt issuance is up 36% y/y to ~$1.5T. A BofA analysis argues the surge in this debt, alongside increased MBS issuance, has driven up the 10-year Treasury yield by ~.3pp this year.

Is a surge in corporate bond issuance raising Treasury yields?

Core argument: Investment-grade companies have sold nearly $1.5 trillion of bonds in 2026, a 36% jump from a year earlier, putting them on pace to eclipse the record set in 2020, when businesses rushed to seize on near-zero interest rates.

[In 2026] investment-grade companies have sold nearly $1.5 trillion of bonds, a 36% jump from a year earlier, putting them on pace to eclipse the record from 2020, when businesses were rushing to seize on near-zero interest rates. As the slew of longer-dated bonds keeps hitting the market, some investors have sold US Treasuries. That freed up cash to buy higher-yielding debt from immensely profitable companies like Alphabet, whose recent 30-year debt was issued at a yield of nearly 6.4%, 1.15 percentage points more than comparable Treasuries. Quantifying the precise impact on interest rates is difficult and estimates vary. Bank of America economists said the AI borrowing is “potentially crowding out long-end Treasury demand” and has played a major role in the rise of bond yields. They estimated that the surge in corporate-debt sales — along with a rise in issuance of mortgage-backed securities — pushed up 10-year rates by about 0.3 percentage point this year.

Takeaways by Macro Roundup® AI

  1. Investment-grade companies have sold nearly $1.5 trillion of bonds in 2026, a 36% jump from a year earlier, putting them on pace to eclipse the record set in 2020, when businesses rushed to seize on near-zero interest rates.
  2. Bank of America economists estimated that the surge in corporate-debt sales, along with a rise in mortgage-backed securities issuance, pushed 10-year rates up by about 0.3 percentage point this year, and said AI borrowing is "potentially crowding out long-end Treasury demand.".
  3. Alphabet issued 30-year debt at a yield of nearly 6.4%, 1.15 percentage points above comparable Treasuries, as some investors sold US Treasuries to free up cash for higher-yielding corporate debt.

Related Articles:

  • Costliest US Bond Sale Since ‘01 Is Investor Warning to Bessent — The US government sold $25bn of 30-year bonds at a yield of 5.216%, the highest at such a sale since 2001, reflecting investors' demand for compensation to finance a growing deficit. The yield came in slightly above the prevailing market level before the bidding deadline, indicating demand modestly lagged expectations.
  • U.S. Treasury Investors Are Long in AI — The authors find U.S. government debt has a large positive productivity beta, because the tax code makes federal revenue highly sensitive to productivity growth while spending commitments change little, leaving bondholders implicitly long productivity growth. They argue greater market confidence in sustained productivity gains from AI could substantially reprice government debt, lowering nominal yields.
  • The Market Is Asking Questions — Markets are weighing whether the trillions committed upfront to AI data centers, chips, and power will generate returns above the cost of capital before those assets depreciate. With the buildout increasingly funded by debt rather than internal cash flow and borrowing spreads widening, financing costs could rise until new data-center spending no longer clears its return hurdle.

AI Summary. A proxy for strain at the long end of government bond markets — the gap between the 10-year yield expected in 10 years' time and today's 10-year yield, scored against its own history — shows Japan as an outlier among the G10 economies.

Robin Brooks Robin Brooks Substack
Date Posted:

Citing the difference between Japan’s artificially depressed 10-year yield and its 10y10y fwd rate, Brooks notes “Japan’s ‘shadow yield’ is much higher than observed yields, which means Japan is – de facto – in a debt crisis.”

Which advanced economy shows the most unusual long-end bond market strain?

Core argument: Japan stands out among G10 economies on a measure of long-end yield curve distortion, defined as the gap between the 10y10y forward rate and the 10-year yield, demeaned and divided by its historical standard deviation.

The blue lines are my proxy for how “broken” yield curves are at the long end. This is the difference between the 10y10y forward [which is what markets price for the 10-year yield in 10 years’ time] and 10-year yield. I demean this difference and divide it by its historical standard deviation. The resulting z-score measures how unusual the slope of the yield curve is at the long end relative to history. The black line in each chart is the median across all G10 z-scores. The gray shaded area is a two standard deviation confidence interval around the black line. If you’re outside this area, something very worrying is going on. Japan sticks out like a sore thumb on this metric.

Takeaways by Macro Roundup® AI

  1. Japan stands out among G10 economies on a measure of long-end yield curve distortion, defined as the gap between the 10y10y forward rate and the 10-year yield, demeaned and divided by its historical standard deviation.
  2. The z-score framework compares each country's long-end curve slope to its own history, with a two standard deviation confidence interval around the median G10 z-score marking the band beyond which conditions are described as very worrying.

Related Articles:

  • Shadow Government Bond Yields in the G10 — The author estimates "shadow" 30-year government bond yields for Germany, Spain, Italy, Greece and Japan — what long-term yields would be if central banks withdrew support such as Japanese government bond purchases and the European Central Bank's backstop facilities. These estimates assume no outright central bank selling, so they represent a lower bound on how high yields could go.
  • Global Debt Report 2026 — Gross central government borrowing across OECD countries rose 7% to a record $17tn, about 23% of GDP, and is projected to reach roughly $18tn, or 24% of GDP. The United States and Japan account for close to 80% of the area's refinancing requirements, with the US share rising from 57% to 70%.
  • Costliest US Bond Sale Since ‘01 Is Investor Warning to Bessent — The US government sold $25bn of 30-year bonds at a yield of 5.216%, the highest at such a sale since 2001, reflecting investors' demand for compensation to finance a growing deficit. The yield came in slightly above the prevailing market level before the bidding deadline, indicating demand modestly lagged expectations.

AI Summary. Tariff refunds are boosting US corporate earnings and economic growth, with the Atlanta Federal Reserve's GDPNow tracker pointing to 4.3% growth this quarter, of which roughly 0.2 percentage points is estimated to come from tariff refunds.

Torsten Sløk Apollo
Date Posted:

Net tariff receipts turned negative in May as refunds outpaced custom duties. The Atlanta Fed projects GDP growth of 4.3% at an annual rate in Q3, of which Sløk estimates ~.2pp will be driven by tariff refunds.

Are tariff refunds adding to corporate earnings and GDP growth?

Core argument: The Atlanta Fed's GDPNow points to 4.3% US growth this quarter, of which roughly 0.2 percentage points are estimated to come from tariff refunds.

The tailwinds behind the US economy are not just AI spending, the industrial renaissance and the One Big Beautiful Bill. Tariff refunds have now joined the list. Not only are tariff refunds boosting corporate earnings, they are also boosting GDP growth. The Atlanta Fed's GDPNow currently points to 4.3% growth this quarter, of which we estimate roughly 0.2 percentage points come from tariff refunds. The bottom line is that the US economy continues to be supported by a growing set of tailwinds. Rates will stay higher for longer.

Takeaways by Macro Roundup® AI

  1. The Atlanta Fed's GDPNow points to 4.3% US growth this quarter, of which roughly 0.2 percentage points are estimated to come from tariff refunds.
  2. Tariff refunds are boosting both corporate earnings and GDP growth, joining AI spending, the industrial renaissance and the One Big Beautiful Bill as tailwinds behind the US economy.
  3. With the US economy supported by a growing set of tailwinds, rates are expected to stay higher for longer.

Related Articles:

  • The State of U.S. Tariffs July 24, 2026 — The average statutory US tariff rate stands at 11.1% and, under current law including scheduled increases, is projected to reach 11.8% by year-end. The authors estimate current-law tariffs will raise consumer prices by about 0.7% and generate roughly $1.9tn over ten years, with revenue somewhat lower after accounting for tariffs' negative effect on economic output.
  • The AI Capex Boom Is Building Twice as Fast as the Housing Boom — Data-center capital spending is projected to rise from 1.4% of GDP to 3.1% within two years, about 0.85 percentage points a year, roughly twice the pace of the housing boom's fastest stretch (0.5 points a year) and far above telecom fiber's (about 0.15). In level, the buildout exceeds twice the fiber peak but remains under half the housing peak.
  • The AI Investment Race — Technology-driven investment booms have historically ended in sharp corrections with wider economic fallout, and the current artificial intelligence build-out is on track to outgrow every previous episode within three years measured from its pre-boom trough. The expansion is increasingly funded by borrowing through bonds, special purpose vehicles and non-bank lenders, alongside circular financing arrangements that create interconnectedness among firms.