Edward Conard

Top Ten New York Times Bestselling Author

  • “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
  • “…a comprehensive explanation of the modern economy.” - Julian Robertson, Founder, Tiger Management
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “Unintended Consequences offers deep and well-argued analyses on almost every issue.” - The New York Times
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “…serious thinking for serious thinkers. …a thought-provoking blueprint for growing middle- and working-class incomes.” - Mitt Romney, former Governor of Massachusetts
  • “A full-throated defense of economic dynamism.” - The Wall Street Journal
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…a fresh argument for the productive value of inequality.” - David Autor, Professor of Economics, Massachusetts Institute of Technology
  • “…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 must-read for serious students of economic policy.” - Glenn Hubbard, Dean, Columbia Business School, and former Chairman of the Council of Economic Advisers
  • “…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
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Monday, August 18, 2025

AI Summary. Artificial intelligence gains have come from a formula of more data and more computing power, but companies have exhausted freely available internet training data and are now striking data-sharing deals with publishers and copyright holders. Training and running large models is energy-intensive, requiring vastly more Nvidia chips with each generation.

Melissa Heikkilä, Tim Bradshaw, Cristina Criddle and George Hammond Financial Times
Date Posted:
Is Database:
Database
Is Important:
Important

The rate of performance improvement of large language models may be slowing, as models face both data availability and compute constraints. GPT-5’s launch underwhelmed many in the sector.

Are data and computing limits slowing AI progress?

Core argument: GPT-4 was trained on several thousand Nvidia chips in 2022, while estimates suggest GPT-5 was trained on hundreds of thousands of Nvidia's next-generation processors, with more powerful chips on the way.

For the past five years, companies such as OpenAI and Anthropic have been able to show consistent gains in the performance of their systems by using a simple formula: more data and more computing power equals bigger, better models. Many AI leaders still believe that these “scaling laws” can continue to hold for years to come. But the approach is starting to reach the limits of the available resources. First, AI companies have sucked up all of the available free training data on the internet. They are now seeking more fuel for their models by making data-sharing deals with publishers and copyright holders, but it is unclear if that is enough to push the frontier forward. AI labs are also constrained by computing power. It is very energy-intensive to train and run large AI models. Back in 2022, GPT-4 was trained on several thousand of Nvidia’s chips. Estimates suggest that GPT-5 was trained on hundreds of thousands of Nvidia’s next-generation processors, with even more powerful chips on the way.

Takeaways by Macro Roundup® AI

  1. GPT-4 was trained on several thousand Nvidia chips in 2022, while estimates suggest GPT-5 was trained on hundreds of thousands of Nvidia's next-generation processors, with more powerful chips on the way.
  2. AI companies have exhausted the freely available training data on the internet and are now striking data-sharing deals with publishers and copyright holders, though it is unclear whether that supply is enough to advance the frontier.
  3. The scaling formula of more data plus more computing power that delivered five years of AI performance gains is reaching the limits of available resources, though many AI leaders still expect scaling laws to hold for years.

Related Articles:

  • The Kanye/Data Center Crossover — Data center capital spending differs from fiber and railroad investment because most of the cost sits in processors that are replaced every 3-4 years rather than in long-lived shells, power, and cooling connections. That short replacement cycle requires returns high enough to exceed a risk-adjusted cost of capital of 12-14% before the hardware must be replaced.
  • ‘Absolutely Immense’: The Companies On The Hook For The $3tn AI Building Boom — Google, Amazon, Microsoft and Meta are projected to spend more than $400bn on data centres in 2026, following more than $350bn this year. Morgan Stanley analysts estimate global data centre spending will approach $3tn through 2029, of which about $1.4tn comes from Big Tech capital expenditure, leaving roughly $1.5tn to be financed by investors and developers.
  • The AI Boom’s Hidden Risk to the Economy — Net income at Alphabet, Amazon, Meta and Microsoft rose 73% to $91bn in a quarter two years on, while their combined free cash flow fell 30% to $40bn, as heavy AI investment absorbs cash these firms once recycled into financial markets. That spare cash previously helped hold down long-term interest rates despite large federal deficits.

AI Summary. Benchmark scores show only modest capability gains in the latest large language model releases compared with the larger jump seen two generations earlier, alongside a broader convergence in performance across competing AI models. Commentators attribute this to inherent limits of the technology and the exhaustion of new training data, raising doubts that the spending on AI hardware is justified.

Joseph Wang Fed Guy Blog
Date Posted:
Is Database:
Database

Following tepid reactions to the release of GTP-5, Joe Wang observes, “It is looking more like companies are spending hundreds of billions on rapidly depreciating GPUs that produce a commoditized product that most clients only modestly benefit from.”

Are AI capability gains slowing enough to question hardware spending?

Core argument: GPT-5 users widely expressed disappointment with the release, which seemed in some ways a step back, and benchmarks show only modest capability gains since the significant improvement delivered by version 4 two years earlier.

GPT-5 users widely expressed disappointment in the capabilities of the new release, which seemed in some ways a step back. This sentiment is reflected in benchmarks that show a modest improvement in capabilities since the significant improvement in version 4 released two years ago. In addition, the benchmarks suggest a broader convergence in the capabilities of AI models. Commentary suggests this could be due to inherent limitations in the LLM technology and exhaustion of new training data. AI is fascinating technology, but it may not justify the enormous sums spent in its pursuit. It is looking more like companies are spending hundreds of billions on rapidly depreciating GPUs that produce a commoditized product that most clients only modestly benefit from. The entire macro landscape would look very different without the support of the AI boom.

Takeaways by Macro Roundup® AI

  1. GPT-5 users widely expressed disappointment with the release, which seemed in some ways a step back, and benchmarks show only modest capability gains since the significant improvement delivered by version 4 two years earlier.
  2. Benchmarks suggest a broader convergence in AI model capabilities, with commentary attributing this to inherent limitations of large language model technology and exhaustion of new training data.
  3. Companies appear to be spending hundreds of billions on rapidly depreciating GPUs producing a commoditized product that most clients only modestly benefit from, and the entire macro landscape would look very different without the support of the AI boom.

Related Articles:

  • Honey, AI Capex is Eating the Economy — Capital spending on AI data centers is around 20% of railroads' peak spending as a share of GDP and is rising quickly, having already surpassed the dot-com era peak in telecom investment. Unlike long-lived rail infrastructure, data centers are short-lived assets requiring frequent hardware replacement, and the capital is being reallocated from other sectors.
  • Back to Our Regularly Scheduled Programming — The four largest US hyperscalers spent $440bn on capital expenditure and research in 2024, with $596bn projected for 2025, a 35% increase, leaving large gaps between AI spending and AI revenues. These firms account for more than half of all capital expenditure growth in the US equity market.
  • AI’s $600B Question — Comparing spending on AI chips to historical railroad construction overlooks that high-fixed-cost, low-marginal-cost businesses without monopoly or oligopoly power tend to see prices competed down to marginal cost, and that semiconductors continually improve in cost versus performance, unlike physical infrastructure. Speculative frenzies accompany technology, but the payoff will take a long time.

AI Summary. Investment-grade corporate borrowing costs stand 0.75 percentage points above benchmark government bond yields in the US and 0.76 points in the Eurozone, the narrowest spreads since 1998 and 2018 respectively. US companies issued $910bn of highly rated debt in the first half of the year, the second-largest first-half total on record.

Ian Smith and Will Schmitt Financial Times
Date Posted:

Borrowing costs for investment grade firms in the US and Eurozone are .75 and .76pp above the benchmark government bond yields, respectively, the lowest levels since 1998 and 2018, respectively.

Are corporate credit spreads at their narrowest levels in decades?

Core argument: US investment-grade credit spreads stood at 0.75 percentage points over benchmark government bond yields, their tightest since 1998, while Eurozone spreads reached 0.76 points, the narrowest since 2018, according to Ice BofA data.

The cost of borrowing for investment-grade companies in US and Eurozone credit markets is 0.75 and 0.76 percentage points above benchmark government bond yields, respectively, according to Ice BofA data. This took spreads in the two markets — a proxy for the risk of default — on Friday to their lowest levels since 1998 and 2018, respectively. “The credit markets are again weirdly confident that the global economy is going to be just fine, and I don’t know why they are as confident as they are,” said Ben Inker, co-head of asset allocation at asset manager GMO. “You just don’t seem to be getting paid that much for taking risk.” US companies have taken advantage of tight spreads to borrow heavily this year, issuing the second-most highly rated debt in the first half of a year on record at $910bn — even with April’s trade war lull.

Takeaways by Macro Roundup® AI

  1. US investment-grade credit spreads stood at 0.75 percentage points over benchmark government bond yields, their tightest since 1998, while Eurozone spreads reached 0.76 points, the narrowest since 2018, according to Ice BofA data.
  2. Ben Inker, co-head of asset allocation at GMO, said credit markets are "weirdly confident that the global economy is going to be just fine" and that "you just don't seem to be getting paid that much for taking risk.".
  3. US companies issued $910bn of highly rated debt in the first half of the year, the second-most on record for that period, even with April's trade war lull.

Related Articles:

  • Has Private Credit Broken the High-Yield Spread Signal? — The high-yield bond market has stayed at about $1.4tn while private credit assets have grown to as much as $2tn, and the highest-rated BB portion of the high-yield index has risen from 41% to 51%. Private markets have shifted riskier borrowers out of high yield, making that index safer.
  • The Growing Role of Private Credit: The Outlook for Corporate Finance — Banks provide a smaller share of corporate credit than before the financial crisis, with bank lending falling from about 27% of total non-financial corporate debt to about 20%.
  • Private Credit: Characteristics and Risks — Private credit loans are negotiated bilaterally outside traditional regulatory requirements and are illiquid because no secondary market exists, leaving investors expected to hold them to maturity or face steep losses if they need to exit. Scarce data on this growing form of nonbank lending makes assessing its risks difficult.

AI Summary. US interest rates have exceeded euro-area rates for well over a decade, with the federal funds rate target range at 4.25%–4.5% versus the European Central Bank's 2% deposit rate. The author argues this gap largely reflects stronger underlying US trend growth relative to Europe's underperforming economy.

Jesper Rangvid Rangvid's Blog
Date Posted:
Is Database:
Database
Is Important:
Important

Due to stronger growth, the underlying real short-term interest rate in the US is ~ 1pp higher than in the euro area. US inflation is also ~1pp higher. Jesper Rangvid concludes the current FFR is appropriate.

Do higher US interest rates than Europe's reflect stronger US growth?

Core argument: The federal funds target range stands at 4.25%–4.5% while the ECB's deposit facility rate is 2%, leaving US policy rates more than two percentage points above the euro area's.

The target range for the federal funds rate currently stands at 4.25%–4.5%. At present, the ECB’s key policy rate, the deposit facility rate, is 2%, more than two percentage points below the fed funds rate.The key point I want to make in this analysis is that US interest rates have been consistently higher than those in the euro area for well over a decade, dating back to 2012, and before that were usually higher too. Since rates began diverging in 2012, underlying US trend growth has risen relative to Europe’s. In essence, higher US interest rates thus largely reflect a stronger US economy. By the same token, there are solid reasons for lower rates in the euro area – namely, that its economy has underperformed relative to the United States.

Takeaways by Macro Roundup® AI

  1. The federal funds target range stands at 4.25%–4.5% while the ECB's deposit facility rate is 2%, leaving US policy rates more than two percentage points above the euro area's.
  2. US interest rates have been consistently higher than euro-area rates since 2012 and were usually higher before that as well.
  3. Higher US rates largely reflect a stronger US economy, as underlying US trend growth has risen relative to Europe's since rates began diverging in 2012.

Related Articles:

  • Data Update 4 for 2025: Interest Rates, Inflation and Central Banks! — Interest rates are ultimately driven by macroeconomic fundamentals, with an "intrinsic risk-free rate" defined as the sum of the inflation rate and real GDP growth tracking the observed 10-year Treasury bond yield closely. On this view, central banks' influence comes from preferential access to data on those fundamentals, their capacity to alter them, and their credibility.

AI Summary. The authors decompose the gender earnings gap among college-educated workers into college major, graduate degree attainment, and graduate field, estimating that education differences were large for early birth cohorts and declined substantially. They estimate changes in fields of study contributed 0.128 to the gap's decline, reduced to 0.055 once shifts in field-specific returns favoring men are included.

Joseph Altonji, John Eric Humphries, Yagmur Yuksel and Ling Zhong National Bureau of Economic Research
Date Posted:
Is Database:
Database

The bulk of the decline in the gender gap among full-time college educated workers born btw the 1931 and 1984 birth cohorts is within occupation. Changes in fields of study contribute on net only ~5.5pp out of a more than 40pp drop.

How much have education choices narrowed the gender earnings gap?

Core argument: Holding relative returns to fields constant, changes in fields of study contribute 0.128 to the decline in the gender earnings gap, but cohort trends in relative returns to specific fields favored men over women, reducing that contribution to 0.055.

We decompose the education gap into the contributions from college major choice, graduate degree attainment, and graduate field. When evaluated at fixed relative returns to each degree type, we find that the education gap is large for the early cohorts but declines substantially and is an important part of the narrowing of the gender gap. We find that much of the large gap in earnings between the 1931 and 1950 cohorts is due to a cohort specific “residual component” that shifts the gender gap in earnings by the same amount for all college graduates. Most of the decline is within occupation, especially for the early cohorts. The residual gap varies little for the 1951 to late 1970s birth cohorts, after which it resumes its decline. When holding the relative returns to fields constant, changes in fields of study contribute 0.128 to the decline in the gender gap, but this decline is partially offset by cohort trends in the relative returns to specific fields that favored men over women, reducing the contribution of field-of-study changes to the decline to 0.055. Degree combinations contribute to the [overall level of the] gender gap, but very little to the decline in the gender gap over the full time period.

Takeaways by Macro Roundup® AI

  1. Holding relative returns to fields constant, changes in fields of study contribute 0.128 to the decline in the gender earnings gap, but cohort trends in relative returns to specific fields favored men over women, reducing that contribution to 0.055.
  2. Much of the large gender earnings gap between the 1931 and 1950 birth cohorts is attributed to a cohort-specific residual component that shifts the gap by the same amount for all college graduates, with most of the decline occurring within occupation.
  3. The residual gender gap varies little across the 1951 through late-1970s birth cohorts before resuming its decline, while degree combinations contribute to the level of the gender gap but very little to its decline over the full period.

Related Articles:

  • One Cohort at a Time: A New Perspective on the Declining Gender Pay Gap — The authors find the entire decline in the gender pay gap comes from newer cohorts entering with smaller-than-average gaps and older cohorts exiting with larger ones, with entry convergence driven by younger men's positional losses in the wage distribution, especially at higher-paying firms. They find labor-market exit became the sole contributor, implying no full convergence ahead.
  • Women’s Evolving Careers Helped Shrink the Gender Pay Gap — The share of women in their late 20s working in the four highest-paying broad occupation groups rose from about 30% to 55%, while the share in professional occupations rose from about 19% to 26%. Women early in their careers shifted away from predominantly administrative work toward better-paying occupations.
  • The ‘Gender Pay Gap’ Is a Myth That Won’t Go Away — The commonly cited figure that women earn 84% of men's pay compares average annual earnings of full-time, year-round workers, a category defined broadly enough to obscure differences in hours, experience, and occupation. Women over 40 average three fewer years of experience than men the same age, which the author says explains about a third of the gap.

AI Summary. The US government is in discussions to take roughly a 10% stake in chipmaker Intel, potentially by converting some or all of the company's federal semiconductor grants into equity, which could make the government Intel's largest shareholder. Intel has been slated to receive $10.9bn in such grants for commercial and military production.

Mackenzie Hawkins, Ryan Gould, and Josh Wingrove Bloomberg
Date Posted:

The Trump administration is considering taking a 10% stake in Intel via converting ~$10.9B in grants committed under the US Chips and Science Act into an equity stake. The USG would be the company’s largest shareholder.

Should the government convert chip grants into an equity stake?

Core argument: The Trump administration is in talks to take a stake of about 10% in Intel, a move that could make the US government the chipmaker's largest shareholder, according to a White House official and others familiar with the matter.

The Trump administration is in discussions to take a stake of about 10% in Intel, according to a White House official and other people familiar with the matter, in a move that could see the US become the beleaguered chipmaker’s largest shareholder. The federal government is considering a potential investment in Intel that would involve converting some or all of the company’s grants from the US Chips and Science Act into equity, said the people, who asked not to be identified because the information is confidential. Intel has been slated to receive a combined $10.9 billion in Chips Act grants for commercial and military production.

Takeaways by Macro Roundup® AI

  1. The Trump administration is in talks to take a stake of about 10% in Intel, a move that could make the US government the chipmaker's largest shareholder, according to a White House official and others familiar with the matter.
  2. The potential federal investment would involve converting some or all of Intel's US Chips and Science Act grants into equity, according to people familiar with the discussions.
  3. Intel has been slated to receive a combined $10.9 billion in Chips Act grants for commercial and military production.

Related Articles:

  • Pentagon to Take Stake in Rare-Earth Company, Challenging China’s Control — Under a deal with the largest U.S. rare earths miner, the Defense Department will take a 15% stake, commit billions of dollars to investment and output purchases, and guarantee a price floor for rare earth minerals and magnets. The agreement calls for a new magnet factory, expected online in 2028, at a scale far exceeding current U.S. magnet production.
  • Nippon Steel Considers ‘Golden Share’ Idea For US Steel Deal — Nippon Steel is considering granting the U.S. government a "golden share" in U.S. Steel to advance its acquisition, a special class of stock whose holder can control board composition and wield influence over major management decisions. U.S.-listed companies generally cannot issue golden shares, but Nippon Steel plans to delist U.S. Steel.
  • The Chips Act Has Been Surprisingly Successful So Far — The US government has spent over half of its $39bn in Chips Act semiconductor incentives, and chip companies and supply chain partners have announced $327bn of investments over the next 10 years, according to Semiconductor Industry Association calculations. US data show a 15-fold increase in construction of computing and electronics manufacturing facilities.

Tuesday, August 19, 2025

AI Summary. The US yield curve is steepening across both the 2-to-10-year and 10-to-30-year segments. The article attributes this to Federal Reserve rate cuts, the risk that markets read those cuts as politically motivated and raise inflation expectations and long-term rates, and growing Treasury debt issuance pushing long rates up.

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

Torsten Sløk attributes the steepening of the US yield curve to accelerating Treasury issuance, an expectation of politically motivated cuts in the short rate, and a rise in expected inflation.

Why is the yield curve steepening?

Core argument: The US yield curve has begun steepening in both 2s10s and 10s30s, attributed to Federal Reserve rate cuts, growing Treasury issuance, and fiscal challenges.

The US yield curve has started steepening, not only 2s10s but also 10s30s. There are three reasons why this is happening: The Fed is cutting rates. If the market thinks the Fed is cutting for political reasons, it puts upward pressure on inflation expectations and ultimately long rates, which also steepens the curve. Growing Treasury issuance is putting upward pressure on long rates. If Fed Chair Powell, in his Jackson Hole speech on Friday, is going to signal less commitment to the Fed’s 2% inflation target, the curve will steepen further.

Takeaways by Macro Roundup® AI

  1. The US yield curve has begun steepening in both 2s10s and 10s30s, attributed to Federal Reserve rate cuts, growing Treasury issuance, and fiscal challenges.
  2. If markets conclude the Federal Reserve is cutting rates for political reasons, inflation expectations and long rates face upward pressure, steepening the yield curve further.
  3. Growing Treasury issuance is putting upward pressure on long rates, and a signal from Fed Chair Powell at Jackson Hole of less commitment to the 2% inflation target would steepen the curve further.

Related Articles:

  • Data Update 4 for 2025: Interest Rates, Inflation and Central Banks! — Interest rates are ultimately driven by macroeconomic fundamentals, with an "intrinsic risk-free rate" defined as the sum of the inflation rate and real GDP growth tracking the observed 10-year Treasury bond yield closely. On this view, central banks' influence comes from preferential access to data on those fundamentals, their capacity to alter them, and their credibility.
  • Rapidly Growing Treasury Supply Crowding Out Other Types of Credit Growth — Roughly half of all fixed income issuance coming to market over the past year has been US Treasuries, meaning investors must allocate a growing share of dollars to financing the government. The argument is that lower government debt would free up more funds for consumer borrowing and business investment.
  • The Long-Term Budget Outlook Under Alternative Scenarios for the Economy and the Budget — The Congressional Budget Office projects that if private investment were more sensitive to federal deficits, debt held by the public would exceed 250% of GDP and real national income per person would be 18% ($21,300) lower than its baseline, with the average interest rate on federal debt 2pp higher.

AI Summary. The equal-weight S&P 500, which treats every company as 0.2% of the index, has hit its weakest performance relative to the market-cap-weighted index in 22 years, as a small group of companies captures a large share of profits. Big Tech's capital spending on AI computing and energy has surged, leaving little free cash flow for buybacks or dividends.

John Authers Bloomberg
Date Posted:
Is Database:
Database
Is Important:
Important

Since the release of ChatGPT in November of 2022, earnings per share of the MAG 7 have risen 125% (through 8/18), while earnings per share of the other 493 companies comprising the S&P 500 have risen just 4%.

Are a small group of companies capturing most stock market gains?

Core argument: The equal-weight S&P 500, which weights each constituent at 0.2% regardless of size, has fallen to its weakest performance relative to the cap-weighted index in 22 years.

The performance of the equal-weight version of the S&P 500 — which effectively measures the average stock by weighting each constituent as 0.2% of the index regardless of their size — compared to the cap-weighted index has dropped to its weakest in 22 years. This is not like the dot-com bubble, when hot companies hoovered up cash from investors. This time, a very small group is siphoning profits from everyone else. Rather than the market, the concern lies with the companies themselves. How can they make such big profits, and can they possibly retain them? The big tech groups have also been spending money far faster than ever before. This is a new development. Tech is usually “capital-light.” Capital expenditures on the computing capacity needed for AI and the energy to back it have surged, so that the companies now invest almost as much of their cash as everyone else combined. Well-deployed capital expenditures pay for themselves with higher profits, but those still have to happen. Meanwhile, Big Tech’s free cash flow yield (operating cash flows minus capital expenditures as a proportion of market cap) leaves little wiggle room for share buybacks or dividends.

Takeaways by Macro Roundup® AI

  1. The equal-weight S&P 500, which weights each constituent at 0.2% regardless of size, has fallen to its weakest performance relative to the cap-weighted index in 22 years.
  2. Unlike the dot-com bubble, when hot companies absorbed cash from investors, a very small group of companies is now siphoning profits from everyone else, raising the question of whether such profits can be retained.
  3. Big Tech's capital expenditures on AI computing capacity and energy have surged so that these companies invest almost as much of their cash as everyone else combined, while their free cash flow yield leaves little room for buybacks or dividends.

Related Articles:

  • Summer Mailbag — The share of total US market capitalization held by young unprofitable companies has fallen from its 2021–2022 peak, while over 40% of small-cap companies have negative earnings and the negative-earnings share among large caps is falling. The 10 largest stocks hold about 35% of market cap, a high since 1990, but their share of market earnings has roughly kept pace.
  • Which U.S. Stocks Generated the Highest Long-Term Returns? — Of the 29,078 U.S. common stocks in a database spanning nearly a century, 51.6% had negative cumulative returns, while 17 stocks returned more than 5m% — led by Altria at 265m%. Those top performers averaged annualized returns of just 13.47%, underscoring the value of time invested.
  • Long-Term Shareholder Returns: Evidence From 64,000 Global Stocks — A study of about 64,000 global stocks estimates net global stock market wealth creation of $75.7tn, with returns highly concentrated: the top 159 firms (0.25%) account for half of net wealth creation and the top 1,526 (2.39%) account for all of it. Fewer than half of stocks beat one-month US Treasury bills, 42.6% of non-US and 44.8% of US stocks.

AI Summary. About a sixth of America's 2% real GDP growth over the past year came from investment in computer and communications equipment and data centres, and one estimate puts the artificial intelligence boom's total contribution to growth at 40%. Average electricity bills rose 7%, attributed in part to data-centre demand, while housebuilding and non-AI business investment slumped and real consumption flatlined.

Economist Staff The Economist
Date Posted:
Is Database:
Database
Is Important:
Important

As AI-related investment has risen since 2023, residential and nonresidential investment have declined or flatlined. This may suggest that a relatively rate-insensitive AI buildout is crowding out more interest-sensitive forms of investment.

Is the AI investment boom crowding out the rest of the economy?

Core argument: Roughly a sixth of the 2% rise in American real GDP over the past year came from investment in computer and communications equipment, including chips, and data centres; one estimate puts the AI boom's contribution at 40% once grid upgrades and software intellectual property are included.

Something like a sixth of the 2% rise in American real GDP over the past year has come from investments in computer and communications equipment, including chips, and data centres. Add in the grid upgrades to power AI models, plus the intellectual-property value of the software itself, and one estimate puts the boom’s contribution to real GDP growth at 40%. The trouble is that the very sector powering so much of America’s economic growth is squeezing the rest of its output. Housebuilders, for instance, cannot afford to be blithe about higher borrowing costs. Data centres have also constrained the rest of the economy by keeping energy prices high. Average American electricity bills have risen by 7% so far in 2025, at least in part due to the extra strain data centres have put on the grid. Real consumption has flatlined since December. Housebuilding has slumped, as has non-AI business investment.

Takeaways by Macro Roundup® AI

  1. Roughly a sixth of the 2% rise in American real GDP over the past year came from investment in computer and communications equipment, including chips, and data centres; one estimate puts the AI boom's contribution at 40% once grid upgrades and software intellectual property are included.
  2. Average American electricity bills have risen by 7% so far in 2025, at least in part due to the extra strain data centres have put on the grid.

Related Articles:

  • Honey, AI Capex is Eating the Economy — Capital spending on AI data centers is around 20% of railroads' peak spending as a share of GDP and is rising quickly, having already surpassed the dot-com era peak in telecom investment. Unlike long-lived rail infrastructure, data centers are short-lived assets requiring frequent hardware replacement, and the capital is being reallocated from other sectors.
  • Back to Our Regularly Scheduled Programming — The four largest US hyperscalers spent $440bn on capital expenditure and research in 2024, with $596bn projected for 2025, a 35% increase, leaving large gaps between AI spending and AI revenues. These firms account for more than half of all capital expenditure growth in the US equity market.
  • The AI Boom’s Hidden Risk to the Economy — Net income at Alphabet, Amazon, Meta and Microsoft rose 73% to $91bn in a quarter two years on, while their combined free cash flow fell 30% to $40bn, as heavy AI investment absorbs cash these firms once recycled into financial markets. That spare cash previously helped hold down long-term interest rates despite large federal deficits.

AI Summary. The author argues the AI infrastructure build-out resembles an arms race fought to avoid defeat rather than to win, with Apple the only major tech platform largely sitting it out. Google's advertising model, which depends on engagement with search result pages, could erode if chatbots answer queries in a single exchange with no click-through.

Brad DeLong Grasping Reality
Date Posted:
Is Database:
Database
Is Important:
Important

Brad DeLong suggests that AI investment is largely defensive in nature, “with fear of AI‑driven disruption defining the moment,” and is skeptical regarding its prospects. ”What, concretely, are the ‘new-and-improved’ services that justify the AI boom?”

Does the AI infrastructure build-out threaten search advertising economics?

Core argument: The author characterizes the AI-infrastructure build-out as an arms race waged to avoid defeat rather than to win decisively, with Apple the only tech platform oligopolist essentially sitting it out.

This build‑out resembles an old-fashioned arms race: not to win decisively, but to avoid defeat. And so of all the tech platform oligopolists, only Apple is essentially sitting this one out. Google’s search monopoly, once an unassailable gold mine, now perhaps faces an existential threat. First, its search is crippled because it swims in an ocean of SEO‑optimized content—a potential Achilles’ heel. Chatbot rivals, free of SEO baggage and possessing a natural-language interface, deliver plausible conversational answers that are not SEO-warped as they bypass click-and-scroll. And even if Google manages to match answer quality, the economics shift. An ad model built on result‑page engagement can evaporate when a single exchange answers the query with no click‑through. Victory here may well be Pyrrhic. The old money flow may not exist in the new régime, no matter how strong the constructed defensive moat. But I am given pause by the fact that I do not understand why AWS today is so massively profitable. There are clearly important things that I do not understand here.

Takeaways by Macro Roundup® AI

  1. The author characterizes the AI-infrastructure build-out as an arms race waged to avoid defeat rather than to win decisively, with Apple the only tech platform oligopolist essentially sitting it out.
  2. Google's search monopoly may face an existential threat, in the author's view, because its results swim in SEO-optimized content while chatbot rivals offer natural-language answers that bypass click-and-scroll.
  3. The author argues that Google's victory could prove Pyrrhic, since an ad model built on result-page engagement can evaporate when a single exchange answers a query with no click-through, and concedes not understanding why AWS is so profitable.

Related Articles:

  • ‘Absolutely Immense’: The Companies On The Hook For The $3tn AI Building Boom — Google, Amazon, Microsoft and Meta are projected to spend more than $400bn on data centres in 2026, following more than $350bn this year. Morgan Stanley analysts estimate global data centre spending will approach $3tn through 2029, of which about $1.4tn comes from Big Tech capital expenditure, leaving roughly $1.5tn to be financed by investors and developers.
  • Is AI Hitting A Wall? — Artificial intelligence gains have come from a formula of more data and more computing power, but companies have exhausted freely available internet training data and are now striking data-sharing deals with publishers and copyright holders. Training and running large models is energy-intensive, requiring vastly more Nvidia chips with each generation.
  • The Kanye/Data Center Crossover — Data center capital spending differs from fiber and railroad investment because most of the cost sits in processors that are replaced every 3-4 years rather than in long-lived shells, power, and cooling connections. That short replacement cycle requires returns high enough to exceed a risk-adjusted cost of capital of 12-14% before the hardware must be replaced.

AI Summary. Canada's unemployment rate has risen to 6.9% from 6.6% at the start of the year, with youth unemployment at 15% and the share of young people employed at its lowest level since 1998, excluding the Covid-19 years. The share of job seekers unemployed longer than six months is also the largest since 1998.

Ari Altstedter Bloomberg
Date Posted:
Is Database:
Database

Alberta Central, a Canadian bank, estimates Canada’s unemployment rate for 16–24 year olds is 15%. The overall unemployment rate according to Statistics Canada is 6.9%.

Are young Canadians facing the weakest job market since 1998?

Core argument: Canada's youth unemployment rate stands at 15%, and the share of young people with jobs has fallen to its lowest level since 1998, excluding the years affected by Covid-19.

Canada’s unemployment rate has crept up to 6.9%, from 6.6% at the beginning of the year, according to Statistics Canada data. The share of job seekers who have been out of work for more than six months is the largest since 1998. Across Canada, the percentage of young people with jobs has collapsed to the lowest level since 1998, excluding the years impacted by Covid-19. Employment growth for 16- to 24-year-olds was the weakest on record in July, according to calculations by economists at credit union Alberta Central. The country's youth unemployment rate stands at 15%.

Takeaways by Macro Roundup® AI

  1. Canada's youth unemployment rate stands at 15%, and the share of young people with jobs has fallen to its lowest level since 1998, excluding the years affected by Covid-19.
  2. Canada's unemployment rate has risen to 6.9% from 6.6% at the start of the year, according to Statistics Canada, with the share of job seekers out of work more than six months the largest since 1998.
  3. Employment growth for Canadians aged 16 to 24 was the weakest on record in July, according to calculations by economists at credit union Alberta Central.

Related Articles:

  • Canada’s Growth Challenge: Why The Economy Is Stuck In Neutral — Canada's labor productivity is about 30% below the United States and roughly 10% below Australia's, and Canada has fallen from the 6th to the 18th most productive economy in the OECD. The authors attribute this to a shortfall in investment outside real estate, construction, and public services, plus negative multifactor productivity growth.
  • A “Heat Dome” of Immigration — Canada has adopted some of the most aggressive immigration policies in the OECD, adding 1.273m residents in a year to reach a population of 41m — proportionally equivalent to the US admitting more than 10m immigrants annually. Canadians are increasingly angry as housing prices soar, sprawl expands, and traffic worsens.
  • Immigration Surge Fuels Male Population Boom in Canada — Canada's adult male population grew 3.4% over the past year versus 2.9% for women, the widest gap in nearly 50 years of records, as a higher share of male newcomers shifts the gender ratio. Among those aged 25 to 44, men outnumber women by 141,000, against a long-run average of zero.

AI Summary. The United Nations Office on Drugs and Crime reports that cocaine production, seizures, and use all hit record highs, with illegal output rising nearly 34% to 3,708 tonnes in a single year and making cocaine the world's fastest-growing illicit drug market. Europe has overtaken the US as the largest cocaine market, and production in Colombia, the biggest producer, rose 53%.

Michael Stott, Michael Pooler, Ciara Nugent and Christine Murray Financial Times
Date Posted:

Cocaine is making a comeback, driven by European demand which has overtaken the US as the world’s largest market. In Columbia, the world’s largest producer, cocaine production rose 53% btw 2022 and 2023.

Is global cocaine production and use reaching record highs?

Core argument: The UN Office on Drugs and Crime reported that illegal cocaine production reached 3,708 tonnes in 2023, nearly 34% more than in 2022, with production, seizures and use all hitting new highs, making cocaine the world's fastest-growing illicit drug market.

The global cocaine business is booming as never before. Europe’s drug habit has grown so fast in the past two decades that it has overtaken the US as the biggest cocaine market. At the heart of Latin America’s vast criminal enterprises is cocaine, by far the most profitable illegal business. “Production, seizures and use of cocaine all hit new highs in 2023, making cocaine the world’s fastest-growing illicit drug market,” said the UN Office on Drugs and Crime in its latest annual report. “Illegal production skyrocketed to 3,708 tonnes, nearly 34% more than in 2022.” In Colombia, the world’s biggest producer, cocaine production rocketed 53% from 2022 to 2023. A senator and presidential hopeful, Miguel Uribe, was shot in June at a rally in Bogotá and later died of his wounds, raising fears that the Andean nation might relapse into the cocaine-fuelled political violence which scarred it in the 1980s and 1990s.

Takeaways by Macro Roundup® AI

  1. The UN Office on Drugs and Crime reported that illegal cocaine production reached 3,708 tonnes in 2023, nearly 34% more than in 2022, with production, seizures and use all hitting new highs, making cocaine the world's fastest-growing illicit drug market.
  2. Colombia, the world's biggest cocaine producer, saw production rise 53% from 2022 to 2023, while Europe's drug habit has grown over two decades to overtake the US as the biggest cocaine market.
  3. Colombian senator and presidential hopeful Miguel Uribe was shot in June at a rally in Bogotá and later died of his wounds, raising fears the country might relapse into the cocaine-fuelled political violence of the 1980s and 1990s.

Related Articles:

  • Drug Overdoses Are On the Decline, in Charts — Preliminary estimates from the Centers for Disease Control and Prevention show US overdose deaths fell 25% over a 12-month period to about 82,000, led by a falling number of fatalities involving synthetic opioids such as bootleg fentanyl.
  • Generation Xanax: The Dark Side of America’s Wonder Drug — Benzodiazepines such as alprazolam, clonazepam, lorazepam, and diazepam are among the most commonly prescribed psychiatric medications in America, with nearly 24m Americans using them and more than 86m prescriptions written in a year. The Food and Drug Administration has required manufacturers to warn of serious risks of abuse, physical dependence, and withdrawal reactions.
  • Importing the Opioid Crisis? International Trade and Fentanyl Overdoses — A study using U.S. mortality and import data finds a positive state-level association between imports per resident and drug overdose death rates, driven by fentanyl and accounting for an estimated 15,000–20,000 deaths per year. The authors estimate that 10% higher imports per resident is associated with an 8.1% higher fentanyl death rate.