Edward Conard

Top Ten New York Times Bestselling Author

  • “A full-throated defense of economic dynamism.” - The Wall Street Journal
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
  • “…a very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard University
  • “…a comprehensive explanation of the modern economy.” - Julian Robertson, Founder, Tiger Management
  • “…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 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
  • “…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 very valuable contribution.” - Larry Summers, former Secretary of the Treasury and director of the National Economic Council, president emeritus, Harvard 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
  • “…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 should be read by anyone who takes for granted the superiority of progressive taxation and has not thought carefully about the trade-offs involved.” - The New Republic
  • “…challenges misconceptions that distort our economic debates.” - Arthur Brooks, President of the American Enterprise Institute
Upside of Inequality Oxford Unintended Consequences

Macro Roundup


Monday, January 27, 2025

AI Summary. Consumers and businesses stand to gain most from a future of effectively free AI products, along with big consumer tech companies, since a world of free AI makes product and distribution the decisive advantages they already hold. China also benefits, and the United States faces a choice between expanding chip export restrictions and permission-based regulation or allowing itself to compete.

Ben Thompson Stratechery
Date Posted:

.@benthompson argues DeepSeek implies “Jevon’s Paradox will rule the day in the long run” as consumers and businesses deploy “effectively-free AI products and services.”

Who benefits most from a future of effectively free AI?

Core argument: The author argues consumers and businesses are the biggest winners from DeepSeek, anticipating effectively-free AI products and services, with Jevon's Paradox ruling in the long run so that everyone who uses AI wins.

The biggest winners are consumers and businesses who can anticipate a future of effectively-free AI products and services. Jevon’s Paradox will rule the day in the long run, and everyone who uses AI will be the biggest winners. Another set of winners are the big consumer tech companies. A world of free AI is a world where product and distribution matters most, and those companies already won that game. China is also a big winner, in ways that I suspect will only become apparent over time. Not only does the country have access to DeepSeek, but I suspect that DeepSeek’s relative success to America’s leading AI labs will result in a further unleashing of Chinese innovation as they realize they can compete. That leaves America, and a choice we have to make. We could, for very logical reasons, double down on defensive measures, like massively expanding the chip ban and imposing a permission-based regulatory regime on chips and semiconductor equipment that mirrors the EU’s approach to tech; alternatively, we could realize that we have real competition, and actually give ourself permission to compete.

Takeaways by Macro Roundup® AI

  1. The author argues consumers and businesses are the biggest winners from DeepSeek, anticipating effectively-free AI products and services, with Jevon's Paradox ruling in the long run so that everyone who uses AI wins.
  2. The author holds that big consumer tech companies win in a world of free AI, because product and distribution matter most there and those firms have already won that game.
  3. The author contends China is a major winner, suspecting DeepSeek's relative success against America's leading AI labs will further unleash Chinese innovation, leaving the U.S. to choose between expanded chip bans and permission-based regulation or permitting itself to compete.

Related Articles:

  • The Year-End Wake-Up: The ‘Chinese Sputnik’ That Demands Our Attention — Chinese AI company DeepSeek trained its Version 3 large language model for roughly $5.6m, about 100x cheaper than Meta's Llama 3, with performance comparable to Claude 3.5 Sonnet and interface access costing 12x less than Anthropic's. The author argues US chip export controls backfired by spurring leaner engineering, and that the gap between open-source and closed-source models is narrowing rapidly.
  • China’s AI Industry Has Almost Caught Up With America’s — Chinese developer DeepSeek trained its V3 large language model using fewer than 3m chip-hours at an estimated cost under $6m—roughly a tenth of the computing power and expense of Meta's Llama 3.1—through a series of marginal efficiency improvements rather than one breakthrough. The model was trained on chips constrained by American export sanctions.
  • 2025 AI Diffusion Export Controls – Microsoft Regulatory Capture, Oracle Tears, Impacts Quantified, Model Restrictions — The US Framework for Artificial Intelligence Diffusion imposes export restrictions aimed at preventing China from accessing the computing power needed to build frontier AI models. It sorts the world into three tiers — the US and 18 allies, 23 arms-embargoed nations plus Macau, and all other countries — with differing access to AI chips and caps on graphics processor purchases.

AI Summary. Chinese artificial intelligence start-up DeepSeek released a large language model with performance comparable to US rival OpenAI while claiming to use far fewer Nvidia chips, raising doubts about whether US AI leadership requires spending billions of dollars on chips. Nvidia lost almost $600bn in market value, the largest recorded drop for any company.

Arjun Neil Alim, Leo Lewis, Eleanor Olcott, Tim Bradshaw and Ian Smith Financial Times
Date Posted:

China’s DeepSeek shocked Silicon Valley with a model rivaling OpenAI’s, developed “with far less computing power than US rivals,” calling into question expectations about the AI capex cycle. Nvidia is down ~ 17% and Siemens fell ~ 20%.

Does US AI leadership require spending billions on chips?

Core argument: Nvidia lost nearly $600bn of market value, the biggest recorded drop for any company, as investors reassessed likely future investment in AI hardware following DeepSeek's model release.

Tech stocks tumbled as advances by Chinese artificial intelligence start-up DeepSeek cast doubt on whether the US could sustain its leadership in AI by spending billions of dollars on chips. DeepSeek last week released its latest large language AI model, which achieved a comparable performance to that of US rival OpenAI, even though the company has previously claimed to use far fewer Nvidia chips. Venture capital investor Marc Andreessen called the new Chinese model “AI’s Sputnik moment”, drawing a comparison with the Soviet Union stunning the US by putting the first satellite into orbit. The results sent a shockwave through markets on Monday, with Nvidia losing almost $600B of market value, the biggest recorded drop for any company, as investors reassessed the likely future investment in AI hardware. The rout extended well beyond traditional tech names. Siemens Energy, which supplies electrical hardware for AI infrastructure, plunged 20%.

Takeaways by Macro Roundup® AI

  1. Nvidia lost nearly $600bn of market value, the biggest recorded drop for any company, as investors reassessed likely future investment in AI hardware following DeepSeek's model release.
  2. DeepSeek's latest large language model achieved performance comparable to that of OpenAI, even though the Chinese start-up has previously claimed to use far fewer Nvidia chips.
  3. The selloff extended beyond traditional tech names, with Siemens Energy, a supplier of electrical hardware for AI infrastructure, plunging 20%, while venture capital investor Marc Andreessen called the Chinese model "AI's Sputnik moment".

Related Articles:

  • The Year-End Wake-Up: The ‘Chinese Sputnik’ That Demands Our Attention — Chinese AI company DeepSeek trained its Version 3 large language model for roughly $5.6m, about 100x cheaper than Meta's Llama 3, with performance comparable to Claude 3.5 Sonnet and interface access costing 12x less than Anthropic's. The author argues US chip export controls backfired by spurring leaner engineering, and that the gap between open-source and closed-source models is narrowing rapidly.
  • SoftBank and OpenAI Back Sweeping AI Infrastructure Project In US — OpenAI and SoftBank plan a US artificial intelligence infrastructure joint venture called Stargate, intending to spend $100bn initially and as much as $500bn over four years. SoftBank holds financial responsibility and OpenAI operational responsibility, with Abu Dhabi state fund MGX and Oracle also funding and Arm, Microsoft and Nvidia as technology partners.
  • A Severe Case of COVIDIA: Prognosis For An AI-Driven US Equity Market — Computing cycles move from infrastructure to platforms to applications, and NVIDIA data center revenues as a share of market-wide capital spending are projected to reach levels last seen at the 1969 mainframe peak and the dot-com boom. Large cloud providers spend more than twice as much training foundational models as running them for customers.

AI Summary. Technology stocks added almost $4.63T in market capitalization in a single year and $13.6T over a decade, with semiconductor and auto gains concentrated in standout firms such as Nvidia and Tesla. Seven companies—Apple, Amazon, Meta, Alphabet, Microsoft, Nvidia and Tesla—added $15.8T over ten years, about 40% of the total gain across US stocks.

Aswath Damodaran Musing on Markets
Date Posted:
Is Database:
Database

Between 2014 and 2024, the Mag 7 added $15.8T in market cap, approximately 40% of the increase in market capitalization across all US equities over the decade. @AswathDamodaran

How concentrated are US stock market gains among a few technology firms?

Core argument: The Magnificent Seven — Apple, Amazon, Meta, Alphabet, Microsoft, Nvidia and Tesla — added $5.6T in market cap in 2024, almost the entire market's gain for the year.

Technology stocks collectively added almost $4.63T in market cap just in 2024, and over the last five (ten) years, the sector has added $11.3T ($13.6T) in market cap. The top two performers, in 2024, on the best performing industry list, semiconductors and auto & truck, owe much of their overall performance to super-performers in each one (Nvidia with semiconductors and Tesla with auto & truck), but airline companies also had a good year, though it may be premature to conclude that they have finally found working business models that can deliver profitability on a continuous basis. In 2024, seven companies (Apple, Amazon, Meta, Alphabet, Microsoft, Nvidia and Tesla) increased in market cap by $5.6T, almost of the entire market's gain for the year. While it is not uncommon for stock market returns to be delivered by a few winners at the top, with the Mag Seven, the domination extends over a decade, and in the last ten years (2014-2024), these seven companies have added $15.8T in market cap, about 40% of the increase in market capitalization across all US stocks over the decade.

Takeaways by Macro Roundup® AI

  1. The Magnificent Seven — Apple, Amazon, Meta, Alphabet, Microsoft, Nvidia and Tesla — added $5.6T in market cap in 2024, almost the entire market's gain for the year.
  2. Over 2014–2024, Apple, Amazon, Meta, Alphabet, Microsoft, Nvidia and Tesla added $15.8T in market cap, about 40% of the increase in market capitalization across all US stocks over the decade.
  3. Technology stocks added almost $4.63T in market cap in 2024, and $11.3T and $13.6T over the last five and ten years respectively, with semiconductors and auto & truck the top 2024 performers, owing much to Nvidia and Tesla.

Related Articles:

  • Data Update 2 for 2025: The Party Continues (for US Equities)! — Applying an internal-rate-of-return approach to the S&P 500 using consensus earnings and payout estimates implies an expected stock return of 8.91% and an equity risk premium of 4.33%. Assuming the premium reverts to 4.5% yields an index value about 12% below current levels, with an estimated 80% chance stocks are overvalued.
  • On Bubble Watch — Data from J.P. Morgan Asset Management covering monthly observations over a 27-year period show a strong relationship between the S&P 500's starting price-to-earnings ratio and its annualized return over the following ten years, with higher starting valuations consistently followed by lower returns. Purchases at a multiple near today's 22x were followed by ten-year returns between +2% and -2%.
  • US Exceptionalism: Drivers of Equity Outperformance and What’s Needed for a Repeat — US companies need about 7% earnings-per-share growth to deliver a normal 4% equity risk premium above the risk-free rate, versus about 3% for other developed markets, with the highest expectations embedded in technology and the largest US tech firms. Chinese companies could generate a 4% risk premium over Chinese bonds with no earnings growth.

AI Summary. The argument holds that large US trade deficits reflect capital flowing in to take advantage of high returns as the US pulled ahead in information technology, since the trade balance plus net foreign investment inflows must sum to zero. On this view the deficits represent international macroeconomics working as intended, though the US has de-industrialized.

Paul Krugman Krugman Wonks Out
Date Posted:

Challenging @michaelxpettis @paulkrugman argues that trade deficits are the obverse side of capital account surpluses. As “the US economy was doing well, it was attracting a lot of foreign investment. And the balance of payments balances.”

Do large trade deficits reflect capital flowing toward high returns?

Core argument: The U.S. pulled ahead of the rest of the world during roughly 1990–2005 by figuring out how to use IT, and remains ahead, with much of the divergence occurring in that relatively brief period.

There was definitely a period of a decade or so [~1990-2005] when the U.S. was really figuring out what to do with IT and the rest of the world was not. And so the U.S. pulled ahead and is still ahead, although a lot of the divergence took place during this relatively brief period. You'll see that that's exactly when the United States began running big trade deficits. And the reason is that because the US economy was doing well, it was attracting a lot of inflows of foreign investment. And the balance of payments balances. The trade balance plus net inflows of foreign investment equals zero. So if we're getting a lot of money coming to America to take advantage of the opportunities, we run a big trade deficit. Hard to see how any of this is objectionable or a problem. This is international macroeconomics working the way it's supposed to. Capital flowing to areas that offer high returns to investment, which for a while was the United States. Okay, now the reason people think it's a problem is that we have definitely de-industrialized.

Takeaways by Macro Roundup® AI

  1. The U.S. pulled ahead of the rest of the world during roughly 1990–2005 by figuring out how to use IT, and remains ahead, with much of the divergence occurring in that relatively brief period.
  2. The onset of large U.S. trade deficits coincided with that period of outperformance, as a strong U.S. economy attracted large inflows of foreign investment and the balance of payments requires the trade balance plus net capital inflows to equal zero.
  3. Capital flowing to the United States in pursuit of high returns on investment represents international macroeconomics working as it is supposed to, and objections to the resulting trade deficits rest on the fact that the U.S. has de-industrialized.

Related Articles:

  • China’s Very Bad, No Good Trillion-Dollar Trade Surplus — China's high investment rates face diminishing returns and the government is unwilling to promote higher consumer spending, so it compensates for weak domestic demand with large trade surpluses supported by a weaker currency. The author argues this attempt to export its policy failure will provoke a protectionist backlash from Europe, the UK, and others.
  • Trade Intervention for Freer Trade — The authors argue the United States should counter surplus countries' trade policies either by restricting their goods exports through tariffs or by restricting their capital inflows through taxes, volume limits, reserve requirements, transaction approvals, or investor eligibility rules. They contend capital controls are a more targeted alternative to uniform tariffs, which impose equal costs on more and less mercantilist countries.
  • A User’s Guide to Restructuring the Global Trading System — During the 2018–2019 U.S.–China trade war, the effective tariff rate on Chinese imports rose 17.9 percentage points while the renminbi depreciated 13.7% against the dollar, leaving after-tariff import prices in dollars up just 4.1%. The author argues this currency offset of more than three-fourths of the tariff explains why inflation pressure was negligible.

AI Summary. German industrial output has fallen 15% since 2018, manufacturing employment is down 3%, and over €300bn of investment capital has flowed out of the country since 2021. Exports support roughly one in four German jobs and equal 43% of GDP, four times the US share and twice China's.

Tom Fairless and Bertrand Benoit Wall Street Journal
Date Posted:
Is Database:
Database

German industrial output has fallen by 15% since 2018 and manufacturing employment is down 3%, as German exports slow amid increasing global competition from Chinese manufacturers.

Is Germany's export-dependent economic model breaking down?

Core argument: German industrial output has fallen 15% since 2018 and manufacturing employment is down 3%, with over €300 billion in investment capital having flowed out of Germany since 2021.

Germany’s industrial output has fallen by 15% since 2018, and the total number of people employed in the manufacturing sector is down 3%. Manufacturers in Germany’s metal and electrical industry, weighed down by costs, could lay off as many as 300,000 workers over the next five years said Stefan Wolf, president of a lobby group for the sector, noting “Deindustrialization is in full swing.” Over €300 billion in investment capital has flowed out of Germany since 2021. Exports support roughly one in four German jobs. More than two-thirds of cars produced in Germany are exported. Since the mid-1990s, exports’ share of Germany’s GDP doubled, reaching 43% of GDP, four times the share in the U.S. and twice as high as China.

Takeaways by Macro Roundup® AI

  1. German industrial output has fallen 15% since 2018 and manufacturing employment is down 3%, with over €300 billion in investment capital having flowed out of Germany since 2021.
  2. Stefan Wolf, president of a lobby group for Germany's metal and electrical industry, said manufacturers weighed down by costs could lay off as many as 300,000 workers over the next five years, adding that "Deindustrialization is in full swing.".
  3. Exports have doubled as a share of German GDP since the mid-1990s to 43%, four times the U.S. share and twice China's, and support roughly one in four German jobs.

Related Articles:

  • Why the U.S. Economy Is Trouncing Europe’s — Only about 12% of Americans score at the highest levels on international academic tests versus 34% at the lowest, yet the U.S. has created nearly 250 companies worth over $10bn ($30tn in value) against Europe's 14, and median U.S. disposable household income is 25% above Germany's. The author argues expanding high-skilled immigration is the fastest way to accelerate growth.
  • How German Industry Can Survive the Second China Shock — The European Union has 30mm manufacturing jobs, and China has grown from a $4tn economy producing under 15% of global manufacturing output to an $18tn economy producing over a third, with well over 50% of global capacity in many sectors. European tariffs on Chinese electric vehicles limit import competition but do not offset lost export markets.
  • Will China Take Over the Global Auto Industry? — The global auto market has historically been regional, with most cars sold in a region also built there. Overcapacity in China's internal combustion sector, expanding electric-vehicle capacity, ample state bank credit, and a weaker yuan could concentrate global auto production in China absent restrictions such as the US 100% tariff on Chinese electric vehicles.

Tuesday, January 28, 2025

AI Summary. The author argues that capital inflows to the US are driven more by conditions abroad than by US economic performance, since the US attracts capital and runs deficits in both good times and bad. China grew faster than the US yet exported capital, a result of capital controls, an undervalued currency, and government-managed credit.

Michael Pettis @michaelxpettis
Date Posted:

Responding to @paulkrugman @michaelxpettis notes that “foreign capital pours into the US not just when times are good, but also when times are bad. US deficits are almost permanent and they don’t reflect US economic conditions.”

Are capital inflows to the US driven by conditions abroad?

Core argument: Pettis argues capital flows into the US in both good times and bad and that US deficits are almost permanent, disputing Krugman's account that late-1990s and early-2000s US productivity gains attracted foreign investment.

We disagree on what drives capital inflows. Krugman argues that rising US productivity in the late 1990s and early 2000s attracted foreign investment, and this is how economics should work: the US economy does well, imports capital, and runs deficits. However, I disagree. Capital flows into the US not just in good times, but also in bad, and US deficits are almost permanent, not reflecting US economic conditions. Another issue is the role of Asia. While China’s economy grew faster than the US’s, it had no capital inflows and instead exported large amounts of capital, which contradicts economic theory. This was due to policy interventions in China, like capital controls, an undervalued currency, and government-managed credit allocation. These conditions forced China to export capital, mostly to the US, even as the US economy weakened by 2005-06. Thus, foreign capital inflows to the US were more about conditions abroad than US performance, including the post-Asian Crisis reserve accumulation by developing countries. The US economy still had to adjust to this imbalance.

Takeaways by Macro Roundup® AI

  1. Pettis argues capital flows into the US in both good times and bad and that US deficits are almost permanent, disputing Krugman's account that late-1990s and early-2000s US productivity gains attracted foreign investment.
  2. China's economy grew faster than the US economy yet exported large amounts of capital rather than importing it, which Pettis attributes to capital controls, an undervalued currency, and government-managed credit allocation.
  3. Pettis contends foreign capital inflows to the US, including post-Asian Crisis reserve accumulation by developing countries, reflected conditions abroad rather than US performance, with the US economy still having to adjust to the imbalance.

Related Articles:

  • Rethinking Trade Imbalances — The argument holds that large US trade deficits reflect capital flowing in to take advantage of high returns as the US pulled ahead in information technology, since the trade balance plus net foreign investment inflows must sum to zero. On this view the deficits represent international macroeconomics working as intended, though the US has de-industrialized.
  • China’s Very Bad, No Good Trillion-Dollar Trade Surplus — China's high investment rates face diminishing returns and the government is unwilling to promote higher consumer spending, so it compensates for weak domestic demand with large trade surpluses supported by a weaker currency. The author argues this attempt to export its policy failure will provoke a protectionist backlash from Europe, the UK, and others.
  • A User’s Guide to Restructuring the Global Trading System — During the 2018–2019 U.S.–China trade war, the effective tariff rate on Chinese imports rose 17.9 percentage points while the renminbi depreciated 13.7% against the dollar, leaving after-tariff import prices in dollars up just 4.1%. The author argues this currency offset of more than three-fourths of the tariff explains why inflation pressure was negligible.

AI Summary. President Trump has said he favors a universal tariff rate "much bigger" than the 2.5% global rate reportedly preferred by his incoming Treasury Secretary, and has pledged sector-specific tariffs on semiconductors, pharmaceuticals, steel, copper, and aluminum. He also suggested tariffs on automobiles from Canada and Mexico, which he has already threatened with 25% across-the-board tariffs.

Stephanie Lai, Billy House, and Josh Wingrove Bloomberg
Date Posted:

Responding to Scott Bessent’s reported advocacy of an initial universal tariff starting at 2.5%, Trump said he favored a “much bigger” rate and in a recent speech pledged tariffs on semiconductors, pharmaceuticals, steel, copper, and aluminum.

Will a universal tariff be set much higher than 2.5%?

Core argument: Trump said he did not think incoming Treasury Secretary Scott Bessent backed a reported plan to start a universal tariff at 2.5%, and that he would not favor it himself, wanting a rate "much bigger" than 2.5%.

Asked about a report that incoming Treasury Secretary Scott Bessent favored starting with a global rate of 2.5%, Trump said he didn’t think Bessent supported that and wouldn’t favor it himself. He said he wanted a rate “much bigger” than 2.5%. Trump spoke aboard Air Force One while he flew back to Washington from a Florida speech where he pledged tariffs on specific sectors, including semiconductors, pharmaceuticals, steel, copper, and aluminum. He also strongly suggested he could also impose them on automobiles from Canada and Mexico, countries he’s already threatened with 25% across-the board tariffs as soon as Feb. 1.

Takeaways by Macro Roundup® AI

  1. Trump said he did not think incoming Treasury Secretary Scott Bessent backed a reported plan to start a universal tariff at 2.5%, and that he would not favor it himself, wanting a rate "much bigger" than 2.5%.
  2. In a Florida speech, Trump pledged tariffs on specific sectors, including semiconductors, pharmaceuticals, steel, copper, and aluminum.
  3. Trump suggested he could impose tariffs on automobiles from Canada and Mexico, countries he has already threatened with 25% across-the-board tariffs as soon as Feb. 1.

Related Articles:

  • Trump Pushes for Early Renegotiation of U.S. Trade Deal With Mexico, Canada — The U.S.-Mexico-Canada trade agreement governs roughly $2tn of trade and was approved with the broadest congressional support of any U.S. trade deal, with more than 200 Democrats backing it over added labor and environmental standards. Mexico and Canada have since supplanted China as the leading U.S. trading partners.
  • A User’s Guide to Restructuring the Global Trading System — During the 2018–2019 U.S.–China trade war, the effective tariff rate on Chinese imports rose 17.9 percentage points while the renminbi depreciated 13.7% against the dollar, leaving after-tariff import prices in dollars up just 4.1%. The author argues this currency offset of more than three-fourths of the tariff explains why inflation pressure was negligible.

AI Summary. Since MSCI data began in 1970, US stocks have outperformed European stocks by an average of 0.8 percentage points per year, with $1 invested in 1970 growing to $268 in the US versus $141 in Europe over 55 years. Before the financial crisis the ranking was reversed, with the US gain equal to 57% of the European gain.

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

Once at par with Europe’s American equity markets pulled away since ~ 2015. One dollar invested in the MSCI US index in 1970 would be worth $268 in 2025, vs. $141 for a dollar invested in the 1970 MSCI Europe index.

Have US stocks consistently outperformed European stocks over five decades?

Core argument: Since MSCI data collection began in 1970, US stocks have outperformed European stocks by an average of just 0.8 percentage points per year.

We often hear about the strength of the US stock market, which is widely perceived to have vastly outperformed its European counterpart. However, since MSCI data collection began in 1970, the difference in performance is relatively modest, averaging just 0.8 percentage points per year. A one-dollar investment in the US in 1970 would have grown to USD 47.8 by 2007, while the same investment in Europe would have reached USD 83.9. This means that the gain from a US investment would have been just 57% of the gain from a European investment. That said, a significant shift occurred around the time of the financial crisis. One dollar invested in the US stock market index in 1970 would have grown to USD 268 by 2025. Over the long term—55 years, 1970-2025—the US stock market has outperformed the European: if you had invested one dollar in the European stock market, it would have grown to only half as much as the US stock market—USD 141.

Takeaways by Macro Roundup® AI

  1. Since MSCI data collection began in 1970, US stocks have outperformed European stocks by an average of just 0.8 percentage points per year.
  2. One dollar invested in US stocks in 1970 would have grown to USD 47.8 by 2007, versus USD 83.9 in Europe, leaving the US gain at 57% of the European gain.
  3. Over 1970 to 2025, one dollar invested in the US stock market index would have grown to USD 268, while the same dollar in European stocks reached USD 141, about half as much.

Related Articles:

  • Why the U.S. Economy Is Trouncing Europe’s — Only about 12% of Americans score at the highest levels on international academic tests versus 34% at the lowest, yet the U.S. has created nearly 250 companies worth over $10bn ($30tn in value) against Europe's 14, and median U.S. disposable household income is 25% above Germany's. The author argues expanding high-skilled immigration is the fastest way to accelerate growth.
  • Data Update 2 for 2025: The Party Continues (for US Equities)! — Applying an internal-rate-of-return approach to the S&P 500 using consensus earnings and payout estimates implies an expected stock return of 8.91% and an equity risk premium of 4.33%. Assuming the premium reverts to 4.5% yields an index value about 12% below current levels, with an estimated 80% chance stocks are overvalued.
  • US Exceptionalism: Drivers of Equity Outperformance and What’s Needed for a Repeat — US companies need about 7% earnings-per-share growth to deliver a normal 4% equity risk premium above the risk-free rate, versus about 3% for other developed markets, with the highest expectations embedded in technology and the largest US tech firms. Chinese companies could generate a 4% risk premium over Chinese bonds with no earnings growth.

AI Summary. DeepSeek appears to have trained its models about 45x more efficiently than other leading-edge models, combining existing techniques—activating only 37bn of 671bn parameters at a time, a smaller memory cache, lower-precision training, and multi-token prediction—alongside self-reinforcement learning. Training cost $5.6m, and its interface charges $0.14 per million tokens versus OpenAI's $7.50.

Michael Cembalest J.P. Morgan
Date Posted:

DeepSeek’s V3 model, the basis for their R1 release, may have been trained 45x more efficiently than other leading-edge models. DeepSeek provides API access at $0.14 per million tokens while OpenAI charges $7.50 per million tokens.

How did one model achieve far greater training efficiency?

Core argument: DeepSeek appears to have trained its models 45x more efficiently than other leading-edge models, at a cost of $5.6M for 2.7M GPU hours, while Llama 3.1 405B used roughly 10x more GPU hours.

DeepSeek appears to have trained its models 45x more efficiently than other leading-edge models. To be clear, most of DeepSeek’s approaches already existed. Its greatest accomplishment: deploying them all at once in the face of a chip ban, and introducing self-reinforcement learning. GPT-3.5 uses its entire model for training/inference despite only needing small parts. GPT-4 and DeepSeek are a mixture of expert (MoE) models activating only necessary parts. DeepSeek V3 has 671B parameters, but just 37B are active at any time.MLA refers to “multi-head latent attention,” jargon for how DeepSeek runs with a smaller memory cache. Other efficiency approaches: parameters stored with BF16/FP32 are reduced to FP8 during training. Multi-token prediction (MTP) reduces accuracy ~10% but doubles inference speed. Training cost $5.6M for 2.7M GPU hours; Llama 3.1 405B used ~10x more GPU hours. DeepSeek API: $0.14/million tokens vs. OpenAI’s $7.50. Performance rivals OpenAI 4o and Anthropic’s Sonnet-3.5, better than Llama’s biggest model.

Takeaways by Macro Roundup® AI

  1. DeepSeek appears to have trained its models 45x more efficiently than other leading-edge models, at a cost of $5.6M for 2.7M GPU hours, while Llama 3.1 405B used roughly 10x more GPU hours.
  2. DeepSeek V3 holds 671B parameters but activates only 37B at any time as a mixture-of-experts model, alongside multi-head latent attention for a smaller memory cache, FP8 training precision, and multi-token prediction that cuts accuracy about 10% while doubling inference speed.
  3. DeepSeek's API charges $0.14 per million tokens against OpenAI's $7.50, with performance rivaling OpenAI 4o and Anthropic's Sonnet-3.5 and bettering Llama's biggest model, though most of its approaches already existed, its accomplishment being deploying them at once under a chip ban and introducing self-reinforcement learning.

Related Articles:

  • DeepSeek FAQ — Consumers and businesses stand to gain most from a future of effectively free AI products, along with big consumer tech companies, since a world of free AI makes product and distribution the decisive advantages they already hold. China also benefits, and the United States faces a choice between expanding chip export restrictions and permission-based regulation or allowing itself to compete.
  • The Year-End Wake-Up: The ‘Chinese Sputnik’ That Demands Our Attention — Chinese AI company DeepSeek trained its Version 3 large language model for roughly $5.6m, about 100x cheaper than Meta's Llama 3, with performance comparable to Claude 3.5 Sonnet and interface access costing 12x less than Anthropic's. The author argues US chip export controls backfired by spurring leaner engineering, and that the gap between open-source and closed-source models is narrowing rapidly.
  • A Severe Case of COVIDIA: Prognosis For An AI-Driven US Equity Market — Computing cycles move from infrastructure to platforms to applications, and NVIDIA data center revenues as a share of market-wide capital spending are projected to reach levels last seen at the 1969 mainframe peak and the dot-com boom. Large cloud providers spend more than twice as much training foundational models as running them for customers.

AI Summary. An analysis of more than 1,400 life narratives of older Americans, combining human readings with large language models, finds sources of life meaning are more varied than previous research suggested, with work and community contributions alongside family and close relationships. Women were significantly more likely than men to cite family, community, resilience, and religion.

David Lagakos, Stelios Michalopoulos and Hans-Joachim Voth National Bureau of Economic Research
Date Posted:

.@JoachimVoth @LagakosDavid reviewed oral histories of 1,400 older Depression-era Americans on life satisfaction and found that work and community contributions are on par with family in living a meaningful life for both men and women.

What do older Americans say gives their lives meaning?

Core argument: Drawing on over 1,400 life narratives of older Americans combining human readings with large language models, the authors find sources of life meaning more varied than prior research suggested, with work and community contributions alongside family and close relationships.

We exploit a unique corpus of over 1,400 life narratives of older Americans. We combine detailed human readings with large language models (LLMs) to extract systematic information on critical junctures, sources of meaning, and overall life satisfaction. Sources of life meaning are more varied than previous research suggested, underlining the importance of work and community contributions in addition to family and close relationships. The central importance of work among our narrators contrasts with the famous Harvard Study of Adult Development which emphasizes the paramount importance of relationships with other people. Sharp differences emerge when comparing what brings happiness and meaning across sub-groups of narrators. These appear in Figure 7, panel (b), which presents odds ratios by sub-group. Family and community were significantly more likely to be mentioned as a source of meaning by women, consistent with widely-held wisdom about gender roles in the late 19th and early 20th century. Resilience and religion were also markedly more common uplifting themes for women than for men.

Takeaways by Macro Roundup® AI

  1. Drawing on over 1,400 life narratives of older Americans combining human readings with large language models, the authors find sources of life meaning more varied than prior research suggested, with work and community contributions alongside family and close relationships.
  2. The central importance of work among the narrators contrasts with the Harvard Study of Adult Development, which emphasizes relationships with other people as paramount.
  3. Odds ratios by sub-group show family and community were significantly more likely to be mentioned as sources of meaning by women, and resilience and religion were markedly more common uplifting themes for women than for men.

Related Articles:

  • The Employment Effects of a Guaranteed Income: Experimental Evidence from Two U.S. States — In a randomized experiment, 1,000 low-income individuals received $1,000/month unconditionally for three years while a control group of 2,000 received $50/month, and the transfers reduced individual income excluding transfers by about $1,500/year. The authors find no significant effects on human capital investment and conclude the transfers produced a moderate labor supply reduction not offset by other productive activities.
  • Post-Pandemic Recovery for America’s Prime Age Labor Force: A Tale of Two Sexes — Prime-age women's labor force participation climbed about 5 points above its pre-pandemic level over a decade, while prime-age men's rose barely a point, only just regaining pre-pandemic levels. Despite economic expansion and labor shortages, work rates for prime-age American men are barely as high as at the trough of the severe Reagan-era recession.
  • The Anti-Social Century — Americans spend less time with other people than in any period with reliable data back to 1965, with in-person socializing down more than 20% from 2003 to 2023 according to the Bureau of Labor Statistics' American Time Use Survey. Among unmarried men and people under 25, the decline exceeded 35%.

AI Summary. US nuclear fusion start-up Helion has raised over $1bn at a $5.4bn valuation, including a $425mn round. The company says it will spend the funds expanding US manufacturing of large-scale pulse capacitors, which account for about a third of the cost of its Polaris reactor and were previously built overseas.

Malcolm Moore Financial Times
Date Posted:

Helion, a fusion startup with a contract to supply Microsoft with electricity by 2028, just raised $425M. The CEO says this funding keeps them “on course” to achieve that goal.

Will new funding help a fusion start-up bring reactor component manufacturing onshore?

Core argument: Helion, backed by Sam Altman and Peter Thiel, raised $425mn in a round that lifts its total funding above $1bn and values the nuclear fusion start-up at $5.4bn.

Helion has now raised more than $1bn and has a valuation of $5.4bn, including the latest fundraising round. The new funds would put it “on course” to achieve the goal, said CEO David Kirtley. Kirtley said the money would be spent on manufacturing pulse capacitors, an important component of its Polaris reactor in the US. “The capacitor is where about a third of the cost of the whole system goes, and about 85% of Polaris’ capacitors were built overseas. We bought them and waited several years to get them in-house,” he said. “We are now the first US manufacturer of large-scale pulse capacitors in decades, and we are going to expand that. So rather than waiting three years, we could get them in a year or less.”

Takeaways by Macro Roundup® AI

  1. Helion, backed by Sam Altman and Peter Thiel, raised $425mn in a round that lifts its total funding above $1bn and values the nuclear fusion start-up at $5.4bn.
  2. CEO David Kirtley said pulse capacitors account for about a third of the cost of Helion's entire system, and about 85% of the capacitors for its Polaris reactor were built overseas, arriving only after a multi-year wait.
  3. Kirtley said Helion is the first US manufacturer of large-scale pulse capacitors in decades and plans to expand that capacity to cut waits from three years to a year or less.

Related Articles:

  • China Outspends the U.S. on Fusion in the Race for Energy’s Holy Grail — Scientists familiar with China's fusion facilities estimate that at its current pace of spending and development, China will surpass U.S. and European magnetic fusion capabilities within three or four years. Much U.S. fusion spending goes to legacy programs, while China's roughly $1.5bn is largely directed at building systems that would compete with leading private U.S. fusion firms.
  • Microsoft Bets That Fusion Power Is Closer Than Many Think — Microsoft has agreed to buy fusion-generated electricity from startup Helion Energy in what is believed to be the first commercial fusion power agreement, even as many experts view fusion as decades away. Helion committed to begin producing fusion electricity within about five years and to supply at least 50 megawatts or pay penalties.
  • Fusion Start-Up Plans to Build Its First Power Plant in Virginia — Commonwealth Fusion Systems, the best-funded of the fusion start-ups, plans to build a fusion power plant in Virginia with a peak capacity of 400 megawatts, which the company says could power about 150,000 homes but would most likely serve large industrial customers. The company is first building a pilot machine in Massachusetts intended to demonstrate the technology's feasibility.