![MarketMetalysis's tweet photo. ok, lots of places i could start from, but ill lay out how i think about nominal bonds first so everything else is clearer. im not saying long duration cant rally or that the trade is stupid. What my point here is going to try is say the chain of assumptions being used to get from “AI works” to “the 20yr goes back to its 2010s average” is unrealistic.
nominal bond yields are broadly a function of:
expected inflation over the life of the bond
expected real short rates over the life of the bond, which expected growth/productivity feed into
expected Treasury issuance and the amount of duration the private market has to absorb
demand from the Fed, foreign official institutions and other buyers
term premium, which is a risk premium but bond dumbos needed a different name cus apparently duration is special
to be more technical, 3 and 4 are not separate additive parts of the yield in the same way expected inflation is. they generally work through expected rates, the amount of duration the market has to warehouse and the term premium. also, none of the term-premium components are directly observable and every model disagrees somewhat. im not pretending this is an exact machine, just that it is a much better starting point than “AI is deflationary, therefore the 20yr returns to its 2010s average.” [1]
the actual question is why yields were so low in the 2010s, why they trended down after 1980, and how much of that AI actually recreates.
everyone knows the basic 1970s story. america eventually broke the inflation regime, inflation expectations became much better anchored and inflation became less volatile. that pulled down both the expected nominal-rate path and the inflation-risk component of the term premium over a very long period.
this then came alongside globalization, reduced sensitivity to energy shocks, the end of the cold war and a long decline in macro uncertainty. globalization reduced prices and production costs in traded goods. the end of the cold war probably reduced some geopolitical/fiscal uncertainty at the margin too, although i wouldnt make that the load-bearing part of the argument.
then add falling estimates of the neutral real rate, demographics, a global savings glut, massive demand for safe assets, regulation which created structural demand for safe/liquid securities, and Treasuries becoming a good hedge against recessions and equity drawdowns. these things together gave you a ridiculously long bond bull market.
post-2000 there were several more factors which extended the bull market even further, and i think these matter cus they are exactly the parts the AI-deflation argument does not automatically recreate.
(images 1, 2)
first, reserves.
with globalization came massive amounts of trade, large current-account surpluses, managed exchange rates and, as a result, massive reserve accumulation. foreign official holdings of US Treasuries went from roughly $400b in 1994 to around $3t by 2010. central banks and reserve managers were recycling trade surpluses and FX intervention into Treasuries and agencies.
that type of buyer is important because a central bank buying for reserve-management or currency-policy reasons does not necessarily require the same yield as a private buyer using scarce or leveraged balance sheet.
the rapid reserve-accumulation impulse then faded through the 2010s. for many countries the existing reserve stock was already enormous and the marginal need for another dollar of reserves was much lower than it had been earlier in the cycle. current-account surpluses also changed, exchange-rate policies changed, capital flows reversed in some countries, and some countries used reserves to defend their currencies rather than continuing to accumulate them.
global FX reserves and foreign official Treasury holdings largely plateaued, while Treasury issuance did not. once official demand is no longer scaling with issuance, more duration has to clear through price-sensitive private balance sheets.
that includes households, mutual funds, banks, insurers, dealers, pensions, foreign private investors and levered relative-value funds. hedge funds are important here, especially in the basis trade and other relative-value strategies, but they did not simply replace foreign central banks one-for-one. the marginal buyer changes depending on the period.
during the 2017–19 Fed balance-sheet reduction, households and dealers absorbed additional supply, with foreign hedge funds appearing to account for part of the household-sector increase. during the first part of the post-2022 period, households, dealers, insurers and foreign buyers were more important, while hedge funds did not absorb the additional supply in the same way.
hedge funds have become a much larger participant again more recently, but a lot of their Treasury exposure is tied to the basis trade, rather than being a simple unhedged long-duration allocation. more supply therefore increasingly has to clear through private buyers who care about yield, financing cost, volatility, liquidity and how much balance sheet the position consumes. those buyers are fundamentally different from a foreign central bank recycling a dollar surplus for reserve or currency-policy reasons. [2]
(image 3)
second is the GFC and QE.
the Fed was obviously already a participant in the Treasury market before 2009, so it was not literally a brand-new buyer. what changed was the scale, maturity and purpose of the buying. QE created an enormous, relatively price-insensitive buyer which deliberately removed longer-duration Treasury and MBS risk from the private market while short rates were at zero.
one Fed estimate put the cumulative reduction in the 10yr term premium from the large-scale asset purchases and maturity-extension programme at around 100bps.
the GFC itself also created a massive negative shock through the world. it damaged balance sheets, reduced private investment and credit creation, lowered expected growth and inflation, and helped push estimates of the neutral real rate lower. then you had QE occurring not only in america but throughout other major developed markets, with zero or negative rates abroad and US duration being one of the few places where global investors could still get a positive nominal yield.
2020 then gave the bond bull its last leg. Emergency QE, zero rates, forced liquidity provision and an enormous risk-off shock all at once. after that, a lot of the underlying structure started reversing. QE became QT, Treasury supply grew rapidly and the long end had to clear at materially higher real yields and risk premiums.
third, post GFC.
the 2010s were not merely decent growth + weak wage pressure + low inflation. they were low inflation, low inflation volatility, a very low neutral real rate, huge central-bank balance sheets, large foreign-official and safe-asset demand, regulation-forced demand, low or negative term premiums and government bonds that were extremely valuable portfolio hedges.
AI could MAYBE recreate some amount of the disinflationary part. it does not automatically recreate the rest.
there is also a current-market point here which i think matters a lot. in one recent Fed decomposition of far-forward Treasury rates, the authors found that far-forward expected inflation and inflation risk had not increased much. they attributed most of the recent rise in the far-forward rate to a higher real risk premium, with perceived adverse supply shocks and future federal deficits among the possible explanations.
obviously this is one model and term-premium decomposition is not holy scripture. but if it is even directionally right, then saying “AI will fix inflation” may be attacking the wrong component of the long yield. even if AI lowers expected inflation, that does not automatically remove the real-rate or real-risk-premium component which is currently keeping the long end higher. [5]
now productivity, because this is one of the parts i disagree with most.
the line that the official statistics said the mobile-internet decade had “no productivity… LOL” is wrong. the official data very clearly did show a major IT/productivity acceleration from roughly 1995 to 2004. productivity then slowed substantially after that.
there are real measurement problems. quality improvements are difficult to measure, free digital goods can create large consumer benefits without having an obvious market price, and some innovation appears in consumer surplus rather than measured output. but research from the SF Fed did not find that worsening measurement error was remotely large enough to explain away the post-2004 productivity slowdown.
BEA also produced an experimental measure which included more free digital content. it increased estimated annual real GDP growth from 1.42% to 1.53% between 2005 and 2015. that is not nothing.
a lot of the benefit from smartphones, search engines, social media, maps, free messaging etc can make people materially better off without producing an equally large increase in measured market output per hour. that does not necessarily mean the productivity statistic failed. sometimes it means people are asking the statistic to measure consumer welfare, which is not exactly what it is designed to measure. [6]
but even granting the premise completely—say official productivity is terrible at seeing innovative booms in real time—the measured productivity print still isnt the first-order observable for this bond argument. the effects of the productivity are.
productivity is basically real output per hour. if AI allows the same amount of labour to produce materially more, then over time we should see some combination of higher real output or volumes relative to hours, slower unit-labour-cost growth, lower prices or slower inflation, higher margins, higher real wages, less labour input, or stronger investment and capital formation.
the measured productivity series is not the first-order variable for this trade. the first-order variables are the effects productivity has on output, costs, inflation and real rates.
if BLS fails to successfully label the boom “productivity,” it doesn't matter. the supposed disinflationary consequences still have to turn up somewhere else, arguably in some of those places before we can confidently identify the structural productivity break in a noisy and revised quarterly series.
but this also creates another problem for the bond bull case, cus productivity works through more than one channel. a persistent productivity shock can initially slow unit-labour-cost growth and inflation. it can also increase expected profits, permanent income, consumption and the marginal return on capital. that increases investment demand and can put upward pressure on the neutral real rate.
the Fed has described basically this exact mechanism. higher productivity means consumers expect higher future income and may save less today. firms see a higher return on capital and want to invest more. both can raise the equilibrium real rate, at least during the transition.
the Fed made the same point specifically in the context of AI in 2026: stronger productivity can increase aggregate supply and allow more output and wage growth without inflation, but the AI investment boom can also increase demand in the near term, and persistent productivity growth can raise the neutral rate through higher consumption and investment demand. [7]
so the nominal-bond result is not simply productivity up = inflation down = yields down. it is lower inflation pressure on one side versus potentially higher real rates, income and investment demand on the other.
this is also why “the buildout is inflationary but deployment is deflationary” is not enough, particularly over the next few years. the sequencing is not that clean. deployment itself creates more complementary capex: more electricity generation, transmission, data centres, networking, semiconductors, cooling, factories and eventually more robots.
even if the eventual ten-year endpoint is strongly disinflationary, the economy can spend years in the investment and demand-heavy part of the transition. “eventually deflationary” and “20yr yields return to their 2010s average over the next couple of years” are two completely different statements.
there is also a distinction between relative prices and aggregate inflation. the price of intelligence collapsing is a relative-price change. it does not mechanically mean the overall price level starts falling or even that aggregate inflation goes materially below target.
if the cost of an AI token drops 99%, firms might consume 100x more of them, improve quality, create new products and spend the savings somewhere else. the economy can receive an enormous real benefit while aggregate inflation remains broadly stable.
some productivity improvements create a one-time reduction in a particular price level or transfer income from workers to margins. sustained aggregate disinflation requires the supply effect to continue outrunning wage adjustment, induced demand, investment demand and the reaction of monetary policy. again, that is possible. it is just another step which has to be demonstrated rather than assumed.
to be clear, none of this means long duration cannot rally. TLT can rip on a recession, a risk-off event, a large labour-displacement demand shock, aggressive Fed easing or a collapse in the term premium.
what i am saying is that “AI works and is deflationary” does not by itself imply “the 20yr returns to its 2010s average.”
the 2010s were a stacked bond-bull regime. AI may recreate the disinflationary part, but you still need to explain why that overwhelms the real-rate effect, the investment boom, Treasury supply, the loss of some foreign-official and QE demand, and a real risk premium which may be elevated for reasons having very little to do with expected inflation.
some sources i thought might be useful
[1] nominal-yield decomposition and the historical decline in yields
https://t.co/cgwJik2gsi
[2] foreign official demand, reserve accumulation and the changing private buyer
https://t.co/frVzwqc5tk
Federal Reserve — Foreign Demand for U.S. Treasury Securities during the Pandemic
https://t.co/KiSLy9nC9w
Federal Reserve — Who Buys Treasuries When the Fed Reduces its Holdings?
https://t.co/AudZ8dnJCV
Federal Reserve — Decomposing Hedge Funds’ U.S. Treasury Exposures
https://t.co/mqCPS5BhVn
[3] QE and the term premium
Federal Reserve — The Effect of the Federal Reserve’s Securities Holdings on Longer-term Interest Rates
https://t.co/pB7od1TNEU
[5] why the current long end may be elevated for reasons other than inflation
Federal Reserve — Why Have Far-forward Nominal Treasury Rates Increased So Much in the Past Few Years?
https://t.co/WUmUEkJMce
[6] productivity measurement, revisions and the digital economy
Bureau of Labor Statistics — Revisions to BLS Quarterly Labor Productivity Estimates: How Large Are They?
https://t.co/GYvHftl2BV
San Francisco Fed — Does Growing Mismeasurement Explain Disappointing Growth?
https://t.co/5zTfm35AmJ
https://t.co/YqEhF004Zq
[7] productivity, inflation and the neutral real rate
Federal Reserve — What Drives Productivity Growth? Implications for the Economy and Prospects for the Future
https://t.co/jkcc18AaQn
Federal Reserve — Economic Outlook and Supply-Side (Dis)Inflation Dynamics
https://t.co/bjpAm4HLoO](https://pbs.twimg.com/media/HQS-DbsW8AACJSj.jpg)
![MarketMetalysis's tweet photo. ok, lots of places i could start from, but ill lay out how i think about nominal bonds first so everything else is clearer. im not saying long duration cant rally or that the trade is stupid. What my point here is going to try is say the chain of assumptions being used to get from “AI works” to “the 20yr goes back to its 2010s average” is unrealistic.
nominal bond yields are broadly a function of:
expected inflation over the life of the bond
expected real short rates over the life of the bond, which expected growth/productivity feed into
expected Treasury issuance and the amount of duration the private market has to absorb
demand from the Fed, foreign official institutions and other buyers
term premium, which is a risk premium but bond dumbos needed a different name cus apparently duration is special
to be more technical, 3 and 4 are not separate additive parts of the yield in the same way expected inflation is. they generally work through expected rates, the amount of duration the market has to warehouse and the term premium. also, none of the term-premium components are directly observable and every model disagrees somewhat. im not pretending this is an exact machine, just that it is a much better starting point than “AI is deflationary, therefore the 20yr returns to its 2010s average.” [1]
the actual question is why yields were so low in the 2010s, why they trended down after 1980, and how much of that AI actually recreates.
everyone knows the basic 1970s story. america eventually broke the inflation regime, inflation expectations became much better anchored and inflation became less volatile. that pulled down both the expected nominal-rate path and the inflation-risk component of the term premium over a very long period.
this then came alongside globalization, reduced sensitivity to energy shocks, the end of the cold war and a long decline in macro uncertainty. globalization reduced prices and production costs in traded goods. the end of the cold war probably reduced some geopolitical/fiscal uncertainty at the margin too, although i wouldnt make that the load-bearing part of the argument.
then add falling estimates of the neutral real rate, demographics, a global savings glut, massive demand for safe assets, regulation which created structural demand for safe/liquid securities, and Treasuries becoming a good hedge against recessions and equity drawdowns. these things together gave you a ridiculously long bond bull market.
post-2000 there were several more factors which extended the bull market even further, and i think these matter cus they are exactly the parts the AI-deflation argument does not automatically recreate.
(images 1, 2)
first, reserves.
with globalization came massive amounts of trade, large current-account surpluses, managed exchange rates and, as a result, massive reserve accumulation. foreign official holdings of US Treasuries went from roughly $400b in 1994 to around $3t by 2010. central banks and reserve managers were recycling trade surpluses and FX intervention into Treasuries and agencies.
that type of buyer is important because a central bank buying for reserve-management or currency-policy reasons does not necessarily require the same yield as a private buyer using scarce or leveraged balance sheet.
the rapid reserve-accumulation impulse then faded through the 2010s. for many countries the existing reserve stock was already enormous and the marginal need for another dollar of reserves was much lower than it had been earlier in the cycle. current-account surpluses also changed, exchange-rate policies changed, capital flows reversed in some countries, and some countries used reserves to defend their currencies rather than continuing to accumulate them.
global FX reserves and foreign official Treasury holdings largely plateaued, while Treasury issuance did not. once official demand is no longer scaling with issuance, more duration has to clear through price-sensitive private balance sheets.
that includes households, mutual funds, banks, insurers, dealers, pensions, foreign private investors and levered relative-value funds. hedge funds are important here, especially in the basis trade and other relative-value strategies, but they did not simply replace foreign central banks one-for-one. the marginal buyer changes depending on the period.
during the 2017–19 Fed balance-sheet reduction, households and dealers absorbed additional supply, with foreign hedge funds appearing to account for part of the household-sector increase. during the first part of the post-2022 period, households, dealers, insurers and foreign buyers were more important, while hedge funds did not absorb the additional supply in the same way.
hedge funds have become a much larger participant again more recently, but a lot of their Treasury exposure is tied to the basis trade, rather than being a simple unhedged long-duration allocation. more supply therefore increasingly has to clear through private buyers who care about yield, financing cost, volatility, liquidity and how much balance sheet the position consumes. those buyers are fundamentally different from a foreign central bank recycling a dollar surplus for reserve or currency-policy reasons. [2]
(image 3)
second is the GFC and QE.
the Fed was obviously already a participant in the Treasury market before 2009, so it was not literally a brand-new buyer. what changed was the scale, maturity and purpose of the buying. QE created an enormous, relatively price-insensitive buyer which deliberately removed longer-duration Treasury and MBS risk from the private market while short rates were at zero.
one Fed estimate put the cumulative reduction in the 10yr term premium from the large-scale asset purchases and maturity-extension programme at around 100bps.
the GFC itself also created a massive negative shock through the world. it damaged balance sheets, reduced private investment and credit creation, lowered expected growth and inflation, and helped push estimates of the neutral real rate lower. then you had QE occurring not only in america but throughout other major developed markets, with zero or negative rates abroad and US duration being one of the few places where global investors could still get a positive nominal yield.
2020 then gave the bond bull its last leg. Emergency QE, zero rates, forced liquidity provision and an enormous risk-off shock all at once. after that, a lot of the underlying structure started reversing. QE became QT, Treasury supply grew rapidly and the long end had to clear at materially higher real yields and risk premiums.
third, post GFC.
the 2010s were not merely decent growth + weak wage pressure + low inflation. they were low inflation, low inflation volatility, a very low neutral real rate, huge central-bank balance sheets, large foreign-official and safe-asset demand, regulation-forced demand, low or negative term premiums and government bonds that were extremely valuable portfolio hedges.
AI could MAYBE recreate some amount of the disinflationary part. it does not automatically recreate the rest.
there is also a current-market point here which i think matters a lot. in one recent Fed decomposition of far-forward Treasury rates, the authors found that far-forward expected inflation and inflation risk had not increased much. they attributed most of the recent rise in the far-forward rate to a higher real risk premium, with perceived adverse supply shocks and future federal deficits among the possible explanations.
obviously this is one model and term-premium decomposition is not holy scripture. but if it is even directionally right, then saying “AI will fix inflation” may be attacking the wrong component of the long yield. even if AI lowers expected inflation, that does not automatically remove the real-rate or real-risk-premium component which is currently keeping the long end higher. [5]
now productivity, because this is one of the parts i disagree with most.
the line that the official statistics said the mobile-internet decade had “no productivity… LOL” is wrong. the official data very clearly did show a major IT/productivity acceleration from roughly 1995 to 2004. productivity then slowed substantially after that.
there are real measurement problems. quality improvements are difficult to measure, free digital goods can create large consumer benefits without having an obvious market price, and some innovation appears in consumer surplus rather than measured output. but research from the SF Fed did not find that worsening measurement error was remotely large enough to explain away the post-2004 productivity slowdown.
BEA also produced an experimental measure which included more free digital content. it increased estimated annual real GDP growth from 1.42% to 1.53% between 2005 and 2015. that is not nothing.
a lot of the benefit from smartphones, search engines, social media, maps, free messaging etc can make people materially better off without producing an equally large increase in measured market output per hour. that does not necessarily mean the productivity statistic failed. sometimes it means people are asking the statistic to measure consumer welfare, which is not exactly what it is designed to measure. [6]
but even granting the premise completely—say official productivity is terrible at seeing innovative booms in real time—the measured productivity print still isnt the first-order observable for this bond argument. the effects of the productivity are.
productivity is basically real output per hour. if AI allows the same amount of labour to produce materially more, then over time we should see some combination of higher real output or volumes relative to hours, slower unit-labour-cost growth, lower prices or slower inflation, higher margins, higher real wages, less labour input, or stronger investment and capital formation.
the measured productivity series is not the first-order variable for this trade. the first-order variables are the effects productivity has on output, costs, inflation and real rates.
if BLS fails to successfully label the boom “productivity,” it doesn't matter. the supposed disinflationary consequences still have to turn up somewhere else, arguably in some of those places before we can confidently identify the structural productivity break in a noisy and revised quarterly series.
but this also creates another problem for the bond bull case, cus productivity works through more than one channel. a persistent productivity shock can initially slow unit-labour-cost growth and inflation. it can also increase expected profits, permanent income, consumption and the marginal return on capital. that increases investment demand and can put upward pressure on the neutral real rate.
the Fed has described basically this exact mechanism. higher productivity means consumers expect higher future income and may save less today. firms see a higher return on capital and want to invest more. both can raise the equilibrium real rate, at least during the transition.
the Fed made the same point specifically in the context of AI in 2026: stronger productivity can increase aggregate supply and allow more output and wage growth without inflation, but the AI investment boom can also increase demand in the near term, and persistent productivity growth can raise the neutral rate through higher consumption and investment demand. [7]
so the nominal-bond result is not simply productivity up = inflation down = yields down. it is lower inflation pressure on one side versus potentially higher real rates, income and investment demand on the other.
this is also why “the buildout is inflationary but deployment is deflationary” is not enough, particularly over the next few years. the sequencing is not that clean. deployment itself creates more complementary capex: more electricity generation, transmission, data centres, networking, semiconductors, cooling, factories and eventually more robots.
even if the eventual ten-year endpoint is strongly disinflationary, the economy can spend years in the investment and demand-heavy part of the transition. “eventually deflationary” and “20yr yields return to their 2010s average over the next couple of years” are two completely different statements.
there is also a distinction between relative prices and aggregate inflation. the price of intelligence collapsing is a relative-price change. it does not mechanically mean the overall price level starts falling or even that aggregate inflation goes materially below target.
if the cost of an AI token drops 99%, firms might consume 100x more of them, improve quality, create new products and spend the savings somewhere else. the economy can receive an enormous real benefit while aggregate inflation remains broadly stable.
some productivity improvements create a one-time reduction in a particular price level or transfer income from workers to margins. sustained aggregate disinflation requires the supply effect to continue outrunning wage adjustment, induced demand, investment demand and the reaction of monetary policy. again, that is possible. it is just another step which has to be demonstrated rather than assumed.
to be clear, none of this means long duration cannot rally. TLT can rip on a recession, a risk-off event, a large labour-displacement demand shock, aggressive Fed easing or a collapse in the term premium.
what i am saying is that “AI works and is deflationary” does not by itself imply “the 20yr returns to its 2010s average.”
the 2010s were a stacked bond-bull regime. AI may recreate the disinflationary part, but you still need to explain why that overwhelms the real-rate effect, the investment boom, Treasury supply, the loss of some foreign-official and QE demand, and a real risk premium which may be elevated for reasons having very little to do with expected inflation.
some sources i thought might be useful
[1] nominal-yield decomposition and the historical decline in yields
https://t.co/cgwJik2gsi
[2] foreign official demand, reserve accumulation and the changing private buyer
https://t.co/frVzwqc5tk
Federal Reserve — Foreign Demand for U.S. Treasury Securities during the Pandemic
https://t.co/KiSLy9nC9w
Federal Reserve — Who Buys Treasuries When the Fed Reduces its Holdings?
https://t.co/AudZ8dnJCV
Federal Reserve — Decomposing Hedge Funds’ U.S. Treasury Exposures
https://t.co/mqCPS5BhVn
[3] QE and the term premium
Federal Reserve — The Effect of the Federal Reserve’s Securities Holdings on Longer-term Interest Rates
https://t.co/pB7od1TNEU
[5] why the current long end may be elevated for reasons other than inflation
Federal Reserve — Why Have Far-forward Nominal Treasury Rates Increased So Much in the Past Few Years?
https://t.co/WUmUEkJMce
[6] productivity measurement, revisions and the digital economy
Bureau of Labor Statistics — Revisions to BLS Quarterly Labor Productivity Estimates: How Large Are They?
https://t.co/GYvHftl2BV
San Francisco Fed — Does Growing Mismeasurement Explain Disappointing Growth?
https://t.co/5zTfm35AmJ
https://t.co/YqEhF004Zq
[7] productivity, inflation and the neutral real rate
Federal Reserve — What Drives Productivity Growth? Implications for the Economy and Prospects for the Future
https://t.co/jkcc18AaQn
Federal Reserve — Economic Outlook and Supply-Side (Dis)Inflation Dynamics
https://t.co/bjpAm4HLoO](https://pbs.twimg.com/media/HQS-DbpW0AEQ5KK.jpg)
Olivia
Online
MarketMetalysis
@MarketMetalysis
Stock research, Semiconductors, Oil, Tech, Macro econ.
Joined March 2024
31 Following
36 Followers
354 Posts
ok, lots of places i could start from, but ill lay out how i think about nominal bonds first so everything else is clearer. im not saying long duration cant rally or that the trade is stupid. What my point here is going to try is say the chain of assumptions being used to get from “AI works” to “the 20yr goes back to its 2010s average” is unrealistic.
nominal bond yields are broadly a function of:
expected inflation over the life of the bond
expected real short rates over the life of the bond, which expected growth/productivity feed into
expected Treasury issuance and the amount of duration the private market has to absorb
demand from the Fed, foreign official institutions and other buyers
term premium, which is a risk premium but bond dumbos needed a different name cus apparently duration is special
to be more technical, 3 and 4 are not separate additive parts of the yield in the same way expected inflation is. they generally work through expected rates, the amount of duration the market has to warehouse and the term premium. also, none of the term-premium components are directly observable and every model disagrees somewhat. im not pretending this is an exact machine, just that it is a much better starting point than “AI is deflationary, therefore the 20yr returns to its 2010s average.” [1]
the actual question is why yields were so low in the 2010s, why they trended down after 1980, and how much of that AI actually recreates.
everyone knows the basic 1970s story. america eventually broke the inflation regime, inflation expectations became much better anchored and inflation became less volatile. that pulled down both the expected nominal-rate path and the inflation-risk component of the term premium over a very long period.
this then came alongside globalization, reduced sensitivity to energy shocks, the end of the cold war and a long decline in macro uncertainty. globalization reduced prices and production costs in traded goods. the end of the cold war probably reduced some geopolitical/fiscal uncertainty at the margin too, although i wouldnt make that the load-bearing part of the argument.
then add falling estimates of the neutral real rate, demographics, a global savings glut, massive demand for safe assets, regulation which created structural demand for safe/liquid securities, and Treasuries becoming a good hedge against recessions and equity drawdowns. these things together gave you a ridiculously long bond bull market.
post-2000 there were several more factors which extended the bull market even further, and i think these matter cus they are exactly the parts the AI-deflation argument does not automatically recreate.
(images 1, 2)
first, reserves.
with globalization came massive amounts of trade, large current-account surpluses, managed exchange rates and, as a result, massive reserve accumulation. foreign official holdings of US Treasuries went from roughly $400b in 1994 to around $3t by 2010. central banks and reserve managers were recycling trade surpluses and FX intervention into Treasuries and agencies.
that type of buyer is important because a central bank buying for reserve-management or currency-policy reasons does not necessarily require the same yield as a private buyer using scarce or leveraged balance sheet.
the rapid reserve-accumulation impulse then faded through the 2010s. for many countries the existing reserve stock was already enormous and the marginal need for another dollar of reserves was much lower than it had been earlier in the cycle. current-account surpluses also changed, exchange-rate policies changed, capital flows reversed in some countries, and some countries used reserves to defend their currencies rather than continuing to accumulate them.
global FX reserves and foreign official Treasury holdings largely plateaued, while Treasury issuance did not. once official demand is no longer scaling with issuance, more duration has to clear through price-sensitive private balance sheets.
that includes households, mutual funds, banks, insurers, dealers, pensions, foreign private investors and levered relative-value funds. hedge funds are important here, especially in the basis trade and other relative-value strategies, but they did not simply replace foreign central banks one-for-one. the marginal buyer changes depending on the period.
during the 2017–19 Fed balance-sheet reduction, households and dealers absorbed additional supply, with foreign hedge funds appearing to account for part of the household-sector increase. during the first part of the post-2022 period, households, dealers, insurers and foreign buyers were more important, while hedge funds did not absorb the additional supply in the same way.
hedge funds have become a much larger participant again more recently, but a lot of their Treasury exposure is tied to the basis trade, rather than being a simple unhedged long-duration allocation. more supply therefore increasingly has to clear through private buyers who care about yield, financing cost, volatility, liquidity and how much balance sheet the position consumes. those buyers are fundamentally different from a foreign central bank recycling a dollar surplus for reserve or currency-policy reasons. [2]
(image 3)
second is the GFC and QE.
the Fed was obviously already a participant in the Treasury market before 2009, so it was not literally a brand-new buyer. what changed was the scale, maturity and purpose of the buying. QE created an enormous, relatively price-insensitive buyer which deliberately removed longer-duration Treasury and MBS risk from the private market while short rates were at zero.
one Fed estimate put the cumulative reduction in the 10yr term premium from the large-scale asset purchases and maturity-extension programme at around 100bps.
the GFC itself also created a massive negative shock through the world. it damaged balance sheets, reduced private investment and credit creation, lowered expected growth and inflation, and helped push estimates of the neutral real rate lower. then you had QE occurring not only in america but throughout other major developed markets, with zero or negative rates abroad and US duration being one of the few places where global investors could still get a positive nominal yield.
2020 then gave the bond bull its last leg. Emergency QE, zero rates, forced liquidity provision and an enormous risk-off shock all at once. after that, a lot of the underlying structure started reversing. QE became QT, Treasury supply grew rapidly and the long end had to clear at materially higher real yields and risk premiums.
third, post GFC.
the 2010s were not merely decent growth + weak wage pressure + low inflation. they were low inflation, low inflation volatility, a very low neutral real rate, huge central-bank balance sheets, large foreign-official and safe-asset demand, regulation-forced demand, low or negative term premiums and government bonds that were extremely valuable portfolio hedges.
AI could MAYBE recreate some amount of the disinflationary part. it does not automatically recreate the rest.
there is also a current-market point here which i think matters a lot. in one recent Fed decomposition of far-forward Treasury rates, the authors found that far-forward expected inflation and inflation risk had not increased much. they attributed most of the recent rise in the far-forward rate to a higher real risk premium, with perceived adverse supply shocks and future federal deficits among the possible explanations.
obviously this is one model and term-premium decomposition is not holy scripture. but if it is even directionally right, then saying “AI will fix inflation” may be attacking the wrong component of the long yield. even if AI lowers expected inflation, that does not automatically remove the real-rate or real-risk-premium component which is currently keeping the long end higher. [5]
now productivity, because this is one of the parts i disagree with most.
the line that the official statistics said the mobile-internet decade had “no productivity… LOL” is wrong. the official data very clearly did show a major IT/productivity acceleration from roughly 1995 to 2004. productivity then slowed substantially after that.
there are real measurement problems. quality improvements are difficult to measure, free digital goods can create large consumer benefits without having an obvious market price, and some innovation appears in consumer surplus rather than measured output. but research from the SF Fed did not find that worsening measurement error was remotely large enough to explain away the post-2004 productivity slowdown.
BEA also produced an experimental measure which included more free digital content. it increased estimated annual real GDP growth from 1.42% to 1.53% between 2005 and 2015. that is not nothing.
a lot of the benefit from smartphones, search engines, social media, maps, free messaging etc can make people materially better off without producing an equally large increase in measured market output per hour. that does not necessarily mean the productivity statistic failed. sometimes it means people are asking the statistic to measure consumer welfare, which is not exactly what it is designed to measure. [6]
but even granting the premise completely—say official productivity is terrible at seeing innovative booms in real time—the measured productivity print still isnt the first-order observable for this bond argument. the effects of the productivity are.
productivity is basically real output per hour. if AI allows the same amount of labour to produce materially more, then over time we should see some combination of higher real output or volumes relative to hours, slower unit-labour-cost growth, lower prices or slower inflation, higher margins, higher real wages, less labour input, or stronger investment and capital formation.
the measured productivity series is not the first-order variable for this trade. the first-order variables are the effects productivity has on output, costs, inflation and real rates.
if BLS fails to successfully label the boom “productivity,” it doesn't matter. the supposed disinflationary consequences still have to turn up somewhere else, arguably in some of those places before we can confidently identify the structural productivity break in a noisy and revised quarterly series.
but this also creates another problem for the bond bull case, cus productivity works through more than one channel. a persistent productivity shock can initially slow unit-labour-cost growth and inflation. it can also increase expected profits, permanent income, consumption and the marginal return on capital. that increases investment demand and can put upward pressure on the neutral real rate.
the Fed has described basically this exact mechanism. higher productivity means consumers expect higher future income and may save less today. firms see a higher return on capital and want to invest more. both can raise the equilibrium real rate, at least during the transition.
the Fed made the same point specifically in the context of AI in 2026: stronger productivity can increase aggregate supply and allow more output and wage growth without inflation, but the AI investment boom can also increase demand in the near term, and persistent productivity growth can raise the neutral rate through higher consumption and investment demand. [7]
so the nominal-bond result is not simply productivity up = inflation down = yields down. it is lower inflation pressure on one side versus potentially higher real rates, income and investment demand on the other.
this is also why “the buildout is inflationary but deployment is deflationary” is not enough, particularly over the next few years. the sequencing is not that clean. deployment itself creates more complementary capex: more electricity generation, transmission, data centres, networking, semiconductors, cooling, factories and eventually more robots.
even if the eventual ten-year endpoint is strongly disinflationary, the economy can spend years in the investment and demand-heavy part of the transition. “eventually deflationary” and “20yr yields return to their 2010s average over the next couple of years” are two completely different statements.
there is also a distinction between relative prices and aggregate inflation. the price of intelligence collapsing is a relative-price change. it does not mechanically mean the overall price level starts falling or even that aggregate inflation goes materially below target.
if the cost of an AI token drops 99%, firms might consume 100x more of them, improve quality, create new products and spend the savings somewhere else. the economy can receive an enormous real benefit while aggregate inflation remains broadly stable.
some productivity improvements create a one-time reduction in a particular price level or transfer income from workers to margins. sustained aggregate disinflation requires the supply effect to continue outrunning wage adjustment, induced demand, investment demand and the reaction of monetary policy. again, that is possible. it is just another step which has to be demonstrated rather than assumed.
to be clear, none of this means long duration cannot rally. TLT can rip on a recession, a risk-off event, a large labour-displacement demand shock, aggressive Fed easing or a collapse in the term premium.
what i am saying is that “AI works and is deflationary” does not by itself imply “the 20yr returns to its 2010s average.”
the 2010s were a stacked bond-bull regime. AI may recreate the disinflationary part, but you still need to explain why that overwhelms the real-rate effect, the investment boom, Treasury supply, the loss of some foreign-official and QE demand, and a real risk premium which may be elevated for reasons having very little to do with expected inflation.
some sources i thought might be useful
[1] nominal-yield decomposition and the historical decline in yields
https://t.co/cgwJik2gsi
[2] foreign official demand, reserve accumulation and the changing private buyer
https://t.co/frVzwqc5tk
Federal Reserve — Foreign Demand for U.S. Treasury Securities during the Pandemic
https://t.co/KiSLy9nC9w
Federal Reserve — Who Buys Treasuries When the Fed Reduces its Holdings?
https://t.co/AudZ8dnJCV
Federal Reserve — Decomposing Hedge Funds’ U.S. Treasury Exposures
https://t.co/mqCPS5BhVn
[3] QE and the term premium
Federal Reserve — The Effect of the Federal Reserve’s Securities Holdings on Longer-term Interest Rates
https://t.co/pB7od1TNEU
[5] why the current long end may be elevated for reasons other than inflation
Federal Reserve — Why Have Far-forward Nominal Treasury Rates Increased So Much in the Past Few Years?
https://t.co/WUmUEkJMce
[6] productivity measurement, revisions and the digital economy
Bureau of Labor Statistics — Revisions to BLS Quarterly Labor Productivity Estimates: How Large Are They?
https://t.co/GYvHftl2BV
San Francisco Fed — Does Growing Mismeasurement Explain Disappointing Growth?
https://t.co/5zTfm35AmJ
https://t.co/YqEhF004Zq
[7] productivity, inflation and the neutral real rate
Federal Reserve — What Drives Productivity Growth? Implications for the Economy and Prospects for the Future
https://t.co/jkcc18AaQn
Federal Reserve — Economic Outlook and Supply-Side (Dis)Inflation Dynamics
https://t.co/bjpAm4HLoO
![MarketMetalysis's tweet photo. ok, lots of places i could start from, but ill lay out how i think about nominal bonds first so everything else is clearer. im not saying long duration cant rally or that the trade is stupid. What my point here is going to try is say the chain of assumptions being used to get from “AI works” to “the 20yr goes back to its 2010s average” is unrealistic.
nominal bond yields are broadly a function of:
expected inflation over the life of the bond
expected real short rates over the life of the bond, which expected growth/productivity feed into
expected Treasury issuance and the amount of duration the private market has to absorb
demand from the Fed, foreign official institutions and other buyers
term premium, which is a risk premium but bond dumbos needed a different name cus apparently duration is special
to be more technical, 3 and 4 are not separate additive parts of the yield in the same way expected inflation is. they generally work through expected rates, the amount of duration the market has to warehouse and the term premium. also, none of the term-premium components are directly observable and every model disagrees somewhat. im not pretending this is an exact machine, just that it is a much better starting point than “AI is deflationary, therefore the 20yr returns to its 2010s average.” [1]
the actual question is why yields were so low in the 2010s, why they trended down after 1980, and how much of that AI actually recreates.
everyone knows the basic 1970s story. america eventually broke the inflation regime, inflation expectations became much better anchored and inflation became less volatile. that pulled down both the expected nominal-rate path and the inflation-risk component of the term premium over a very long period.
this then came alongside globalization, reduced sensitivity to energy shocks, the end of the cold war and a long decline in macro uncertainty. globalization reduced prices and production costs in traded goods. the end of the cold war probably reduced some geopolitical/fiscal uncertainty at the margin too, although i wouldnt make that the load-bearing part of the argument.
then add falling estimates of the neutral real rate, demographics, a global savings glut, massive demand for safe assets, regulation which created structural demand for safe/liquid securities, and Treasuries becoming a good hedge against recessions and equity drawdowns. these things together gave you a ridiculously long bond bull market.
post-2000 there were several more factors which extended the bull market even further, and i think these matter cus they are exactly the parts the AI-deflation argument does not automatically recreate.
(images 1, 2)
first, reserves.
with globalization came massive amounts of trade, large current-account surpluses, managed exchange rates and, as a result, massive reserve accumulation. foreign official holdings of US Treasuries went from roughly $400b in 1994 to around $3t by 2010. central banks and reserve managers were recycling trade surpluses and FX intervention into Treasuries and agencies.
that type of buyer is important because a central bank buying for reserve-management or currency-policy reasons does not necessarily require the same yield as a private buyer using scarce or leveraged balance sheet.
the rapid reserve-accumulation impulse then faded through the 2010s. for many countries the existing reserve stock was already enormous and the marginal need for another dollar of reserves was much lower than it had been earlier in the cycle. current-account surpluses also changed, exchange-rate policies changed, capital flows reversed in some countries, and some countries used reserves to defend their currencies rather than continuing to accumulate them.
global FX reserves and foreign official Treasury holdings largely plateaued, while Treasury issuance did not. once official demand is no longer scaling with issuance, more duration has to clear through price-sensitive private balance sheets.
that includes households, mutual funds, banks, insurers, dealers, pensions, foreign private investors and levered relative-value funds. hedge funds are important here, especially in the basis trade and other relative-value strategies, but they did not simply replace foreign central banks one-for-one. the marginal buyer changes depending on the period.
during the 2017–19 Fed balance-sheet reduction, households and dealers absorbed additional supply, with foreign hedge funds appearing to account for part of the household-sector increase. during the first part of the post-2022 period, households, dealers, insurers and foreign buyers were more important, while hedge funds did not absorb the additional supply in the same way.
hedge funds have become a much larger participant again more recently, but a lot of their Treasury exposure is tied to the basis trade, rather than being a simple unhedged long-duration allocation. more supply therefore increasingly has to clear through private buyers who care about yield, financing cost, volatility, liquidity and how much balance sheet the position consumes. those buyers are fundamentally different from a foreign central bank recycling a dollar surplus for reserve or currency-policy reasons. [2]
(image 3)
second is the GFC and QE.
the Fed was obviously already a participant in the Treasury market before 2009, so it was not literally a brand-new buyer. what changed was the scale, maturity and purpose of the buying. QE created an enormous, relatively price-insensitive buyer which deliberately removed longer-duration Treasury and MBS risk from the private market while short rates were at zero.
one Fed estimate put the cumulative reduction in the 10yr term premium from the large-scale asset purchases and maturity-extension programme at around 100bps.
the GFC itself also created a massive negative shock through the world. it damaged balance sheets, reduced private investment and credit creation, lowered expected growth and inflation, and helped push estimates of the neutral real rate lower. then you had QE occurring not only in america but throughout other major developed markets, with zero or negative rates abroad and US duration being one of the few places where global investors could still get a positive nominal yield.
2020 then gave the bond bull its last leg. Emergency QE, zero rates, forced liquidity provision and an enormous risk-off shock all at once. after that, a lot of the underlying structure started reversing. QE became QT, Treasury supply grew rapidly and the long end had to clear at materially higher real yields and risk premiums.
third, post GFC.
the 2010s were not merely decent growth + weak wage pressure + low inflation. they were low inflation, low inflation volatility, a very low neutral real rate, huge central-bank balance sheets, large foreign-official and safe-asset demand, regulation-forced demand, low or negative term premiums and government bonds that were extremely valuable portfolio hedges.
AI could MAYBE recreate some amount of the disinflationary part. it does not automatically recreate the rest.
there is also a current-market point here which i think matters a lot. in one recent Fed decomposition of far-forward Treasury rates, the authors found that far-forward expected inflation and inflation risk had not increased much. they attributed most of the recent rise in the far-forward rate to a higher real risk premium, with perceived adverse supply shocks and future federal deficits among the possible explanations.
obviously this is one model and term-premium decomposition is not holy scripture. but if it is even directionally right, then saying “AI will fix inflation” may be attacking the wrong component of the long yield. even if AI lowers expected inflation, that does not automatically remove the real-rate or real-risk-premium component which is currently keeping the long end higher. [5]
now productivity, because this is one of the parts i disagree with most.
the line that the official statistics said the mobile-internet decade had “no productivity… LOL” is wrong. the official data very clearly did show a major IT/productivity acceleration from roughly 1995 to 2004. productivity then slowed substantially after that.
there are real measurement problems. quality improvements are difficult to measure, free digital goods can create large consumer benefits without having an obvious market price, and some innovation appears in consumer surplus rather than measured output. but research from the SF Fed did not find that worsening measurement error was remotely large enough to explain away the post-2004 productivity slowdown.
BEA also produced an experimental measure which included more free digital content. it increased estimated annual real GDP growth from 1.42% to 1.53% between 2005 and 2015. that is not nothing.
a lot of the benefit from smartphones, search engines, social media, maps, free messaging etc can make people materially better off without producing an equally large increase in measured market output per hour. that does not necessarily mean the productivity statistic failed. sometimes it means people are asking the statistic to measure consumer welfare, which is not exactly what it is designed to measure. [6]
but even granting the premise completely—say official productivity is terrible at seeing innovative booms in real time—the measured productivity print still isnt the first-order observable for this bond argument. the effects of the productivity are.
productivity is basically real output per hour. if AI allows the same amount of labour to produce materially more, then over time we should see some combination of higher real output or volumes relative to hours, slower unit-labour-cost growth, lower prices or slower inflation, higher margins, higher real wages, less labour input, or stronger investment and capital formation.
the measured productivity series is not the first-order variable for this trade. the first-order variables are the effects productivity has on output, costs, inflation and real rates.
if BLS fails to successfully label the boom “productivity,” it doesn't matter. the supposed disinflationary consequences still have to turn up somewhere else, arguably in some of those places before we can confidently identify the structural productivity break in a noisy and revised quarterly series.
but this also creates another problem for the bond bull case, cus productivity works through more than one channel. a persistent productivity shock can initially slow unit-labour-cost growth and inflation. it can also increase expected profits, permanent income, consumption and the marginal return on capital. that increases investment demand and can put upward pressure on the neutral real rate.
the Fed has described basically this exact mechanism. higher productivity means consumers expect higher future income and may save less today. firms see a higher return on capital and want to invest more. both can raise the equilibrium real rate, at least during the transition.
the Fed made the same point specifically in the context of AI in 2026: stronger productivity can increase aggregate supply and allow more output and wage growth without inflation, but the AI investment boom can also increase demand in the near term, and persistent productivity growth can raise the neutral rate through higher consumption and investment demand. [7]
so the nominal-bond result is not simply productivity up = inflation down = yields down. it is lower inflation pressure on one side versus potentially higher real rates, income and investment demand on the other.
this is also why “the buildout is inflationary but deployment is deflationary” is not enough, particularly over the next few years. the sequencing is not that clean. deployment itself creates more complementary capex: more electricity generation, transmission, data centres, networking, semiconductors, cooling, factories and eventually more robots.
even if the eventual ten-year endpoint is strongly disinflationary, the economy can spend years in the investment and demand-heavy part of the transition. “eventually deflationary” and “20yr yields return to their 2010s average over the next couple of years” are two completely different statements.
there is also a distinction between relative prices and aggregate inflation. the price of intelligence collapsing is a relative-price change. it does not mechanically mean the overall price level starts falling or even that aggregate inflation goes materially below target.
if the cost of an AI token drops 99%, firms might consume 100x more of them, improve quality, create new products and spend the savings somewhere else. the economy can receive an enormous real benefit while aggregate inflation remains broadly stable.
some productivity improvements create a one-time reduction in a particular price level or transfer income from workers to margins. sustained aggregate disinflation requires the supply effect to continue outrunning wage adjustment, induced demand, investment demand and the reaction of monetary policy. again, that is possible. it is just another step which has to be demonstrated rather than assumed.
to be clear, none of this means long duration cannot rally. TLT can rip on a recession, a risk-off event, a large labour-displacement demand shock, aggressive Fed easing or a collapse in the term premium.
what i am saying is that “AI works and is deflationary” does not by itself imply “the 20yr returns to its 2010s average.”
the 2010s were a stacked bond-bull regime. AI may recreate the disinflationary part, but you still need to explain why that overwhelms the real-rate effect, the investment boom, Treasury supply, the loss of some foreign-official and QE demand, and a real risk premium which may be elevated for reasons having very little to do with expected inflation.
some sources i thought might be useful
[1] nominal-yield decomposition and the historical decline in yields
https://t.co/cgwJik2gsi
[2] foreign official demand, reserve accumulation and the changing private buyer
https://t.co/frVzwqc5tk
Federal Reserve — Foreign Demand for U.S. Treasury Securities during the Pandemic
https://t.co/KiSLy9nC9w
Federal Reserve — Who Buys Treasuries When the Fed Reduces its Holdings?
https://t.co/AudZ8dnJCV
Federal Reserve — Decomposing Hedge Funds’ U.S. Treasury Exposures
https://t.co/mqCPS5BhVn
[3] QE and the term premium
Federal Reserve — The Effect of the Federal Reserve’s Securities Holdings on Longer-term Interest Rates
https://t.co/pB7od1TNEU
[5] why the current long end may be elevated for reasons other than inflation
Federal Reserve — Why Have Far-forward Nominal Treasury Rates Increased So Much in the Past Few Years?
https://t.co/WUmUEkJMce
[6] productivity measurement, revisions and the digital economy
Bureau of Labor Statistics — Revisions to BLS Quarterly Labor Productivity Estimates: How Large Are They?
https://t.co/GYvHftl2BV
San Francisco Fed — Does Growing Mismeasurement Explain Disappointing Growth?
https://t.co/5zTfm35AmJ
https://t.co/YqEhF004Zq
[7] productivity, inflation and the neutral real rate
Federal Reserve — What Drives Productivity Growth? Implications for the Economy and Prospects for the Future
https://t.co/jkcc18AaQn
Federal Reserve — Economic Outlook and Supply-Side (Dis)Inflation Dynamics
https://t.co/bjpAm4HLoO](https://pbs.twimg.com/media/HQS-DbzXIAArWve.jpg)
ok, this is very confusing post. are you saying earnings as in actual after tax earnings? and when you say excess are you comparing to potential extrapolated pre ai growth? because there are so many issues if you are comparing it to an extrapolated possibility due to them possibly having share loss in a scenario they did 0 capex.
@dampedspring u wana add sub systems really they have the most leverage to buildout bullwhips. so MKSI ICHR UCTT. also would weight more to memory share ones, less logic node. AMAT LRCX kokusai electric
@HFI_Research I think you need to accept a scenario that you're completely wrong on directionality of oil price because of hormuz, its reasonably possible china continues to use SPR for continued months and wait for cheap oil to ramp back up, demand in poorer nations could have easilly been negatively effected by this crisis and will have lasting effects as well.
if even few mbpd more of oil starts coming out of strait its going to make up for the US exports, especially as its going to be heavier grades which is what the world was starved for. also shorter routes to asian countries.
you have been in the oil market long enough to know when oil sells off it sells off hard and fast and leaves everyone behind in fundamental land, which is exactly whats happening right now and also what happened in 2022 when everyone was screaming about oil draws PAST the peak in summer as we made new lows into december then went sideways.
the risk of war continuing is 100% a reality, i dont think isreal is going to stop, but iran before the war was very lenient on responses and isnt going to risk kinetic war with its own country to save its organizations around the region, since the supreme leader has been reported as coming back its definitely true that the IRGC hardliners impact onto negotiations has calmed down, the intensity of the firing at boats in the strait, and especially the intensity of the responses to US aggression have calmed, IRGC definitely have more power but the facts point to less then mid war which does make sense because supreme leader does have the power of the people and the IRGC isnt going to risk a internal war just to continue waring against america when they are now way more empowered and can build up strength.
@BobEUnlimited @thedailyshot same for all asian currencies, probably mix between china yuan strengthening and complete inaction from central banks to strengthen their currencies against the USD in fear of some sort of retaliation by trump maybe
@dampedspring and iran now has massively upped political power in the middle east implicitly by already taking control of the hormuz once. which just further cements breaking away from any deal made now in the future.
@ncitayim because these accounts are managed by someone in america not the actual people, people have tracked him down.
@dampedspring all of their AI division people left for oai years ago, politics has destroyed the ability for them to keep up in AI as no one wants to work there even if money is big because its absolutely terrible environment.
@ncitayim is this person trustable source of information?
demand shifting also relies on it actually being able to be supplemented, we would also see it in other markets as it would have signficant short term impact moving even 500kbpd worth of demand of a oil to another source like coal/electricity demand via EV.
on EV by itself i dont think it has any meaning to the discussion for oil demand, EV adoption has been rigirously studied in european states and asian states, it doesn't 1-1 work as a replacement for oil demand especially heavier products.
the biggest point against EV helping this demand shift though is the fact that china exports of vehicles (primarilly ev) has been making record numbers literally every month since 2020, till last 2 months, which is kinda suspect in of itself but ignoring that, all of those EV's have had to find an owner and this hasn't resulted in materially lower demand for oil in that time period. https://t.co/ynqxSQ7QGG
to get the argument of EV shifting helping demand you have to work some kind of scenario where ICE vehicles were set to the side by owners and they switched to EV's but most people especially asian dont own multiple vehicles and just switch based on the prevailing oil price, its a wealth level sort of thing where wealthier groups own EV's.
i would further argue against other sources of demand shifting, there is definitely arguments for industrial switching to coal or other but this is also only if they CAN be substituted. China and other industrial powerhouses arent burning diesel to run their factories, they could definitely be using petroleum based feedstocks as part of the process though, which i think is more likely the answer, the industrial facilities are just lowering utilization gradually as they burn through supplies, or turn off completely like some examples in asia petrochemical complex have.
@dampedspring https://t.co/RWewlEdgDS
@dampedspring isnt it a short timeframe since you first posited this idea? like 1-1.5 months ish right
@ncitayim no sell side firm can tell its clients the strait is guna stay closed for longer than expected because then they will look stupid, they look less stupid if they continually pushback the timeline.
@DannyDayan5 wonder if market finally starting to care about inflation? i dont trust it till yields start moving materially above 4.5% but it looks like it might have legs.
@BobEUnlimited well as long as the fed redo's 2021 no one cares about this for a couple of months and the equity party can keep going hard.
The biggest problem for the hyperscalers is the fact that 50%+ of their RPO is from 2 customers, its also repaying the 10s of billions of investments they made into said customers at a high roi. hyperscalers are massively over exposed to marginal supply of tokens which is not a good business short term, long term they dont care as long as the biggest and best models require ridiculous amounts of memory because they can own the infrastructure and as such own the platform like they do for databases.
@DannyDayan5 i've never wanted more in my life for all the dumbos saying warsh is "hawkish" because of his 2009 comments to be right. if only the next fed chair wasn't just following the prevailing political narrative to get him clout.
ye idno its odd, like i understand the hate towards axios for over promising but trusting obvious saber rattling or some dumbass iranian guy who has no power in determining shit just feels really fucking weird especially when the same people say the iranian regime is terrible. its pretty easy to see whos stock's exposure is weighting their favouratism
Most Popular Users
Elon Musk 
@elonmusk
241.5M followers
Barack Obama
@barackobama
119.1M followers
Cristiano Ronaldo
@cristiano
113.6M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.2M followers
Rihanna
@rihanna
98.5M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.7M followers
KATY PERRY
@katyperry
89.5M followers
Taylor Swift
@taylorswift13
83.4M followers
Lady Gaga
@ladygaga
74.9M followers
Virat Kohli
@imvkohli
72.6M followers
Kim Kardashian
@kimkardashian
70.7M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
65.5M followers
Bill Gates
@billgates
64.8M followers
Selena Gomez
@selenagomez
62.5M followers
The Ellen Show
@theellenshow
62.3M followers
CNN
@cnn
61.8M followers
X
@x
60.8M followers