No Relief Until 2027—Then Stocks Face Reset | Peter Berezin

NVIDIA

Across the shows: 31 speakers in other episodes, 24 bullish and 9 bearish calls, past 30 days

• Peter Berezin · BCA ResearchNeutral

  Nvidia's buyback pledges make circular financing a risk, but current chip demand is real.

  Berezin compares Nvidia's arrangements to AIG: Nvidia guarantees returns for customers by pledging to buy back chips or rent them back if demand doesn't materialize. Circular financing may have inflated some of the demand. But the underlying demand is real, since there is a shortage of compute, companies struggle to get enough chips, and GPU rental rates, including for older Blackwell chips, are rising rather than falling. So the risk is present but not an immediate problem.

  “I mean, it does to some extent, and I think you can sort of make parallels between, say, AIG, and NVIDIA in the sense that NVIDIA is, like, guaranteeing returns for all its customers by pledging to back... Buy back chips or [rent] chips if the demand doesn't materialize.”at 22:41

Micron Technology

Across the shows: 5 speakers in other episodes, 3 bullish and 1 bearish calls, past 30 days

• Peter Berezin · BCA ResearchBearish

  Micron's 80-90% margins are probably not sustainable as chip supply increases.

  Berezin says that if there is a bubble, it is in earnings rather than P/E ratios, and asks whether the margins Micron earns, of 80 to 90%, are sustainable. His answer is probably not. As more chips come onto the market from the large data-center investment, the pricing power of the companies that sell into AI demand should fade. That implies downside risk to its margins and earnings.

  “Today I would say that if there's a bubble, the bubble is more in earnings rather than in the PE ratios. So you have to kind of ask yourself like the margins that Micron has, these sort of 80, 90% margins... Probably not.”at 24:09

US consumer

• Peter Berezin · BCA Research

  US economy is K-shaped; real disposable income barely grown since Liberation Day.

  Berezin says real personal disposable income has barely grown since Liberation Day, so consumers are not seeing increases in take-home pay. Stock prices have risen, but that helps only those who own a lot of stock, and most people do not. The economy is not firing on all cylinders across the income distribution, so the lower half is struggling while the upper half still spends.

  “Well, we have a k shaped economy. Real personal disposable income has barely grown since liberation day, so consumers aren't seeing increases in their take home pay.”at 2:12

• Peter Berezin · BCA Research

  A tech-led equity bust would hit upper-income spending hard; equity wealth now far above 2000 levels.

  US households hold about $76 trillion in equity wealth, versus $13 trillion in 2000. As a share of GDP that is about 230%, up from about 130%, and more of it is now in tech than in 2000. The upper end of the income distribution has been the main driver of consumption. If tech busts, their wealth falls sharply, consumer spending would almost certainly slow, and there is no fiscal room to offset it.

  “...the bigger impact would be through the so called wealth effect. US households currently hold about 76,000,000,000,000 in equity wealth. Just for comparison purposes, in 2000 they held 13,000,000,000,000 in equity wealth...”at 18:50

US stock market outlook

• Peter Berezin · BCA Research

  More than half of stocks trade below their 200-day average; index strength rests on tech alone.

  Berezin points to weak breadth: more than half of US stocks are trading below their 200-day moving average. The index is near an all-time high only because tech stocks have been powering ahead. For many companies it has not been a good market, so the headline index overstates the health of the broader market.

  “More than half of stocks are trading below their two hundred day moving average right now. The stock market is near an all time high, but that's because tech stocks have been powering ahead.”at 3:49

• Peter Berezin · BCA Research

  Stocks could peak in 2027, ahead of AI capex peaking around 2028; neutral, not bearish, for now.

  Berezin argues AI data-center capex will keep rising for another 12 to 24 months because compute is short and GPU rental and memory prices are elevated. Stocks have historically peaked before capex peaks, so if capex peaks around 2028, stocks could peak in 2027. He is neutral on stocks today, not bearish, and suggests a benchmark equity allocation (60/40), but expects he will need to get more defensive at some point. He worries investors all assume they can exit before an AI bust, and that market timing is hard.

  “...Historically, stocks have peaked before the CapEx peaks. So if the CapEx peaks maybe in 2028, stocks could peak in 2027. So right now, I'm neutral on stocks. I'm not bearish.”at 20:50

Oil & fuels

• Peter Berezin · BCA Research

  US gasoline demand has barely fallen; the US is drawing down inventories that cannot last.

  Berezin says US gasoline volumes have declined only a little despite higher prices, so demand destruction in the US has been limited. Instead the US is relying on inventories, which are being steadily depleted. Demand destruction has been larger in countries outside the US that are less integral to the global economy, such as Bangladesh and the Philippines. He concludes the Middle East situation must improve because inventories cannot be run down indefinitely.

  “We haven't seen much of a decline in volumes of gasoline purchased in The US, despite higher prices... Right now, The US is relying on inventories, which are being steadily depleted.”at 5:58

• Peter Berezin · BCA Research

  Little scope for oil price relief this year; refined product margins and shipping costs keep prices high.

  Berezin says oil flows through the Strait of Hormuz have increased, but shipping costs have soared because oil must be transferred between tankers. Consumers buy refined products, so crack spreads matter more than crude. Those spreads are very high because Ukrainian strikes on Russian refineries have cut product supply, keeping diesel and gasoline prices high. China's export pace of refined products is another variable. He sees no huge scope for relief this year and says next year could improve.

  “...export of refined products. So there's a lot of variables at work. I don't see a huge scope for relief in oil prices this year.”at 9:39

US growth & recession risk

• Peter Berezin · BCA Research

  When the AI boom ends, US growth should fall significantly; oil alone won't drive growth.

  Berezin argues the US economy is largely being driven by AI, which makes for a dangerous setup. Until the AI boom ends, growth is likely to hold up, and oil prices would have to reach something like $200 a barrel before oil became the dominant driver. Once the AI boom ends, he expects a pretty significant decline in growth.

  “So when the AI boom ends, at that point, I think we will see a pretty significant decline in growth, but until then, unless oil prices go to like $200 a barrel...”at 7:32

• Peter Berezin · BCA Research

  Underlying US real growth is about 2%, not 5%; higher only with an AI singularity.

  Berezin disagrees with Treasury Secretary Bessent's claim of 5% GDP growth. Underlying real growth is nowhere near 5%. The labor force is growing more slowly and productivity growth has been mediocre, so underlying growth is about 2%. Only if a technological singularity with recursive self-improvement arrived might 5% be too low, and that is not the situation now.

  “Well, underlying GDP growth is nowhere near 5% in real terms. The labor force is growing more slowly... Underlying growth right now is 2%.”at 33:11

US Treasuries & yields

• Peter Berezin · BCA Research

  AI capex is the main driver of higher bond yields, more than oil.

  Oil is part of the story for bond yields because diesel prices feed transport and machinery costs and so affect core inflation more than gasoline does. But oil is not most of the story. Berezin argues the AI capex boom has created a massive demand for investment that crowds out other spending through higher interest rates. Standard theory says rates rise when investment demand rises or savings fall, and both have happened: AI investment demand has risen and stock gains have led people to save less.

  “So I think, you know, if you're looking for an explanation for why bond yields have gone up, oil is part of the story. Now, is it most of the story? No, I don't think so. I think most of the story really has to do about the fact that we're in the midst of a massive AI CapEx boom.”at 11:44

• Peter Berezin · BCA Research

  A 10-year yield near 5.5% would be high enough to turn bullish on bonds.

  Berezin has long said that a 10-year yield of about 5.5% would be high enough to make him bullish on bonds. At that level investors lock in a very attractive return, and a TIPS bond offers close to 3% real yield, which is good. The 10-year is not far off that level, so he might put on a trade to go long the 10-year in the next couple of weeks.

  “I'm not sure if five and a half is the peak, but five and a half is the level on the ten year that I've been saying that would be high enough for me to get bullish on bonds.”at 16:34

AI pricing & unit economics

• Peter Berezin · BCA Research

  Hyperscalers' high compute rental prices will fall as data-center supply grows.

  Berezin argues that current compute margins are high because supply is short. As all the data-center investment adds capacity, Microsoft, AWS and Google will lose their ability to charge high prices for access to compute. That will reduce the margins on AI infrastructure, creating a lot of risk for stocks where margins are elevated.

  “Probably not. You know, at some point, we're going to have more chips on the market. At some point, because of all this investment in data centers, the ability of Microsoft and AWS and Google to charge really high prices for access to compute, that's going to diminish as the supply of compute increases.”at 24:34

• Peter Berezin · BCA Research

  AI token prices are flattening in nominal terms; early pricing pressure, not yet severe.

  Berezin treats token prices as a direct gauge of AI demand. Token volumes keep rising, but ramp data on token prices and token spending has flattened out in nominal terms because of deflationary pressure. He says it is not severe so far. If it continues, he would need to get more concerned about the AI demand outlook.

  “The token volumes continue to rise, that'll continue. But if we're starting to see pricing pressure, we are starting to see some of that... Like it has sort of flattened out in nominal terms because of this sort of deflationary impact. Now it's not severe so far.”at 30:20

US housing

• Peter Berezin · BCA Research

  US home prices have stopped rising, and rising mortgage rates have strained housing.

  Berezin says most people's main asset is their house, and home prices are no longer rising. Housing has become increasingly strained by rising mortgage rates. This is part of why the economy is not firing on all cylinders across the income distribution, and why most households are not benefiting from the stock market gains.

  “Yes, stock prices have gone up and that's great if you own a lot of stock, but most people don't. For most people, the house is their number one asset, and home prices aren't rising anymore. If anything, the housing market has become increasingly strained by rising mortgage rates.”at 2:27

Middle East & Iran

• Peter Berezin · BCA Research

  Iran conflict is an unstable equilibrium likely to simmer for months, with oil still getting through.

  Berezin says the BCA house view is a kinetic equilibrium in which both sides keep firing but oil still gets through, which he calls very unstable. The Houthis have already expanded the conflict to Saudi infrastructure. Iran's weapons are degraded, but a few good hits can cause a lot of damage, for example in Qatar or elsewhere in the Gulf, so the risk remains. He expects the conflict to simmer for at least the next few months.

  “I think the house view is that we're in this sort of kinetic equilibrium where the two sides are still shooting at each other, but oil gets through... My guess is that this conflict continues to simmer at least for the next few months.”at 7:59

US rates & the Fed

• Peter Berezin · BCA Research

  Fed funds sits below Taylor-rule levels of 5-6%, so policy is not yet restrictive.

  Taylor rules, which derive the policy rate from the neutral rate and inflation and unemployment gaps, would put the Fed funds rate at 5 to 6%. The Fed is below that threshold. The 10-year yield is about 5.2-5.3%, but nominal GDP growth is around 6.5%, so long-end yields are not yet above nominal growth. Rates are rising and the Fed has just raised them, but the economy is not yet in restrictive territory.

  “Right now, of these Taylor rules would tell you that the Fed funds rate should be five, 6%. We're below kind of that threshold.”at 14:30

US deficit & government debt

• Peter Berezin · BCA Research

  A 7%-of-GDP deficit and above-target inflation leave little room for a muted policy response.

  In 2000 the Fed cut rates about 12 times in 2001, and fiscal policy also turned more stimulative. Berezin says those resources no longer exist. The Fed would cut if the economy slowed, but inflation has been above target for six years, and the budget deficit is 7% of GDP, so there is little room to loosen further. That means the policy response to a slowdown could be more muted than in 2000, a recipe for a weaker economy.

  “...unlike in 2000 when the Fed was able to start cutting rates... fiscal policy turned more stimulative. We don't have those resources anymore... The budget deficit is 7% GDP so how much more can it loosen?”at 19:33

AI infrastructure & data centers

• Peter Berezin · BCA Research

  AI data-center capex boom likely runs another 12 to 24 months; data centers still very profitable.

  Berezin says it is hard to time the turn from AI boom to bust, but he does not expect it soon. Data centers are still wildly profitable, compute is in short supply, and hyperscalers say so. GPU rental rates are elevated, including for older Blackwell chips, and memory prices are elevated too. All of this signals demand for data-center inputs is strong relative to supply, so he expects capex to keep growing for about another 12 to 24 months.

  “I don't think it's going to happen anytime soon simply because data centers are still wildly profitable, there's a shortage of compute out there, all the hyperscalers will tell you that.”at 20:20

Earnings & corporate profits

• Peter Berezin · BCA Research

  S&P 500 at 2019 profit margins would trade near 27x forward earnings, not 19x; margin risk is key.

  Critics say the S&P 500 at 19 times forward earnings is not that high. Berezin says to ask where the P/E would be at pre-pandemic economy-wide profit margins. At 2019 margins, today's forward P/E would be 27, not 19. If margins come down, stocks will fall even if current P/E ratios do not look crazy. He thinks margins are at risk because the AI-era margins of chipmakers and data-center operators are probably not sustainable.

  “Yeah 19 times isn't that high but ask yourself where would the P ratio be if we had the same economy wide profit margins that we had before the pandemic. If we had, if the S and P had the same profit margin that it had in 2019, today's forward PE ratio would be 27.”at 25:04

AI & jobs

• Peter Berezin · BCA Research

  AI productivity gains are not yet visible in economy-wide data; they must grow to justify capex.

  Berezin says AI productivity gains have not shown up in economy-wide data. Productivity grew 2.2% year over year in Q2, versus a 10-year average of 2%, and total factor productivity is very low. Most productivity gains have come from capital deepening. Surveys show people save time from generative AI, so gains are starting to appear, but they must become much larger to justify the scale of data-center spending.

  “So far, I would say we haven't really seen that in the economy wide data, like productivity growth in Q2 was 2.2% year over year... If you look at so called total factor productivity, it's very, very low.”at 26:29

US inflation

• Peter Berezin · BCA Research

  Oil has not yet lifted inflation expectations, but persistent above-target inflation could force a recession.

  The critical question for central banks is whether high oil prices lift inflation expectations as they did in the 1970s, leading to stagflation. So far that has not happened. But US inflation has been above target for six years. If inflation stays elevated because of oil or other factors, expectations will eventually shift and force a more forceful Fed response. Monetary policy would then tighten a lot, and a recession would follow.

  “...what's gonna be critical is whether today's high oil prices lift inflation expectations in the way that they did in like the 70s... The good news is so far it hasn't happened. The bad news is that we've had above target inflation for six years in The US.”at 28:12

AI chips & semiconductors

• Peter Berezin · BCA Research

  Memory is the most acute AI supply bottleneck; easing memory prices would signal looser AI supply chains.

  Berezin names memory prices as the first of two direct measures of AI demand. Memory is the part of the AI supply chain where the shortage is most acute. If memory prices begin to ease, that would probably mean the rest of the AI supply chain is becoming less constrained. He presents this as a signal to watch as part of his criteria for when to turn more defensive.

  “The first is memory prices. Memory is the part of the AI supply chain where the shortage is most acute. If that begins to ease, that probably means that the rest of the AI supply chain is becoming less constrained.”at 30:03

AI bubble & tech valuations

• Peter Berezin · BCA Research

  AI capex needs about $10 trillion in annual revenue; OpenAI and Anthropic make only about $130B.

  In his last report Berezin estimated AI companies need roughly $10 trillion in annual revenue to justify their data-center capex. OpenAI and Anthropic together, which he says represent about two-thirds of enterprise AI demand, generate only about $120 to 130 billion. That revenue was close to zero not long ago and still has a long way to grow. Any sign that their revenue is flattening would be quite ominous given how far they must go to monetize the spending. He argues the bubble is in earnings more than in valuation multiples.

  “Well, eventually they will because they've gone up so much. It's just a question of when. But you know, I estimated in my last report that AI companies will need to generate something in the order of $10,000,000,000,000 in annual revenue to justify all the CapEx that they're putting into data centers.”at 30:59

US jobs & labor market

• Peter Berezin · BCA Research

  Inflation is falling partly because wage growth is slowing, squeezing household income.

  Inflation is coming down, but partly because wage growth is coming down too. Average hourly earnings grew at the slowest pace in the past four years. So households are not getting the income they need to keep spending. He describes it as a better economy than expected, but one largely driven by AI.

  “Now the good news is inflation has sort of coming down, but it's coming down partly because wage growth is coming down. Average hourly earnings were up at the slowest pace in the last [past] four years.”at 33:43

Pharma & biotech

• Peter Berezin · BCA Research

  Reproductive tech set for a multi-year bull market as governments subsidise IVF.

  Fertility rates in most developed countries are below two and have fallen dramatically, so governments will get desperate to reverse them. Japan and Denmark already use IVF heavily, and more than 10% of children born in those two countries were born through IVF. Governments will subsidise these technologies very significantly, and the technologies are advancing rapidly. He expects a bull market in reproductive tech over the next three to five years.

  “...more than ten percent of kids born in those two countries were born because of IVF. Governments are going to subsidise these technologies very, very significantly... There is going to be a bull market in reproductive tech that I think we'll see play out over the next three to five years.”at 36:00

Robotics & autonomous vehicles

• Peter Berezin · BCA Research

  Robots will not replace much human labor over the next five years; governments will fund automation.

  Berezin rejects the view that a lot of robots means we do not need more people. GDP exists to improve living standards, and a world with few people has lost the plot. He also thinks a lot of human labor is still a long way from being replaceable by robots. That will come, but not over the next five years. In the meantime he expects governments to spend heavily on related efforts.

  “...I think we're still like ways away where a lot of human labor can be replaced by robots. That will come, but that's not gonna happen, I think, over the next five years.”at 37:10

Education

• Peter Berezin · BCA Research

  Falling global test scores threaten the economy, which depends on a large, skilled workforce and top talent.

  Berezin contrasts Malthus's failed prediction with today's trends. Population growth is slowing, and educational achievement is also starting to decline. He cites the recent OECD PISA results, which show a big drop in average maths, science and reading scores. He argues this is a worry for the global economy. The division of labour it relies on needs many workers, and it also needs the know-how of its most talented people. He says the world is moving backwards on these dimensions and expects governments to try to reverse the trend.

  “educational achievement is starting to decline. If you look at test scores, the OECD just released their PISA exam results a few weeks ago, big decline in average test scores in math, science, reading. That is a worry, I think, for the global economy”at 38:22

Emerging markets

• Peter Berezin · BCA Research

  Frontier markets' large populations won't drive growth because their human capital is very low.

  Berezin says fertility rates have fallen in Southeast Asia. Thailand is well below replacement, and Bangladesh's rates have also dropped. Sub-Saharan Africa's rates are still high but also declining. He says some frontier-market opportunities could exist, but the core problem is that human capital is very low. These countries have more people, but their educational attainment is poor. He concludes that this lag limits their growth opportunities.

  “the problem with a lot of these frontier markets is that the level of human capital is very, very low. ... they're really lagging behind, and that limits their growth opportunities.”at 39:43

Europe's economy

• Peter Berezin · BCA Research

  Recent immigrants to Europe have low educational attainment, so Europe is falling behind.

  Berezin argues that importing workers does not solve Europe's problem by itself. Educational attainment among recent immigrants to Europe is very low, according to his report's chart. In France, about half of second-generation immigrants are innumerate, according to the OECD. He concludes that Europe needs people who will excel at school and at work, not just more people. He says Europe has been falling behind on this dimension.

  “if you look at educational attainment amongst recent immigrants to say Europe, it's actually very low. ... Half of those kids are innumerate according to the OECD. ... Europe has sort of been falling behind on that dimension.”at 40:17