The AI Bubble Will Pop from Equity Supply Says This $80 Billion Asset Manager | Ben Inker | GMO

CoreWeave

Across the shows: 1 speaker in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Ben Inker · GMOBearish

  CoreWeave faces a bigger risk in an AI bust, since its business is almost entirely AI and it invests heavily.

  The damage from an AI bust depends on how much a company has invested relative to its existing business. Companies investing heavily on the back of a business that is entirely about AI, or that is much smaller than what they are trying to build, have the most at risk. CoreWeave fits that profile, so it faces a bigger issue than the large hyperscalers with established businesses. Some of these businesses could go down 100% if they cannot fund their obligations.

  “It's a bigger issue for a CoreWeave or an EBS or somebody who is investing a ton on the back of either a business which is entirely about AI or is much smaller than the business they are trying to build.”at 13:24

• Ben Inker · GMOBearish

  In an AI overbuild bust, CoreWeave would not survive while Microsoft would.

  In a scenario where data center returns prove lousy, the most single-line AI companies are most exposed. Microsoft will go down if its data center returns are poor but will be okay because it is diversified. CoreWeave, a pure-play AI cloud, will not be okay. Inker says this is not guaranteed but would not be surprised by it, given the historical pattern of technology booms.

  “Microsoft will be okay. Core Weave will not.”at 1:08:59

Oracle

Across the shows: 8 speakers in other episodes, 0 bullish and 7 bearish calls, past 30 days

• Ben Inker · GMOBearish

  Oracle's AI build-out is large relative to its core business, making it vulnerable if AI hopes disappoint.

  If AI is a bubble that bursts, the AI beneficiaries are most at risk. Hyperscalers like Meta have excellent businesses that predate ChatGPT and would survive if their AI revenue hopes fail. Oracle has a good business, but the scale of its AI ambitions is large relative to its current business, which makes it more exposed than the hyperscalers. The risk is conditional on the AI bust.

  “That may be tougher for an Oracle. Oracle has a good business, but the scale of what they are trying to do now relative to the size of their current business is big.”at 13:24

Space Exploration Technologies

Across the shows: 1 speaker in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Ben Inker · GMOBearish

  SpaceX at about 100 times sales could fall a long way even if its business stays fine.

  The impact of an AI bust depends on how expensive each company is. SpaceX, trading at about 100 times sales, has room to fall a long way even if its underlying business is fine. This applies to SpaceX as one of the AI-linked mega-IPOs, which Inker is treating as a source of equity supply.

  “If you are trading like SpaceX at whatever, a 100 times sales, there's a long way down you could go even if the business is fine.”at 13:24

Micron Technology

Across the shows: 3 speakers in other episodes, 0 bullish and 2 bearish calls, past 30 days

• Ben Inker · GMOBearish

  Now is a bad time to buy memory makers like Micron; their earnings are at a cyclical peak.

  Memory makers are commodity producers and complete price takers. Historically they look cheapest on a PE basis at the peak of their cycle, which is a bad time to buy. Their margins are so high, reportedly around 88% gross margin for Micron, that if margins fall toward historic levels, revenue falls too even if they sell many more chips. Demand growth for most commodities is slow and governed by economic growth, but it is hard to handicap when this memory shortage ends. Inker guesses the other side arrives soon enough and these stocks will not be great investments. Unlike some other AI beneficiaries, they trade at levels where current cash flow is a meaningful part of total investment cost, so if the shortage lasts three years that is great, though he does not know whether it will.

  “this would be a bad time to buy a company like the memory makers. They are commodity producers, right? They are complete price takers.”at 1:02:19

Super Micro Computer

Across the shows: 1 speaker in other episodes, 1 bullish and 0 bearish calls, past 30 days

• Ben Inker · GMOBearish

  Super Micro may not survive an AI overbuild bust, as a single-line AI supplier.

  In the AI-bust scenario where data center returns disappoint, the more single-line AI companies are at risk. Inker says Super Micro may not be okay, along with CoreWeave, while diversified giants like Microsoft would get through. He stresses this is not guaranteed but says it should not be a surprise if it happens.

  “Super Micro may not be.”at 1:09:05

US stock market outlook

• Ben Inker · GMO

  Record AI IPO and lockup supply will weigh on the market, since heavy-issuance years have historically underperformed.

  Demand for shares is price-inelastic: index funds, leveraged ETFs and underweight active managers will not sell into a rising price, so new supply needs a bigger price drop to clear. Supply is now surging. SpaceX's $75B IPO is small against a $75T market, but about $2T more becomes sellable over the next twelve months. OpenAI and Anthropic listings could add 6 to 7.5% of US market cap, more than any 12-month period on record. Historically the market does about 4 points worse for each percent of market cap issued in a year. Years with IPO issuance above 4% (1999, 2000, 2021) were all followed by difficult markets. At 6% issuance, that implies roughly 24% worse than normal. This is not a guarantee of a crash, but it is a significant headwind.

  “In general, the market does about 4% worse per percent of market cap that comes in an IPO in a given year. So, if we are seeing 6% coming in, that's 24% worse than normal.”at 11:19

• Ben Inker · GMO

  Value beats growth: the valuation spread is wide and value has outperformed since 2018.

  GMO would rather own value stocks than growth stocks because the valuation spread between them is very wide. Even if the spread stays wide, that favors value. Neither group is static: value stocks that rise can graduate into the growth universe, which is wonderful when the discount is large. Growth stocks that graduate into the value universe hurt if the buyer paid a large premium. Since 2018 the value-growth gap has been widespread, and in recent years it has worked out well for value. Growth holders also tend not to sell as prices rise, which may keep the spread wide.

  “We'd rather own value stocks than growth stocks because that spread is really wide.”at 48:35

• Ben Inker · GMO

  Largest US stocks have historically underperformed; a multi-year sideways run in mega-caps is normal.

  Over the last 150 years the largest stocks have generally underperformed. One reason is that size is defined by market cap, and one way to become the largest company is to be overvalued. Historically there have also been limits to scale, since a company must grow about 10% a year, which requires huge investment and makes being wrong catastrophic. Today's giants are less vertically integrated than past giants, so they can spread risk, but they are still making very large bets that could disappoint. If the mega-caps stop being world beaters, investors may hold less of their portfolios in US stocks and in stocks generally, and these changes sneak up on you.

  “If you look over the last one hundred and fifty years, in general, the largest stocks underperformed.”at 58:01

AI infrastructure & data centers

• Ben Inker · GMO

  Data center investment will be overbuilt, causing giant write-offs; investing in data centers now looks risky.

  Building a large data center takes several years, so builders must forecast demand years ahead for something that barely existed a few years ago. The stakes are high and it is hard to get right. Inker expects the market to peak at the top of capex before the scale of overinvestment is understood, as in the railroad and internet cycles. He expects too much investment in data centers and giant write-offs. He concedes the opposite risk, not building enough, but argues that as returns improve the build-out pulls the economy in further, so overbuilding is the likelier outcome. He sees better places to invest than AI infrastructure.

  “So, by this, I am saying, I think in the end, we will wind up investing too much in these data centers, and there will be giant write offs.”at 17:43

• Ben Inker · GMO

  If the planned AI build-out goes ahead, AI will concentrate corporate debt to an unprecedented, risky degree.

  AI-related borrowing is growing fast, and lenders increasingly need a premium to lend to AI borrowers because the sector's concentration is becoming a problem. Lending comes from banks, private credit firms and non-bank lenders that re-lend, so exposure is large. Credit is asymmetric: it earns a little if things go well and loses a lot if they go badly. Even without a wave of defaults, the concentration will create trouble. The 2001 fiber overbuild did little damage because it was not financed by leveraged entities or banks.

  “And if we get all of the investment that is slated to come, this will become a profound problem, right? Because we will see a concentration in the AI space within corporate debt that is greater than anything...in the modern history of the credit markets.”at 30:52

• Ben Inker · GMO

  Data centers and compute are commoditising; hyperscalers turned capital-heavy with unclear returns.

  Data centers are inherently an undifferentiated business because customers must be able to move between providers, so there is little ability to differentiate or take share. Speaker 1 notes that five companies are trying to commoditise compute into futures, and Inker agrees compute is being sold as a commodity. The biggest tech companies were historically capital-light with huge returns on investment. They have now made themselves much more capital-heavy and are investing in a competitive space. It is hard to see how an extraordinary return on capital will come from that, and if it does, it is hard to see who will be able to afford all the spending.

  “One really important thing about data centers is you really want interoperability because you need it to be the case that if somebody wants to move to you, they can.”at 54:56

AI bubble & tech valuations

• Ben Inker · GMO

  AI is likely a bubble, since transformational technologies historically produce overinvestment and collapsing returns.

  Transformational technologies such as canals, railroads, autos, electrification and the internet draw in huge capital because everyone sees the change coming but cannot tell who will profit. Most of the money invested earns a poor return. Early returns look good because demand ramps up, which attracts more capital until supply exceeds demand. The problem is that the whole corpus of investment can be destroyed, as in the UK railway boom, when six lines were built where one was needed and none earned money. AI fits this pattern at the first level, and its core businesses (compute and models) look competitive rather than winner-take-all. Inker concludes the AI boom is likely to prove a bubble. In the opening teaser he says that if the hyperscalers' return on investment turns out lousy, a market drop of 35 to 50% is the outcome, not the 18% figure.

  “I'd say kind of at the highest level, it is because it is a transformational technological change. Transformational technological changes normally create bubbles.”at 1:54

• Ben Inker · GMO

  Overbuilt AI spending could trigger credit defaults and a 35–50% market fall, as past tech busts did.

  Much of the spending is circular: half of compute may currently go to training AI models. If that spend turns out to be excessive, training drops to perhaps 10% of compute and inference spend falls because companies need to see returns, which are not coming in. Suddenly the world is awash in compute rather than short of it, so older data centers become uneconomic and switch off. That would bring massive credit defaults, equity companies going bust, and hyperscalers whose returns turn out lousy. The market would fall not 18% but 35% to 50%. Inker is not sure it will happen but says it would be easy. History supports the risk: railroads, autos, electrification and the internet all changed the world, and the Nasdaq fell at least 80% after the internet boom. The bust could start before anyone understands overinvestment was a problem, as it did in the internet bubble.

  “we will have massive defaults on the credit side. We will see companies on the equity side go bust... That is a minus 35% to 50% in the market.”at 1:08:08

Frontier AI labs & model race

• Ben Inker · GMO

  AI compute and models look like competitive, low-switching-cost businesses, so high returns on capital are unlikely.

  Winner-take-all businesses earn high returns on capital through network effects and returns to scale. AI compute and model provision instead look like businesses with low switching costs and comparable offerings. The host notes that users switch between models, including open-source ones, across queries and tasks, and Inker agrees. Broad adoption of AI requires it to be cheap, and something that is cheap is hard to make wonderfully profitable. Together these point to weak returns for the model and compute providers.

  “What we see is a lot of this is about either the provision of compute, or the provision of AI models. And what we see in both cases is those look more likely to be competitive businesses, than winner take all businesses.”at 1:54

• Ben Inker · GMO

  AI model delivery is highly competitive; customers face almost no lock-in and switch models freely.

  AI model delivery is pretty competitive, with each model leapfrogging the last. Inker uses a model router that picks among more than a dozen models for each task to find the most efficient option. Lock-in is essentially nonexistent, because the software moves each question to a different model rather than the user simply picking one from a drop-down. That is why the industry feels competitive, and it is also why the profits that made the US the envy of the world do not come from here.

  “it gets to pick among over a dozen models. The lock in is completely nonexistent.”at 55:55

Credit markets

• Ben Inker · GMO

  Credit spreads are unusually tight for quality junk debt, while weak CCC structures are already distressed.

  Spreads widened in September but remain near the tight end of history: high yield is at about the 13th percentile and investment grade at about the 13th percentile. BB and B spreads are very tight. CCC spreads are wider than average, around the 74th percentile, with 100 as the widest ever. Many of the riskiest capital structures were built in 2020 and 2021 when rates were zero, and they do not work with rates near 5%, so restructurings are needed. GMO's credit team is only starting to dip its toes in.

  “I am worried about credit markets full stop. Credit spreads are sort of weirdly tight these days from our perspective.”at 30:52

• Ben Inker · GMO

  Public credit shows little distress; private credit default rates are too uncertain to call.

  Public credit markets show little evidence of distress, and the global economy is doing better than Inker expected. Private markets have far less visibility. He cites a range of private credit default estimates from 1% to 19% depending on who you ask and says he does not know who is right. Public credit is the more reliable gauge.

  “we aren't seeing a lot of distress.”at 42:35

International equities

• Ben Inker · GMO

  GMO prefers non-US stocks to US stocks because non-US valuations are lower.

  GMO prefers non-US stocks to US stocks today because their prices are lower. Valuation is the crucial test when building portfolios: you do not want to own something when everyone is piling into it. GMO also overweights emerging markets, which is part of the non-US tilt.

  “we say we'd rather own non US stocks than US stocks because the prices are lower.”at 48:31

• Ben Inker · GMO

  Korea and Taiwan AI winners' profits outran prices; GMO holds less as profits look unsustainable.

  GMO is still holding Korean and Taiwanese companies that have benefited from AI, but holds less of them than before. Their valuations have not risen much while their profits have exploded. Taiwan is not full of Palantir-like names priced as if they will be the most amazing companies in history. These are companies whose profits rose tenfold while their share prices rose about sevenfold. GMO holds them because profits have grown faster than prices, but it does not believe those profits are fully sustainable, which is why it holds less than before.

  “In general, their valuations haven't gone up that much. What has happened is their profits have exploded... We hold a little less because we do not believe that those profits are fully sustainable.”at 50:52

US housing

• Ben Inker · GMO

  High interest rates have killed housing and non-AI investment, while data centers are the only unaffected area.

  With rates where they are, the housing market is dead and most non-residential construction is dead. The only part of the economy unaffected is data centers. Investment outside AI has been falling because the cost of capital is high, while AI investment is not falling. Long-dated, non-AI investments are being starved of capital, and the rest of the economy is being squeezed by the high cost of borrowing.

  “With interest rates where they are, the housing market is dead. Most non residential construction is dead.”at 17:43

US rates & the Fed

• Ben Inker · GMO

  Fed has ample room to cut rates in a recession; inflation keeps rates high for now.

  In a recession, governments and central banks lower rates, and the Fed has plenty of room to do so from where rates are now or where they will be in 2027. Right now, the economy is running hot, inflation is higher than desired, and heavy investment demand, much of it AI-driven, keeps pushing up goods and services prices and the price of money. That changes quickly in a recession, when the Fed can step in and lower rates.

  “And the federal reserve from where rates are now, or at least where rates are going to be in 2027, had plenty of room to lower the cost of capital if the economy comes to a screeching halt.”at 38:31

US inflation

• Ben Inker · GMO

  AI-driven investment demand is keeping inflation and the price of money elevated.

  The economy is running pretty hot and inflation is higher than desired. There is a lot of demand for investment, which may all come from AI, and that keeps pushing up both goods and services prices and the cost of money. This is the current danger, and it reverses quickly in a recession.

  “the economy is running pretty hot, inflation is higher than we would like, and there's a lot of demand for investment. It may all be by AI, but it's a lot of demand for investment.”at 38:31

US deficit & government debt

• Ben Inker · GMO

  The government cannot afford to backstop an AI bust given its debt and deficits, and bailing out would destabilise bonds.

  The US has a lot of debt and is running a very large deficit, and it is unclear where the money would come from to rescue AI losses. Funding that without destabilising the government bond market would be very hard, and destabilising bonds would do more damage than it helps. The 2001 fiber overbuild lost a lot of money but did not destabilise much, because the losses lived in places that were not levered or bank-financed. Losses can be absorbed as long as they sit in the right places, and the Fed can cut rates rather than the government bailing out companies.

  “I don't think the government can afford to be the backstop anymore. We have a lot of debt. We are running a very large deficit.”at 38:31

Oil & fuels

• Ben Inker · GMO

  Oil still flows via Hormuz and the Red Sea, just costlier; high prices are pushing central banks to hike.

  Oil is still getting through the Strait of Hormuz and through the Red Sea, by both the northern and southern routes. It is more expensive and more of a hassle, but it is happening, so it is unclear how much oil is actually held off the market. The world has been drawing down reserves, which cannot continue indefinitely, and nobody knows how low reserves can go before problems start. High energy prices are feeding into prices generally, including diesel, which raises the cost of moving goods and of agriculture. That is forcing central banks to raise rates more than they expected. Inker does not know when rising costs become outright shortages or how long it would take to turn into a food security crisis.

  “Oil is getting through the Strait Of Hormuz. It is getting through the Red Sea... It's more of a pain. It's more expensive, but it is happening.”at 44:44

Commodities (broad)

• Ben Inker · GMO

  Resource stocks are a long-run diversifier; GMO has liked them for a long time.

  GMO has liked energy stocks, and more broadly resource stocks, for quite a while because they diversify a portfolio. Most stocks are highly correlated with each other over daily, weekly, monthly and yearly horizons. The one group where that correlation keeps falling as the timeframe lengthens is the resource companies. They are not protected against every inflationary shock, but a chunk of them are well positioned for the bad cases that hurt everything else. On a secular basis, resources are one of the two groups Inker would overweight if he knew nothing else about the market. Valuation still matters when building portfolios, so they would not own resources when everyone is piling in.

  “We have definitely liked energy stocks, and let's say more broadly resource stocks for quite a while because of their diversifying characteristic. ...the one group of stocks where that correlation keeps dropping as the timeframe gets longer is the resource companies.”at 46:43

Earnings & corporate profits

• Ben Inker · GMO

  US profits are increasingly driven by market power, and that pattern is unlikely to reverse.

  S&P 500 profits and return on capital are at all-time highs, but the strength is concentrated. The top 10 and top 50 stocks have never made money like this, and stocks ranked 250 to 500 look about 70th percentile versus history. Mid caps sit near the 40th percentile and small caps near the 20th. At a true cyclical peak, small caps would look wonderful relative to their norm, so this does not look like a cyclical peak. Profits seem increasingly driven by market power, as the US has become a much less competitive place in most industries. Antitrust enforcement is not aggressive, so Inker does not expect the pattern to reverse. The caveat is that the tech monopolists are now investing in much more competitive areas, such as data centers and AI models.

  “they seem increasingly to be driven by market power. The US has become a much less competitive place in most industries.”at 54:03