UGO13: How Much Longer Can AI Carry the Market ft. Christian Mueller-Glissmann

Christian Mueller-Glissmann · Goldman Sachs

Rising yields are starting to weigh on parts of equities as earnings tailwinds moderate.

Earnings growth was enormous in the first half of the year, but it is now moderating. Rates headwinds are starting to bite. Combined with the speed of the bond sell-off, this has made equities consolidate, especially in certain parts of the market. The rates impact falls unevenly, hitting some equity segments harder than others.

“...the rates headwinds are starting to bite a bit and they're starting to weigh on equities a bit, especially certain parts of equities”at 2:13

Cem Karsan · Kai Volatility Advisors

Markets have become the economy's biggest driver, creating a fragile, reflexive liquidity loop.

Roughly $50 trillion of equity collateral, public and private, is priced off the S&P 500. In April and May, a 20% market rally added about $50 trillion of collateral in two months, which is far larger than the roughly $10 trillion of COVID-era Fed QE. Capex is funded by market gains, which feed earnings and reflect back into the market, creating a reflexive loop. The loop depends on how reliable market liquidity is. Equities are now too big to fail, so the economy becomes fragile and concentrated in the assets getting appreciation.

“Markets themselves are arguably the biggest driver of the economy, period.”at 22:20

Christian Mueller-Glissmann · Goldman Sachs

US household equity holdings are at a record share of wealth, creating a negative feedback loop if stocks fall sharply.

Equity holdings as a share of US household wealth are at their highest level ever, above the tech bubble and the COVID recovery. The gap between equity and residential real estate as a share of net wealth is the widest on record. In long equity bull markets, household assets drift toward equities, and the wealth effect grows and supports spending. The more extreme this exposure gets, the larger the negative feedback loop to households and the economy after a sharp equity correction, through financial conditions tightening.

“If you look at the value of equity holdings for US households as a proportion of their wealth, you're at the highest level ever.”at 24:33

Christian Mueller-Glissmann · Goldman Sachs

Nearly 40% of S&P 500 stocks show negative beta, a record that echoes the tech bubble.

Over the last three months nearly 40% of S&P 500 stocks have had negative beta to the index, and roughly 20-25% over the last year. He says this has never happened and was last seen in the tech bubble. The stocks are rate-sensitive companies that have suffered from higher rates without earnings tailwinds: bond proxies, low-volatility names and high-dividend payers. Earnings dispersion is also among the largest on record, which he sees as a fundamental driver alongside index-level selling. He calls it a chicken-and-egg problem, with dispersion carry trades attracting index selling and vice versa.

“nearly 40% ... within the S and P 500, 40 of stocks had a negative beta to the S and P 500. ... That has never happened.”at 47:55

Christian Mueller-Glissmann · Goldman Sachs

Defensive, bond-proxy equities should diversify well if AI goes wrong, as in the 2000s tech bust.

Low-volatility, high-dividend and rate-sensitive bond-proxy stocks currently move negatively with the index, which he links to AI creating winners and losers. He thinks they will be a fantastic diversifier later if something goes wrong with AI. He points to the tech bubble, when the Nasdaq fell about 80% and the S&P 500 about 46%, while the S&P Dividend Aristocrats and low-vol indices rose about 10%, and those defensive areas eventually snapped back. Market timing and diversification are the two value-adds an asset allocator has, so factor diversification is a useful tool.

“they're gonna be a fantastic diversifier later down the road if something goes wrong with AI”at 48:55

Cem Karsan · Kai Volatility Advisors

Structurally stable businesses with solid earnings look cheap on DCF and can be defensive.

Certain businesses look very cheap on discounted cash flow, particularly relative to their earnings growth. Many have structurally stable, solid earnings that are not built on capex or on the market rising. They will still face some blowback if the economic cycle turns, but they can be much more resilient and can serve as a defensive asset when money needs to go somewhere.

“certain businesses are really cheap right now. ... there's a lot of structurally stable that are real solid earnings.”at 56:33

Cem Karsan · Kai Volatility Advisors

Equities could go nowhere in nominal terms, as in 1968–1982, as margins and multiples pull in opposite directions.

Cem draws on a study of the 1960s and 1970s. Rising discount rates compress multiples while profit-margin pressure cuts earnings, and these forces pull equities in opposite directions. He cites 1968–1982, when equity markets went nowhere in nominal terms, and presents that as a parallel to today's environment, which he calls a dramatic departure from recent history. His view is a historical analogy, not a firm forecast, and he does not give a number.

“Right? Equities are nominal... those things end up pulling in two directions for equities. And as we saw sixty eight to eighty two, equity markets went nowhere. Literally nowhere in nominal, terms.”at 1:07:33

Christian Mueller-Glissmann · Goldman Sachs

AI capex boom is historically large and cannot be extrapolated; capex is peaking, fading the earnings tailwind.

The AI capex investment phase may be the largest and longest in history as a share of GDP growth, so its earnings contribution cannot be extrapolated indefinitely. The market is already questioning how much of the resulting earnings level and growth to extrapolate. AI capex is peaking, so its second derivative, which is what drives earnings, becomes less favourable. The tailwind from capex to earnings is therefore fading.

“the problem is you cannot extrapolate the investment phase for too long. Like, this kind of investment phase might be both the largest and longest we've seen in history with regards to how much it is as a proportion of GDP growth.”at 6:45

Christian Mueller-Glissmann · Goldman Sachs

Hyperscalers' negative free cash flow needs AI monetization shown in Q3 earnings, or profitability falls.

Hyperscalers' free cash flow is negative because of AI capex. Without that capex they would be highly profitable, and much of that profitability is not cyclical. Over time the capex must be depreciated, but depreciation is spread over several years and they can earn revenue from the investments. High-frequency indicators such as cloud computing show AI revenue starting to come through. Q3 earnings should show how they monetise the capex. If monetisation does not come through, negative free cash flow yields and future depreciation headwinds will need to be reflected in lower valuations, because profitability will fall.

“if you look at some very high frequency indication of AI monetization, like cloud computing ... they're actually starting to see revenue come in. And this will, I think, be a key focus for the q three earnings season.”at 10:39

Cem Karsan · Kai Volatility Advisors

Debt-funded hyperscaler capex crowds out the long end and is unsustainable as a base case.

Earnings growth is increasingly financed by hyperscaler debt, and CDS are breaking out. Borrowing to sustain earnings growth crowds out the long end of the curve, creating a self-reinforcing loop. It can last for quarters but probably not years. The way out is for non-capex earnings to boom as AI is adopted. That is possible, but Karsan says it is hard to make the base case given current market pricing.

“It's a bit of a circle, isn't it? ... to continue the earnings growth, we... They need to borrow more, which is gonna crowd out the long end of the curve more. This is a death loop, right, at some point.”at 38:59

Cem Karsan · Kai Volatility Advisors

AI capex is increasingly government-funded, with direct US investment in equities as the next step.

Karsan argues the capex cycle is moving from market-funded to state-backed. He points to a September 2025 table where the government agreed to large capex commitments, and to plans for a sovereign wealth fund at scale and Trump accounts. Incentives leave few options other than printing money and buying stocks, which he compares to Japan's fifteen years of similar policy. He thinks this direct government investment is the next logical step in the AI capex bubble.

“How about you? I'll do 500,000,000,000 in CapEx. Right? It's going to be direct investment coming into equities from the US government.”at 1:05:01

Cem Karsan · Kai Volatility Advisors

Recent AI-driven earnings growth is asset appreciation and capex sugar rush, not operational growth.

Much of the earnings growth reflects asset appreciation: private AI valuations, such as Anthropic going from about $380B to $2T, flow into reported earnings as marks. Earnings are also boosted by liquidity-driven capex, which feeds back into earnings. Hyperscaler cash flows are falling sharply while they are the main contributors to earnings growth. Normally earnings and cash flow are highly correlated, but they have diverged. This resembles 1999, when capex-driven earnings growth of 27% was followed by a bust.

“And it really isn't operational earnings growth. Actually, cash flows from these hyperscalers is falling out of bed.”at 9:37

Christian Mueller-Glissmann · Goldman Sachs

Not a valuation bubble; more likely an earnings bubble in parts of the market, which deflates slowly.

Investors already discount the earnings: MSCI World 12-month forward P/E is down about 15% year to date, and the S&P 500 forward P/E fell from 23x to 19x. Semiconductor and tech valuations have de-rated. The earnings driving the boom come from cyclical businesses, so they can fall as well as rise. Data center construction and compute contracts are long-dated, giving unusual visibility. Whether compute demand and pricing hold up is the open question. The Shiller P/E is high because it uses 10-year trailing earnings, which lag a sharp rise in profits. S&P 500 return on equity is about 24%, versus below 20% five years ago and before the AI capex boom. Hyperscalers already earned ROE in the 40s. Late-1990s ROE was less than half today's, so the comparison to the tech bubble is misleading. A valuation bubble can burst quickly, whereas an earnings bubble takes a long time to deflate.

“We would say we're certainly not in a valuation bubble. We are more likely to be a bit in an earnings bubble, at least in some of these parts of the market.”at 10:39

Cem Karsan · Kai Volatility Advisors

AI-driven market extremes are a bubble that depends on current trends holding and can reverse sharply.

Vol compression forces momentum, concentrates gains in part of the market and drives more capex, creating a self-reinforcing loop that produces extreme outcomes. These outcomes are the basis for calling it a bubble. The market depends on everything maintaining this direction, but the same forces work just as strongly in reverse, and a small idiosyncratic shock could start a sharp unwind. Bubbles can persist for years, as the 1998 tech bubble and the 2005-06 housing bubble showed, but that does not make the market sustainable. The move is overdone and unsustainable even if it can continue.

“And those extreme outcomes is why I call it a bubble, really. It's at the end of the day, it is dependent on everything maintaining this direction.”at 51:28

Christian Mueller-Glissmann · Goldman Sachs

AI/tech is a boom rather than a bubble; the peak is very hard to time.

He says he would not call the AI-driven market a bubble, but a boom that will end at some point. Tech is harvesting roughly 30% annualised earnings growth, so a flat share price implies about a 30% de-rating over a year. Strong earnings growth makes it hard to generate large downside volatility unless earnings growth falls off, which is not his base case. He argues that timing the peak of a bull market is extremely difficult. Calling the late-1990s bubble in 1998 would have been about two years early and costly for investors. He therefore recommends preparing portfolios with risk management and shock-absorbing tools rather than betting on a top.

“I think we all know that we're in a boom, and a boom at some point comes to an end. I wouldn't call it a bubble. As I said, let's say a boom for now.”at 52:45

Cem Karsan · Kai Volatility Advisors

Near-zero 2021 debt rolling over at much higher rates, plus new supply, will pressure the bond market.

Everybody financed at near-zero rates in 2021, and the average duration of debt is five to eight years. Higher rates therefore only fully bite as the world refinances. Debt taken at 1–2% to fund investments earning 5% now has to be refinanced at about 7.5%, forcing liquidation or re-financing. Dramatic new debt issuance is competing for financing at the same time, creating a supply-demand imbalance. Policy measures such as stablecoin legislation, SLR changes and Treasury buybacks have not kept it in check.

“everybody financed at zero percent five years ago in 2021, and the average duration of debt is five to eight years... It doesn't matter fully until the world has to refinance and roll over that debt.”at 29:28

Christian Mueller-Glissmann · Goldman Sachs

The structural cycle is now unfriendly to long-duration bonds; the 10-year yield sits above its long-run average.

Innovation is a fight between productivity, which raises real yields over the long term, and inflation. Near term, AI innovation crowds out bond demand as investors move into equities. Inflation volatility is higher and hurts bonds. Central banks cannot ease as they did with negative rates and QE. Fiscal concerns and high debt-to-GDP add pressure. The US 10-year is at about 5.3–5.4%, versus a 300-year average of about 4.7% (about 4.5% excluding the 1970s). The open question is whether the current term premium adequately compensates for these risks.

“And the structural cycle has just become much less friendly for bonds.”at 32:14

Christian Mueller-Glissmann · Goldman Sachs

Yields above 5% are damaging rate-sensitive equities and have passed the historical indigestion threshold.

Yields have moved past about 5%, which he calls a moving target because innovation and capital competition have let equities absorb that level so far. The damage is visible below the index: small and mid caps, real estate, homebuilders, rate-sensitive parts and selective financials are under pressure. The speed test also flags stress: the three-month rise in yields exceeds two standard deviations, which historically marks the point where equities start to show indigestion. The source matters too. Yield rises driven by policy and front-end real yields are unfriendly to equities, while growth-driven rises are easier to absorb, and recent growth news, especially consumer data, has been softening.

“there is damage, like the higher bond yields are causing damage. So I think level wise, we are definitely getting closer”at 36:40

Christian Mueller-Glissmann · Goldman Sachs

Bonds are becoming a less reliable equity hedge, so the belly may work better than long duration.

In a world of anchored low inflation, deflation risk and fiscal consolidation, bonds were a good buffer for portfolios. Now there are fiscal concerns in most countries and policy and political volatility in many more. Combined with the rate and yield issues discussed earlier, this makes the bond market less reliable as an equity partner. The 60/40 portfolio will not return to its pre-COVID form. Duration choice depends on the country, and maybe the belly of the curve is more useful than long-duration bonds.

“I think suddenly the bond market, on top of the reasons we discussed earlier, is becoming a less reliable partner in for equities in the portfolio.”at 1:01:12

Christian Mueller-Glissmann · Goldman Sachs

Equity volatility looks cheap versus credit, rates and commodity vol; favours longer-dated upside calls into year-end.

Benchmarked against credit spreads, rates volatility and commodity volatility, equity vol looks cheap. He cites several sources of convexity: earnings, AI applications, and reactions to bonds, which can move equities in both directions. Dispersion is a major reason vol is low, since tech is harvesting around 30% annualised earnings growth and there is huge divergence under the surface. Single-stock vols on the hyperscalers have reset lower since the summer momentum unwind. He leans toward upside convexity being cheap into year-end and sees value in slightly longer-dated calls, including in tech. His base case is not a sharp downside move, though the earnings season could be a catalyst for one. He sees a growing wall of worry (midterms, Middle East war, bond sell-off, AI earnings) that could either be climbed or end in capitulation.

“No doubt. I think you highlighted it. If you benchmark it to credit spreads, to rates volatility, to commodity volatility, I think it does look a bit cheap.”at 42:26

Cem Karsan · Kai Volatility Advisors

Structured-product selling compresses index vol, pinning the index and forcing record dispersion since 2017.

Index volatility has been compressed by the growth in structured products and the systematic selling that goes with them. That pins the index relative to its constituents and forces dispersion. Realised index dispersion has stayed in a tight band for about 120 years, including through WWII and other macro shocks. Since 2017 it has risen dramatically, and that increase almost one-to-one tracks index-level vol selling. Momentum and passive flows then amplify the effect, concentrating gains in a few names.

“Index volatility compression, which has been happening because of increases in structured products ... is pinning the index relative to the constituents, which is forcing that dispersion.”at 45:57

Christian Mueller-Glissmann · Goldman Sachs

FX volatility and FX-linked positions will become much more valuable for protecting non-home-currency assets.

Christian argues that as portfolios face more macro and geopolitical risk, FX volatility and instruments tied to it will gain value as a way to protect assets held outside the investor's home currency. He offers no numbers in this segment, so the case rests on the general argument about diversification and the changing macro backdrop discussed earlier in the episode.

“FX volatility and expressions linked to FX volatility will also become much more valuable in the portfolio to protect assets outside your home currency.”at 1:06:38

Christian Mueller-Glissmann · Goldman Sachs

Global asset owners are increasingly home-biased because FX risk from policy shifts is rising.

In discussions with global asset allocators, his team has seen more home bias driven by FX risk. Large policy shifts make currency risk matter more. Asset owners with liabilities or a need to manage returns in a specific currency migrate closer to home. This has already happened in bond markets and is increasingly happening in equities. The cost is that the best opportunities are often in other countries, so investors need to be nimble to capture them.

“there's a bit more home bias because of FX risk.”at 59:42

Christian Mueller-Glissmann · Goldman Sachs

FX volatility looks remarkably cheap and will become more valuable for hedging foreign assets.

FX used to be boring and mean-reverting, with no violent convex moves before COVID. That has changed in recent years, with sharp moves in the euro, the Swiss franc and earlier the yen, including in the last few days around Europe. Given how cheap FX vol looks relative to that new behaviour, he expects FX volatility and related expressions to become much more valuable for protecting assets held outside the home currency.

“FX volatility, we talked a lot about equity volatility, to me also looks remarkably cheap”at 1:06:05

Christian Mueller-Glissmann · Goldman Sachs

Gold is a useful geopolitical and trust-crisis diversifier, but day-to-day timing is too hard.

Gold has convexity when trust breaks down. It was supported in the 1970s when investors doubted central banks on inflation, during the financial crisis when they doubted the financial system, and since COVID when trust in governments fell. His team has been quite constructive on gold as a portfolio diversifier against broad geopolitical risk. But he says trading the day-to-day moves tied to policy and politics is too hard, so they stay away from that and treat gold as a structural holding.

“And I feel like that has supported gold, and we've been quite constructive on gold as a portfolio diversifier to deal with broad geopolitical risk.”at 1:01:58

Cem Karsan · Kai Volatility Advisors

In the 1960s–70s, gold, FX and bond volatility outperformed while industrial commodities and equity vol lagged.

Cem says his firm's study of the 1960s and 1970s found that the volatility of FX, bonds and gold, which he treats as essentially currencies, far outperformed any other period in history. Industrial commodities and equity volatility underperformed over the same stretch. He sees similarities with today, which suggests gold and currency-like assets could again behave differently from industrial commodities and equities. The view rests on one historical study and he gives no current forecast.

“During that period, the volatility of FX and bonds and gold, which is essentially currency, dramatically outperformed any other period in history. Meanwhile, counterintuitively, industrial commodities and equity vol dramatically underperformed.”at 1:06:58

Christian Mueller-Glissmann · Goldman Sachs

The Middle East war, via energy prices, is a bigger inflation driver this cycle than AI capex.

Some inflation has come from the AI capex boom, but the bigger driver has been the Middle East war. The war pushed up energy prices, which boosted inflation. Bond yields have had to adjust to this higher-inflation regime. The energy shock, rather than AI investment, is the main source of the inflation pressure on the bond market.

“I think the bigger [driver of] inflation has been the Middle East war. So the Middle East war pushing up energy prices has then boosted inflation.”at 2:13

Christian Mueller-Glissmann · Goldman Sachs

AI productivity gains are still early on the J-curve, not yet visible in earnings or macro data.

Every technology revolution moves through phases: first investment, then application and adoption. Productivity gains lag and follow a J-curve. The AI boom is still in the investment phase. Productivity benefits are not yet visible in corporate earnings or in macro productivity measures. Early signs include some CEOs quantifying cost savings. Christian's macro team still sees material productivity benefits over the next decade or so, but near-term gains are hard to see.

“I would say that we are probably still at the early part of the j curve.”at 6:45

Cem Karsan · Kai Volatility Advisors

Earnings are about 50% above trend; history implies mean reversion well below average over 2-5 years.

Real earnings growth has been about 6% over the long run, and earnings are now roughly 50% above that structural trend on a rolling basis. In past bubbles, earnings did not simply plateau at a new high. They mean-reverted sharply, and the reversion went well beyond the long-run mean for several years. Earnings boosts from lower tax rates could reverse under populism, and capex-driven earnings are not structurally higher.

“We see actually mean reversion in earnings dramatically... whenever we see that history, the mean reversion is not just to the mean. It's well beyond the means, the other side, for the next several two, three, four, five years.”at 18:00

Christian Mueller-Glissmann · Goldman Sachs

Because of wealth effects, a sharp equity correction would likely bring a policy response as a buffer.

Given how large the wealth effect now is and the linkage between equities and household spending, a sharp equity correction would create a negative feedback loop. The Fed would probably react, which would provide some buffer from policy. Once policymakers set things in motion, however, stabilising them is not always easy.

“considering the importance of these wealth effects and considering exactly that linkage you were highlighting, there would probably be a bit of a buffer from policy.”at 24:33

Christian Mueller-Glissmann · Goldman Sachs

Corporate AI spend is only about 0.1% of COGS, so monetisation gap risks narrow AI earnings.

Citing the US strategy team's work, S&P 500 companies spend only about 0.1% of cost of goods sold on AI, via subscriptions and token use. He thinks that figure is too low and that companies are not paying for much of their usage. That leaves a large potential spend that has to be bridged into real monetisation. If it is not, AI earnings stay narrow, with the market unwilling to pay for earnings that carry cyclicality and rising credit risk across the AI ecosystem. He wants spenders to show revenue plans and adopters to show productivity use cases, and sees the upcoming earnings season as key to this.

“So there's a large amount of potential spend from the corporates, and they're already using it. It's probably that they are not paying for it fully.”at 40:56

Cem Karsan · Kai Volatility Advisors

CDS and bond volatility are breaking out while the VIX stays complacent, a warning sign.

CDS are starting to break out for the first time in a long time. The MOVE index, which tracks bond volatility, is also starting to break out. The VIX, however, shows nothing unusual. The divergence between credit and bond-vol stress and equity volatility is a signal that equity markets are not pricing in.

“I'll highlight that CDS are starting to break out as we've seen. ... But the VIX continues to just, you know, nothing to see here.”at 41:35

Christian Mueller-Glissmann · Goldman Sachs

Real assets can cushion medium-term inflation volatility, though real estate and infrastructure often do not protect.

Asset-owner discussions have concluded that real assets deserve a larger place in portfolios. Policy choices that cause market volatility often generate inflation. Real assets deliver stable cash-flow growth in real terms. They do not always protect against inflation, and real estate and infrastructure in particular do not, but commodities may do a bit better. Over the medium term, real assets have the potential to cushion inflation volatility.

“One of the conclusions we've had with a lot of asset owners discussions on is real assets, the value of real assets in the portfolio”at 59:02

Cem Karsan · Kai Volatility Advisors

US debt cannot be resolved without debasement, arriving within the next year as a convex market shock.

The US cannot deal with its debt while competing with China and driving growth and investment without some degree of debasement. That relies on the exorbitant privilege of the dollar. Debasement, which people have discussed for years, is now close. When it begins it will be a dramatic convex outcome for global markets, not necessarily a fall in prices. The likely policy response is massive QE with a cap on the long end of the curve, which creates a structural inflationary loop.

“that debasement that people have been talking about for some time is not... Is no longer two, five, ten years away. ... It's coming in the next year,”at 1:03:13