Macro & Strategy - November 2025

Read the full article (PDF)open_in_new

99 Red Balloons

“99 dreams I have had / In every one, a red balloon” – Nena

Section 1: The anatomy of a bubble

Every market cycle has its soundtrack. In 1983, Nena’s “99 Red Balloons” captured Cold War paranoia and the unintended consequences of overreaction. Today, it feels eerily relevant again— only, this time, the balloons are valuations, and the battlefield is the AI arms race.

The question haunting investors this fall is deceptively simple: Are we in a bubble? The answer, as always, is nuanced. Bubbles are defined not by price alone, but by the narrative that drives them—by the dreams we attach to red balloons floating ever higher.

What is a bubble, really?

A bubble is a psychological phenomenon as much as a financial one. It begins when prices detach from fundamentals, fuelled by a compelling story and a belief that “this time is different.” The narrative becomes self-reinforcing: Rising prices validate the story, attracting more capital, which pushes prices even higher. Eventually, reality intrudes—often abruptly.

Let’s go back to basics. Bubbles typically exhibit three traits:

1. Rapidly rising asset prices.
2. Extreme valuations.
3. Systemic risk via leverage or concentration.

But not all bubbles are created equal. Some leave behind useful infrastructure, such as railways or fibre optics. Others, like Japan’s 1980s asset bubble, leave decades of stagnation in their wake. But one thing stands out from history: Every major technological breakthrough has been marked by speculation leading to a bubble, whether involving subcomponents of the theme or the theme as a whole. Not only is it right to ask whether the AI theme has taken us into bubble territory, but it is also a vital question.

As we’ve learned, ample money can be made on the way up. The trick is to manage risks appropriately along the way and not be left holding the bag when markets come to an inevitable reassessment of valuations.

The question is: Where are we in the investment cycle driven by the AI theme? And how will we know when it’s time to scale down our investments?

Highlights

• Massive AI-driven investment is powering U.S. growth, with the Fed supporting risk assets and a macro backdrop that remains constructive.
• Tech giants continue to deliver strong profits and maintain solid balance sheets, which justifies elevated valuations despite some signs of overheating.
• Our positioning stays assertive: overweight equities, underweight the U.S. dollar, a preference for emerging markets, and we’ve taken profits on gold after its recent surge.

A museum of mania: historical bubbles and their echoes

Let’s revisit the greatest hits of financial euphoria. Each episode began with a dream: a red balloon. But the physics of euphoria is unforgiving. What goes up without earnings support eventually comes down.

Section 2: Is AI a bubble? Not yet, but the setup is forming

The AI boom is one of the largest investment cycles in modern history. Some estimates put global AI data centre spending at $600 billion this year alone. And it’s expected to grow substantially. Nvidia’s CEO is projecting $3 trillion to $4 trillion by 2030. That’s not hype; it’s hard capital. But is it sustainable?

Valuations: Elevated but not extreme

The tech sector within the S&P 500 trades at 43x earnings—rich, but well below the 68x seen in March 2000 at the peak of the dotcom bubble. The Magnificent Seven tech giants dominate market cap, but they also dominate earnings. Their balance sheets are pristine, their cash flows robust. Looking at many different metrics yields the same answer: The technology sector is less expensive than it was in 2000, and its profitability is much better.

Narrative versus monetization

The narrative is powerful: AI will transform productivity, unlock new business models, and perhaps even deliver artificial general intelligence. Adoption is growing with more and more people using AI tools in their daily life and at work.

Markets can trade on such a narrative for some time, but eventually this new technology will need to be monetized to justify current valuations and investments.

Signs of such monetization are emerging. A recent example is OpenAI, which will increasingly rely on advertising and recently signed agreements to allow online shopping on its application. Such deals are insufficient, though. Businesses will not only have to adopt and implement AI at scale but also be willing to pay for it. There is scant evidence that this is happening yet. For example, a recent paper by MIT found that although 80% of surveyed businesses explored or piloted AI tools, only 5% of businesses saw a positive P&L impact from their implementation¹.

The disconnection between narrative and monetization can’t last too long. If model improvements plateau, or enterprise adoption stalls, the narrative could unravel. But we aren’t there yet.

Capex: The fuel and the fuse

The other factor that can blow a bubble is excessive leverage. As discussed above, the capex boom is staggering. But here’s the nuance: Most of the capex is still being financed through free cash flow, not debt.

That’s a critical distinction. In past bubbles—railways, telecoms, housing—debt was the accelerant. Today, the hyperscalers, key drivers of the AI cycle and the data centre boom, are using their own cash. Their ROEs are strong. Their balance sheets are built for war.

But the story is shifting. We’re starting to read reports of a growing appetite for debt issuance as cash reserves thin. In 2025 alone, AIlinked firms have issued $141 billion in corporate credit, already surpassing 2024’s total. The asset-backed-securities market is seeing a surge in data centre deals. Vendor financing arrangements seem to echo the dot-com bubble. Private credit funds are circling.

And the capex boom itself is an increasingly important driver of economic growth. In the first two quarters of the year, AI-related capex contributed more than one percentage point to GDP growth, or about 40% of total growth. This amount is the same as personal consumption, a remarkable feat for a consumer-led economy such as the U.S.

At the same time, concentration risks are building in the financial markets. The Mag Seven represented about a third of the S&P 500 in the second quarter. But let’s not forget that the concentration still holds when looking at profits and at capex. In fact, capital expenditure by the Mag Seven is now more than 28% of the index, while profits are close to 25% of the entire S&P 500². Big tech is making a lot of money, is reinvesting it in AI capex, and is being rewarded for it by the market. As long as all three factors are aligned, and they have been moving in tandem in recent years, the case for “too much market capitalization concentration” in tech, and AI in particular, is weak at best.

This isn’t a bubble, but the U.S. economy and financial markets are betting big on the AI theme. The momentum is sustainable if AI fulfills its promises, but the macro consequences could be severe if it doesn’t.

Section 3: If it’s not a bubble, then what is it?

If we accept that AI isn’t yet a bubble, then what are we witnessing?

The answer lies in the concept of a capital supercycle, a prolonged period of elevated investment driven by a transformative technology.

Think railways in the 19th century, electricity in the 1920s, and the internet in the 1990s. These cycles often begin rationally, with investment justified by productivity gains, but they can tip into excess if expectations outrun reality.

A prisoner’s dilemma of innovation

The AI race isn’t just about growth, it’s about survival. Tech giants are locked in a prisoner’s dilemma: If they don’t invest, they risk falling behind. If they all invest, they risk overcapacity. But the potential prize—AGI, dominance in enterprise software, and control of the data layer—is too large to ignore.

This dilemma leads to circular investment logic. Nvidia invests in OpenAI to ensure that future technological breakthroughs will continue to support the AI capex cycle, and therefore that OpenAI will continue to buy its chips. Oracle builds data centres to support its cloud ambitions. Meta borrows to expand infrastructure, betting that future monetization will justify today’s spending.

Their behaviour is sensible. After all, this emerging sector of economic activity is attracting immense investments and generating equally massive profits, and it becomes not only rational but optimal for every company involved in the field to accelerate the growth of the other pegs in the machine. This approach will raise the odds of success in other parts of the supply chain as well as accelerate the coming of potentially equally massive profits.

But, at the same time, it plants the seeds of potential issues if the industry finds itself wrong on most revenue forecasts.

For example, if investment in data centres were to slow abruptly, Nvidia and Microsoft could be hit on two fronts: a decline in revenue and a loss of value in the equity investments in their partners. The upside is that this is still very different from what we called vendor financing, one of the weakest spots of the 1990s and early-2000s tech boom. Back then, a common pattern emerged: Telecom equipment suppliers would provide loans or extend credit to their customers, enabling the customers to purchase their products.

The productivity multiplier

The optimistic case is that AI will deliver a productivity renaissance. Some have estimated that AI could lift annual global GDP growth by 20%³ . Although such an outcome is unlikely, the productivity boost could be such that demand from AI would grow exponentially. If that’s the case, today’s valuations are not only justified, but they’re also conservative.

But the burden of proof is high. So far, new data centres are being used at full capacity, an indication that demand continues to outpace supply. But, with the continued build-out of data centres, what happens if the productivity boost fails to materialize?

The next phase of the cycle depends on whether AI moves from promise to profit. If it does, we’re in a capital supercycle. If it doesn’t, we’re in a prebubble setup.

Another important question is: Who exactly will extract the (potentially) very large financial benefits from AI? In the 1990s, as communication networks were being deployed, very few investors expected that advertising would be the main source of income from the web, and that a little-known firm named Google would play such a dominant role.

Will Nvidia follow Cisco's trajectory once the infrastructure boom subsides? And is OpenAI about to be dethroned by an unknown emerging player that is already building the next killer app via an open-source model, such as Deep Seek? Or will it be the data centre owners who will hold the keys to the kingdom? Or maybe the energy providers? That’s where the real risk lies: not in knowing whether this emerging industry will be lucrative, but rather where the money will be.

Strategic positioning: Diversify within the theme

For investors, the message is: Be exposed to AI but diversify within it. Chips are currently the profit centre, but that may not last. Applications, energy, infrastructure— these could be the next cycle’s winners.

International markets also offer leverage to the AI theme. Taiwan and South Korea are deeply embedded in the supply chain. Europe and Canada have less exposure but may nevertheless benefit from second-order effects.

The key is to avoid overexposure to the narrative. As history shows, the biggest names in a bubble rarely remain dominant after it bursts. Of the top 10 S&P companies in 1985, none remained in the top 10 by 2020. Only one from 2000 survived.

Our opinion is clear: Current valuations, although elevated, aren’t yet at bubble levels. But the margin for error is shrinking. So, our recommendation to investors is to focus on the signal, ignore the noise, and look for any clues that the theme is still accelerating, or decelerating. The next steps will most likely be key, as the foundational phase nears an end.

The key metrics to track include the following:

• Model improvement rates – essential for maintaining interest in AI’s benefits
• Enterprise adoption, ROI, monetization – indicative of real productivity gains
• Data centre utilization – useful to monitor overcapacity risks
• Leverage use – indicative of a shift from cash-flow to debtfinanced cycles

Closing reflection: Watch the balloons

“99 dreams I have had / In every one, a red balloon.” The lyrics capture the essence of this moment: a cycle driven by dreams, narratives, and the seductive promise of transformation.

We’re not in a bubble—yet. The AI capex supercycle should continue into 2026, supported by the promises of this new technology. But, beyond that, AI will have to deliver on its potential and justify the resources that have been poured into it. Investors should remain vigilant, diversify exposures, and watch for signs of narrative fatigue.

If the metrics hold, the cycle continues. If they falter, the balloons may pop.

Positioning

We remain overweight equities, backed by a macro backdrop that continues to favour risk assets. The U.S. Federal Reserve is easing into a cyclical upswing, while the AI-driven capex boom fuels strong growth in U.S. mega-cap tech. Markets have absorbed headline risks—from U.S.-China tensions to sector-specific shocks—and are rewarding solid earnings. Third-quarter results have exceeded expectations, reinforcing our conviction. Valuations are elevated, but fundamentals remain supportive.

We continue to underweight the U.S. dollar. Despite recent investor optimism, structural headwinds persist. Tariff threats from Washington have injected negative supply shocks just as the Fed loosens policy. Global investors are diversifying away from U.S-dollar assets, and non-dollar currencies are buoyed by more supportive policy stances. We see further downside ahead.

Within equities, we favour emerging markets. A weaker dollar supports capital flows, and EM economies are gaining traction amid policy easing, resilient demand, and attractive valuations—especially in South Korea and Brazil. Real rates remain high, leaving room for further cuts, which historically boost local equities. Positioning trends suggest more upside, despite lingering geopolitical risks.

Finally, we exited our long gold position. With spot prices nearing $4,500 an ounce, sentiment looked frothy. Even though macro drivers such as fiscal concerns and debasement risk persist, we saw growing euphoria and stepped aside. We’ll reassess if positioning normalizes, because the medium-term case for gold remains intact.

¹ MIT, The GenAI Divide – State of AI in Business 2025, July 2025, v0.1 State of AI in Business 2025 Report.pdf
² In this case, profit is measured by earnings before interest and taxes (EBI