The Daily Spark

Compute demand is significantly greater than supply, so the speed of the data center buildout matters. Who are the winners and losers if the buildout takes longer?

Winners

• Owners of operating data centers. Capacity that is already online can raise rents and fill every megawatt, since new competitors cannot arrive quickly.
• Hyperscalers with capacity already secured. Those that locked in power, land and chips early can keep serving customers while rivals wait.
• Holders of powered land and grid access. Sites with secured power, interconnection and permits are the scarcest input and hardest to replicate, so their value rises.
• Suppliers of long-lead equipment. Makers of transformers, switchgear, generators and cooling gear keep long backlogs and pricing power, with lead times of up to 26 months for transformers.
• Power producers and on-site generation. Where the grid is the bottleneck, gas turbines, fuel cells and batteries become more valuable.

Losers

• Hyperscalers’ growth and earnings. Revenue is capped by supply rather than demand, and committed capex keeps accruing while capacity sits idle. Pricing power on existing capacity only partly offsets this, so earnings growth slows versus plan.
• AI labs, startups and enterprise buyers. They face rationing, higher prices and slower model training and product launches, with smaller players crowded out first.
• Developers and construction-stage investors. Cash flows start later while interest and carrying costs continue, lowering returns. Leveraged or pre-lease-dependent developers face covenant and refinancing stress.
• Equity investors in the AI trade. Valuations that assume on-time delivery are exposed to earnings revisions and multiple compression.
• The broader economy. Near-term GDP growth, which leans on AI investment, is lower, and the productivity gains from AI arrive later. Households and businesses on the same grid may also see higher electricity costs.

For chipmakers, the outcome of a delay is more mixed. Chip demand stays strong, but if sites cannot energize fast enough, chip deliveries could outrun installation and delay revenue.

The key distinction for investors is delay versus cancellation. A delay hurts returns by pushing cash flows out, but the asset remains and may become more valuable. The real risk is that scarcity unwinds, for example, if supply catches up in a rush or demand softens, which would erode the premium on existing capacity.

The bottom line for investors is that a delay would favor assets tied to scarce physical capacity, such as operating data centers, power producers and long-lead equipment makers, while assets priced on on-time delivery would underperform, notably AI-linked equities with high multiples and the debt of leveraged developers. Chipmakers face mainly timing risk, and hyperscaler bonds face mostly spread risk rather than default risk. The broader market would be exposed through the weight of AI names in major indices and through slower growth.

Silicon Valley operates on the premise that compute demand is unlimited, yet the recent slowdown in token spending growth suggests that this appetite may exist only when compute is effectively free, see first chart below.

In reality, compute appears to face a flat, highly price-sensitive demand curve, where even modest changes in token prices can drive buyers to throttle usage or substitute cheap open-source alternatives.

Silicon Valley’s conviction of infinite demand for compute collapses because it fundamentally ignores price, as productivity returns for firms on their AI investments are either arriving far slower than projected or failing to materialize at all, turning what was pitched as an essential growth engine into a margin-draining cost center.

The squeeze is showing up on the supply side too. Return on invested capital for the five largest hyperscalers is declining, as invested capital grows faster than the earnings it generates, see second chart below.

The core economic flaw in the AI bull case is that the marginal cost of producing compute exceeds the marginal benefit it delivers to users, raising questions about claims of unlimited demand for data centers. How much compute and how many data centers are needed depends on the returns firms are able to generate with that compute. So far, the AI payoff for firms has been slow to materialize, implying a growing reluctance to pay for AI services. That is consistent with the ongoing slowdown in token spending among corporate customers.

Ultimately, the number-one variable governing the viability of data center construction models is not physical capacity or power; it is the actual price users are willing to pay for the compute being built, and companies will only pay a high price for that compute if they are able to turn the compute into higher top-line growth, bigger profit margins or substantial cost savings, little of which is showing up at scale yet. Maybe that will come in the future. Maybe not.

M&A is on pace for a near-record year, driven by the ongoing rapid technological change that is forcing firms to adapt to a new world.

In the first nine months, 42 deals topped $10 billion, worth over $1.1 trillion, or 36% of all M&A value, see charts below. Big is getting bigger.

1. AI is the trigger of the M&A boom, but it is lighting up far more than tech. Buyers are chasing data centers, power generation, grid infrastructure, cooling and connectivity. Utilities and energy posted a record first half, as AI infrastructure demand and energy security accelerated. Scale-seeking in industrials and health care adds fuel.
2. Corporates lead, while private equity holds back because AI has made software harder to value. AI cuts both ways. It spurs investment in some areas but raises doubts about the durability of software business models. Sponsor tech deal value fell 67%, while non-tech PE rose 17%. Net, sponsor M&A is down 6%.
3. The M&A boom in Europe looks different from America’s.
   • US: Fewer, bigger deals. Large strategic transactions have returned while broader dealmaking stays subdued. The action is in AI, power and tech.
   • Europe: A consolidation play. In the second quarter, financial services was the highest-value sector, ahead of industrials. Banks and telecoms are bulking up as policymakers back scale, and defense is attracting capital that once steered clear.

The bottom line is that the speed of change in AI, energy, infrastructure, defense and financial services has triggered action among many corporates, and the need for long-term, flexible capital to finance this change is enormous.

For more discussion, see also here.

Written by Huw van Steenis and Torsten Slok

Germany is hosting its Invest in Germany Summit this week, against the backdrop of rising investment needs in AI, infrastructure, defense and energy resilience.

Meeting these longer-term investment needs will require public and private capital across both credit and equity markets. And Germany starts with a relatively small pool of domestic institutional capital.

Germany has the least-funded pension system of any major advanced economy. Funded pension assets are only around 7% of GDP against 149% in Sweden and 185% in Canada. Furthermore, German occupational pension schemes allocate just 14% to equities, compared with 34% across the EU and around 50% in the US, see chart below.

Yet Germany has the highest household savings rate in Europe, with much of household wealth held in deposits.

This is why pension reform matters beyond retirement policy. The reforms now under way could ultimately generate around €90 billion of new annual funded pension flows, equivalent to roughly 2% of GDP.

The bottom line is that Germany is trying to increase investment at the same time as it builds a much larger domestic pool of long-term savings. Pension reform could help connect the two.

We look at the numbers, the reforms and the implications for capital markets in our new presentation on German pension reform, available here.

Written by Huw van Steenis

Duration measures how sensitive a bond’s price is to changes in interest rates, and for the 30-year Treasury, it has fallen from 25 years in 2020 to 14 today because of higher yields and higher coupons.

That means long Treasuries offer less cushion per dollar in a downturn.

As a result, investors may need to own more fixed income to hedge against the risk of a slowdown. Pairing Treasuries with high-quality credit can help.

Investment-grade private placements and infrastructure debt are often long-dated and fixed-rate, so they can carry meaningful duration, while asset-backed finance can add income, with duration varying by structure.

The bottom line is that when interest rates stay higher for longer, lower duration weakens the hedge that long Treasuries provide.

Let’s call the ongoing correction in AI assets "The AI Reset," where investors mark down their expectations for future cash flows.

What are the implications for the economy?

A sharp reset in AI expectations could undermine corporate confidence, trigger a pullback in capex and weigh on GDP growth, just as higher oil prices, tariffs and tighter immigration restrictions push inflation higher, raising the risk of a stagflationary environment in which growth slows even as price pressures persist.

Higher interest rates have not slowed the data center buildout. Over the past year, 10-year yields are up 125 bps, yet consensus forecasts for 2027 capex growth have only been revised higher, see chart below.

The reason is that many AI investors expect AI investments to return more than 10%, well above the 4% Fed funds rate. With borrowing costs far below expected returns, the AI spending boom is not sensitive to interest rates, and tighter policy is losing its bite on the economy.

The implication for investors is that upward pressure on rates persists until either the Fed raises rates above expected AI returns, or expected AI returns fall below the Fed funds rate. With the consensus still very bullish on AI, see chart again, investors should expect continued upward pressure on rates.

A Middle East deal after the midterm elections could push yields lower via lower oil prices. But the fundamental direction of travel for rates remains higher, especially with ongoing fiscal challenges and heavy Treasury issuance adding to the term premium.

The bottom line for both public and private markets is that higher for longer continues.

A new survey by Mavvrik and Benchmarkit of 396 enterprises finds that companies are adopting AI faster than they are building the financial controls to keep its costs in check.

Only 11% of firms forecast their AI spending within 10% of actual, down from 15% last year, see chart below.

The survey also shows that those misses are changing decisions, with 62% of firms saying an unexpected AI cost altered a business decision and one in four delaying or canceling an AI initiative.

The impact is showing up in profitability too, with 81% of companies reporting moderate-to-high gross margin impact from AI costs, including 63% who see a 6 to 15 percentage point decrease.

The bottom line is that we are still waiting for the ROI from AI investments, and this survey shows that while we wait, costs continue to surprise and become harder to predict and control.

For more see here and here.

US 10-year Treasury yields have risen by about 100 bps since the beginning of the year. A model developed by Fed economists D’Amico, Kim and Wei decomposes Treasury yields to show what is driving the move. Updated estimates released point to two main factors:

1. The expected path of real short rates has moved higher. The model estimates that the average real short rate expected over the next 10 years has increased. Lower oil prices or weaker growth could ease some of this pressure and help bring longer-term rates down.

2. The term premium has risen. The term premium is the additional yield investors demand to hold a long-dated bond beyond what the expected path of Fed policy would imply. It can therefore capture compensation for longer-term risks, including concerns around fiscal sustainability or Fed credibility.

The bottom line is that higher long rates this year reflect both the higher expected policy path from the Fed and investors asking for more compensation to hold duration.

Written by Allison Boxer

The US and Europe have fundamentally different roles in the AI economy. One is building it, and the other is using it. The US dominates on every supply-side measure: hyperscaler capex approaching $800 billion in 2026, data center capacity three times that of Europe and the dominant share of global AI models and semiconductor value, see the first three charts below.

Yet on adoption, several European countries are actually ahead of the US, and many European firms are using AI in their businesses (see the fourth and fifth charts). This suggests Europe is an eager user of infrastructure it did not build and cannot easily replicate.

The bottom line is that the US is the dominant AI builder, but both the US and Europe are AI adopters. For more, see our new chart book: The macro implications of the AI investment boom: US and Europe

Written by Huw van Steenis