Finans

Kommentator: Uden AI-investerings-boomet havde USA været i recession

Artiklen er bearbejdet på grundlag af kommentar fra Authers, FT

Den amerikanske økonomi ser umiddelbart robust ud, men under overfladen er væksten i stigende grad blevet båret af én ekstraordinær kraft: de enorme investeringer i kunstig intelligens. Ifølge Stijn Van Nieuwerburgh fra Columbia Business School ville USA allerede have været i recession, hvis det ikke var for AI-investeringerne.

Han vurderer, at investeringerne i AI-infrastruktur nu svarer til omkring 2,8 procent af amerikansk BNP – en investeringsbølge, der relativt set er større end jernbane-boomet i 1800-tallet. Dermed fungerer AI-kapløbet reelt som en massiv økonomisk stimulus, der både holder BNP-væksten oppe og løfter aktiviteten i blandt andet datacentre, chips, energiinfrastruktur og byggeindustrien.

Men vækstmaskinen bliver stadig dyrere at holde kørende. De fem store hyperscalere Amazon, Alphabet, Meta, Microsoft og Oracle ventes næste år at investere mere end 1.000 mia. dollar. JPMorgan vurderer samtidig, at hele 4.100 mia. dollar af de forventede 5.500 mia. dollar i samlede AI-investeringer vil blive finansieret med gæld.

Det gør udviklingen i kreditmarkederne afgørende. Kreditspændene på flere af de store AI-selskabers obligationer er begyndt at stige. Der er endnu ikke tale om krisesignaler, men investorerne kræver en højere risikopræmie for at finansiere investeringerne. Det kan blive et problem, hvis de lange renter fortsætter op, eller Federal Reserve igen strammer pengepolitikken.

Samtidig begynder de enorme investeringer at kunne ses direkte på selskabernes pengestrømme. Alphabet havde senest negativt frit cash flow for første gang siden børsnoteringen, mens Metas frie cash flow faldt 91 procent år over år i andet kvartal. Investorerne vil derfor i stigende grad kræve dokumentation for, at AI-investeringerne faktisk kan skabe et tilfredsstillende økonomisk afkast.

Det centrale spørgsmål er, om efterspørgslen kan følge med kapacitetsudbygningen. Brugen af AI vokser eksplosivt: tokenforbruget er mere end tredoblet siden januar, og virksomhedernes adoption er fortsat stigende. Men indtjeningen følger ikke nødvendigvis med. Priserne på AI-modeller falder, og brugerne søger mod billigere løsninger, hvilket betyder, at kraftigt stigende anvendelse ikke automatisk giver tilsvarende vækst i omsætningen.

Historien viser samtidig, at store teknologiske gennembrud godt kan være samfundsøkonomiske succeser, selv om de bliver dårlige investeringer for de første kapitaludbydere. Jernbaneboomet i 1800-tallet og teleinvesteringerne omkring årtusindskiftet er oplagte paralleller.

Dermed er paradokset i den amerikanske økonomi blevet markant: AI-investeringerne er muligvis det, der lige nu holder USA ude af recession – men den samme investeringsbølge er samtidig ved at skabe en betydelig finansiel risiko.

Hvis kreditmarkederne mister appetitten på at finansiere de enorme investeringer, eller hvis selskaberne ikke kan dokumentere tilstrækkelige afkast, kan den vækstmotor, der i dag holder amerikansk økonomi oppe, hurtigt blive dens største svaghed.

Uddrag fra Authers, FT:

AIndicators

Investors have awakened to the financial risks of the artificial intelligence buildout, and they have a deep well of precedent to draw on. Like all transformational technologies, it must be financed with borrowed money long before it can generate a return. More than 150 years ago, financier Jay Cooke’s ambitious campaign to fund the Northern Pacific Railroad began to unravel not because railroads were a bad idea, but because the bonds financing it traded at steep discounts for months. His firm’s collapse triggered the Panic of 1873. As Alberto Gallo of Andromeda Capital Management points out, history is full of worthy projects that weren’t worthy investments:

These projects benefited the wider population later on but ended up bringing insufficient financial rewards to their initial capital providers, especially when capital was in the form of credit.

Today’s debt-financed AI buildout isn’t necessarily destined for the same reckoning. But credit markets have a habit of registering doubt before broader markets and should be heeded. Spreads revealed mounting strain long before newspapers declared a financial panic in 1873. They are performing much the same function today. As Barclays shows, main AI players’ credit spreads have risen sharply in recent weeks. It’s not clear whether this is temporary or the start of something consequential, but July’s huge turbulence, which has nonetheless left the S&P 500 at a fresh all-time high, leaves both interpretations open:

Virtually nothing matters more to markets at present than the AI buildout. It’s such a sudden and massive stimulus for the US that it has shifted macroeconomic data. Columbia Business School’s Stijn Van Nieuwerburgh argues that without it, the US would be in recession. He estimates AI infrastructure investment at roughly 2.8% of GDP — larger than the railroad boom — and it’s projected to keep rising:

Will credit markets permit spending on the scale that AI’s backers want? Points of Return is starting a monitor of metrics of credit market confidence in AI, and in spending itself. We’ll call it AIndicators (brand names are always good) and hope this will be useful in the months ahead. Feedback will be gratefully accepted as we try to get this right.

Spreads aren’t signaling a crisis just yet. Their recent widening nevertheless suggests investors are demanding greater compensation for financing the AI buildout. State Street Investment Management’s Michael Arone argues the spending boom is reaching a mature phase, even if capital expenditures keep rising:

Credit is quietly repricing the AI boom: Spreads are wider and yields higher, not at crisis levels, but enough to show investors want to be paid for the risk that big AI spenders may not be the eventual winners.

CapEx

The concentration of debt financing is hard to ignore. The big five hyperscalers – Amazon.com Inc., Alphabet Inc., Meta Platforms Inc., Microsoft Corp. and Oracle Corp. – are expected to spend more than $1 trillion on capex next year. That figure begins to approach the Pentagon’s proposed $1.5 trillion expenditure:

An expenditure of that magnitude will inevitably move bond and corporate credit markets. Barclays strategists Dominique Toublan and Andrew Keches find that six tech groups (the five above plus SpaceX) account for 8.4% of the duration-times-spread (DTS) – which combines a bond portfolio’s credit spread with its interest-rate duration to measure overall spread risk – of the US investment-grade corporate bond market, based on data through July 31. By comparison, the six largest banks — historically far more important to credit — account for just 7.3% of the Bloomberg US Corporate Bond Index’s DTS:

Cash Flow

Hyperscalers’ free cash flows have fallen, giving them much less room for maneuver. Hence investors’ demand for proof that the model is paying off, and willingness to punish any slippage. Alphabet reported negative free cash flow in its latest quarter, for the first time since going public more than two decades ago. Meta Platforms disappointed as its second-quarter free cash flow plunged 91% year-over-year to $784 million. Others weren’t immune, although the complaint that Apple hasn’t done enough on AI does have the corollary that it still enjoys very healthy free cash flows:

Stock Markets

Meanwhile, stock markets constantly reassess who will profit most. Since ChatGPT’s 2022 debut, investors have successively rewarded companies building AI models, then those supplying the infrastructure, and those making the chips to power it. That’s now turned into unease about whether they can ever earn an adequate return.

So far this year, concerns that soaring capital spending was depriving the companies of free cash flow have combined with worries about profits to curb the Magnificent Seven stocks. Blowout second-quarter earnings, particularly for Microsoft, which had lagged badly, reignited some enthusiasm:

Winners and Losers

The spending weighing on cash flows will pay for itself if demand for computing power increases as the hyperscalers hope. But past industrial revolutions suggest linear extrapolation won’t necessarily happen. Drawing parallels to telecoms in the late 1990s, Man Group’s Matt Rowe argues some degree of overinvestment in AI infrastructure is inevitable. What matters is how it’s financed and absorbed by credit markets:

You’re going to see some M&A, partnerships, some failure, and ultimately more consolidation and prices coming down… I think we’re partway through that. We haven’t seen any marquee failures, but seeing credit spreads widen could be the next phase of the cycle.

If the evidence investors seek doesn’t arrive or proves unconvincing, they’ll probably do what they’ve done throughout the AI cycle: pivot toward areas with stronger near-term fundamentals (as in the recent boom in chipmakers), rather than abandon AI altogether. This could favor “pick-and-shovel” firms involved in data center construction, memory chips, or providing power infrastructure:

Token Usage

The bottom line is whether people use the product once it’s ready, and what they pay for it. In Field of Dreams terms, the industry is building it; will they come? A Federal Reserve study published in April estimates that AI adoption among US businesses reached 18% by the end of 2025, pointing to substantial room for further expansion. Demand tells a similar story. Data from OpenRouter, which tracks consumption of tokens (the basic units of text processed by large language models), show usage has more than tripled since January:

Source: OpenRouter

Does soaring token usage necessarily translate into soaring revenues for providers of foundational models? Not quite. The LLM Token Expenditure Index, which measures the effective expenditure on large language models by combining token prices and usage, reveals the disconnect. As token volumes surge, falling prices and migration toward cheaper models suggest monetization has not kept pace with adoption:

What Next?

In the near term, the greatest risks come from rising long-term Treasury yields, and the possibility of tighter monetary policy by the Fed. This could make life harder for those who borrowed most heavily.

JPMorgan estimates that $4.1 trillion of the $5.5 trillion in AI capex will be debt-financed, so rate hikes would essentially pile on the costs of this buildout, putting a spoke in the wheels of the massive investment cycle.

Historically, hyperscalers have been very profitable and had no difficulty funding most investment from their own balance sheets using cash flows, equity and traditional debt markets. The AI buildout has forced them to look beyond conventional financing.

Columbia Business School’s Van Nieuwerburgh argues that while these financing structures have protected the mega caps’ balance sheets, they may be shifting risks elsewhere. Their opacity, he says, makes it hard to assess where leverage sits — and whether a reverse for AI could trigger losses across the broader economy:

We were making the exact same argument in the financial crisis when we talked about mortgage-backed securities… The argument was instead of having all these mortgages sit on bank balance sheets, why don’t we securitize them and spread them to other parts of the financial system that have less leverage. What we didn’t appreciate at the time is that a lot of that risk transfer actually didn’t take place and the banks kept, often through the back door, exposures to these structures.

We will update the AIndicators in due course.

Få dagens vigtigste
økonominyheder hver dag kl. 12

Bliv opdateret på aktiemarkedets bevægelser, skarpe indsigter
og nyeste tendenser fra Økonomisk Ugebrev – helt gratis.

Jeg giver samtykke til, at I sender mig mails med de seneste historier fra Økonomisk Ugebrev. Lejlighedsvis må I gerne sende mig gode tilbud og information om events. Samtidig accepterer jeg ØU’s Privatlivspolitik. Du kan til enhver tid afmelde dig med et enkelt klik.

[postviewcount]

Jobannoncer

Økonomichef til Børne-og Ungdomsforvaltningen i Københavns Kommune

Region Hovedstaden

Annonce

Hent udgivelser

Nyt magasin Danmarks 100 Rigeste: Se danmarksrigeste.dkNYT

Hent udgivelser

Morten W. Langer

Velkommen til Økonomisk Ugebrev

Vores erfarne journalister stiller hver dag skarpt på ...

Gratis white papers:

Kommentator: Uden AI-investerings-boomet havde USA været i recession

Morten W. Langer

AIndicators

What Next?

Få dagens vigtigste
økonominyheder hver dag kl. 12

Relaterede nyheder

FOM Technologies offentliggør halvårsrapport for første halvår 2026

Space-X fortsætter op med raketfart – plus 9 procent i dag

Pandora leverer 3% organisk vækst i andet kvartal – forventninger opgraderet

Jobannoncer

Mere fra ØU Finans

Dybdegående og original
journalistik siden 1994

OM ØU

Sektioner

KONTAKT

Log ind

FÅ VORES STORE NYTÅRSUDGAVE AF FORMUE

Her er de 10 bedste aktier i 2022

Analyser med aktiestrategi, enkeltaktier, consensus, crypto og råvarer

Analyse af og prognoser for Fixed Income (statsrenter og realkreditrenter)

Analyse og prognoser for kort rente, samt for centralbankernes politikker

Nyt magasin Danmarks 100 Rigeste: Se danmarksrigeste.dkNYT

Hent udgivelser

Morten W. Langer

Velkommen til Økonomisk Ugebrev

Vores erfarne journalister stiller hver dag skarpt på ...

Gratis white papers:

Kommentator: Uden AI-investerings-boomet havde USA været i recession

Morten W. Langer

AIndicators

What Next?

Få dagens vigtigste økonominyheder hver dag kl. 12

Relaterede nyheder

FOM Technologies offentliggør halvårsrapport for første halvår 2026

Space-X fortsætter op med raketfart – plus 9 procent i dag

Pandora leverer 3% organisk vækst i andet kvartal – forventninger opgraderet

Jobannoncer

Mere fra ØU Finans

Dybdegående og original journalistik siden 1994

OM ØU

Sektioner

KONTAKT

FÅ VORES STORE NYTÅRSUDGAVE AF FORMUE

Her er de 10 bedste aktier i 2022

Analyser med aktiestrategi, enkeltaktier, consensus, crypto og råvarer

Analyse af og prognoser for Fixed Income (statsrenter og realkreditrenter)

Analyse og prognoser for kort rente, samt for centralbankernes politikker

Få dagens vigtigste
økonominyheder hver dag kl. 12

Dybdegående og original
journalistik siden 1994