Investing/Nov 21 edition/Podcast
StocksBonds & creditFed & ratesMacro

StocksEquities, earnings & valuations

Andy Constan says three pressures from the AI buildout are weighing on stocks: share and bond issuance, a profit gap and 5% Treasury yields

Damped Spring's Andy Constan argues the AI buildout has turned from a market tailwind into a drag. Companies are moving from buybacks to issuing stock and debt, profit hopes outrun what flattening deficits can fund, and 10-year yields stay above 5%.

From the Excess Returns podcast episode of November 20, 2026 and cited sources: Constan says companies that once spent about 2% of GDP on net share buybacks are now net issuers of stock and bonds to pay for chips and data centres. [2] [9] [7]

SOURCES[2] Full Transcript: Andy Constan on AI and Shifting Market… excessreturnspod.substack.com[9] Full Transcript: Andy Constan on IPOs, AI CapEx, and the… excessreturnspod.substack.com[7] Yahoo Finance: Banks, investors grow wary of AI debt finance.yahoo.com

AI-assisted, not human-edited. Methodology Corrections Who owns this site

KEY FACTS
What happened
Constan says companies that once spent about 2% of GDP on net share buybacks are now net issuers of stock and bonds to pay for chips and data centres. [2] [9] [7]
Original source
Excess Returns podcast episode, independent perspective
Brief by
AT Pulse research desk AI-assisted; publisher and contact on the desk page
When
November 20, 2026 podcast episode published
Evidence
11 sources across 9 sites; no transcript

Reach: not measured yet

THE CORE IDEAS4 TAKEAWAYS
01

Constan says companies that once spent about 2% of GDP on net share buybacks are now net issuers of stock and bonds to pay for chips and data centres. That adds supply that becomes a drag if AI returns only meet expectations. The 2% figure is his own estimate. The bond totals are independently reported: hyperscalers have sold about $370.7B of bonds this year, roughly a fifth of record US investment-grade issuance. [2] [9] [7]

02

His second pressure is a profit gap. Using the Kalecki-Levy profits framework, which counts government deficits as a source of corporate profits, he argues that deficits that have stopped growing cannot support current earnings expectations. Projections of negative combined free cash flow for big tech in 2026 and 2027 sharpen that concern. Strong cloud growth at Azure, AWS and Google Cloud cuts the other way. [2] [4]

03

The third pressure is high borrowing costs that last. The 10-year Treasury yield passed 5% in early November, its highest in 24 years. Spreads on hyperscaler bonds have also widened about 50bp this year as banks and credit investors look harder at AI debt deals. [2] [6] [7]

04

Constan is optimistic about the technology itself. His deciding test is commercial: whether customers can use AI profitably at prices that also reward the companies selling it. On Ramp, the top 1% of businesses account for about 80% of spending on major AI platforms, so that test is still open across the wider customer base. [2] [4]

WHY IT MATTERS

Hyperscaler bond issuance and wider spreads are documented. Whether they drive the sell-off is Constan's interpretation.

Read the full assessment

If financing tightens, the cost of AI compute and the pace of the buildout could shift, making customer return on investment the key metric to watch.

Research is done; the dossier is below.

Executive brief

Andy Constan of Damped Spring argues that the AI boom is starting to weigh on stock markets, not just lift them. Companies that once spent cash equal to about 2% of GDP buying back their own shares are now issuing new stock and bonds to pay for chips and data centres, according to the Excess Returns transcript. He calls this the first of "three poisons." The second is profit expectations that GDP growth can't fill. The third is borrowing costs that stay high: the 10-year yield has been above 5% for about two months. This is one analyst's commentary, recorded after a sharp November tech sell-off. Other analysts read the same market moves differently.

What changed and event timeline

  1. Constan calls a bubble regime

    His essay argues that AI investment has pushed equities into a bubble. In his view that calls for smaller positions and an exit plan made in advance, not attempts to time the top.

  2. Hyperscaler borrowing spree

    Five hyperscaler bond deals raised about US$67B. Oracle also paired a US$10B bond sale with a US$10B stock offering.

  3. Tech rout

    The Philadelphia Semiconductor Index fell about 12% and the Nasdaq 5.5%, wiping out roughly $2.5T in market value. Reports linked the drop to fears that Democrats' midterm gains could slow the AI buildout.

  4. Yields above 5%

    The 10-year Treasury yield passed 5%, its highest in 24 years. Robin Brooks says fiscal risk, not AI borrowing, is the main cause.

  5. Lenders pull back

    Banks and credit investors are looking harder at AI debt deals. Spreads on 10-year hyperscaler bonds have widened about 50bp so far this year.

  6. The episode airs

    Constan sets out his three-pressure framework on Excess Returns' monthly First Principles show.

Capabilities and access

This is market commentary, not a product launch, so no model or version is involved. The episode is free on Spotify and the Excess Returns Substack, which also carries a full transcript. On AI itself, Constan says he is optimistic about what the technology can do.

Read the full section

This is market commentary, not a product launch, so no model or version is involved. The episode is free on Spotify and the Excess Returns Substack, which also carries a full transcript. Damped Spring's detailed research sits behind a subscription at dampedspring.com. On AI itself, Constan says he is optimistic about what the technology can do. His test is commercial: can customers use it profitably at prices that also reward the companies selling it?

Technical analysis for researchers and developers

  • Supply of stocks: When companies issue shares instead of buying them back, the number of shares for sale goes up.
  • Kalecki-Levy profits equation: This framework treats government deficits as a major source of corporate profits.
  • Market signals of demand: He judges investor appetite by the discounts companies must offer to sell new securities, by credit spreads, and by how new issues trade afterwards.
Read the full section
  • Supply of stocks: When companies issue shares instead of buying them back, the number of shares for sale goes up. Constan argues that if AI spending only meets expectations rather than beating them, that extra issuance becomes a net drag on prices (transcript).
  • Kalecki-Levy profits equation: This framework treats government deficits as a major source of corporate profits. Constan reads deficits that are flattening, not growing, as a drag on earnings expectations (transcript).
  • Market signals of demand: He judges investor appetite by the discounts companies must offer to sell new securities, by credit spreads, and by how new issues trade afterwards. All of these are observable and can be checked.
  • The underlying numbers: FactSet projects big tech's combined free cash flow at –US$45.5B for 2026 and –US$119.4B for 2027 (Beansprout).

Claims and evidence

  • Net buybacks have flipped from about 2% of GDP to net issuance: Constan's figure, from the transcript.
  • Heavy hyperscaler bond issuance: independently reported. Goldman projects $420B of global hyperscaler issuance in 2027.
  • High rates are lasting: the 10-year above 5% is consistent with Fortune. The transcript also cites mortgage rates of 7.25%.
Read the full section
  • Net buybacks have flipped from about 2% of GDP to net issuance: Constan's figure, from the transcript. It is not checked against primary flow-of-funds data.
  • Heavy hyperscaler bond issuance: independently reported. Hyperscalers have sold $370.7B of bonds this year, about a fifth of a record $1.88T in US investment-grade issuance (Yahoo/Reuters search coverage). Goldman projects $420B of global hyperscaler issuance in 2027.
  • High rates are lasting: the 10-year above 5% is consistent with Fortune. The transcript also cites mortgage rates of 7.25%.
  • AI demand is still strong: Azure grew 48% year on year, AWS 39% and Google Cloud 103% (Beansprout). This sits awkwardly with the profit-gap thesis.

Context and prior work

Constan had already warned in May that markets were in a bubble regime (Substack). He has made the issuance argument before, in an earlier episode on IPOs and AI capex. He has also said that Treasury bond buybacks hold down long-term rates (Scuttlebutt). Shortly before the episode, he said the so-called "circular" AI deals, where suppliers invest in their own customers, ultimately depend on end customers paying (Gokhshtein).

Limitations, safety and contested findings

  • What is driving yields is disputed. Brooks attributes the rise to fiscal stress: a $1.99T deficit and $1.05T in net interest costs.
  • The trigger is disputed too. Coverage tied the Nov 5 drop to midterm politics (CNN), not mainly to financing strain.
  • Some of Constan's figures are unsourced. The 10–15% further fall before policymakers step in, and the stock-issuance examples he gives, are his estimates.
Read the full section
  • What is driving yields is disputed. Brooks attributes the rise to fiscal stress: a $1.99T deficit and $1.05T in net interest costs. He rejects the idea that AI borrowing is crowding out Treasuries (Fortune).
  • The trigger is disputed too. Coverage tied the Nov 5 drop to midterm politics (CNN), not mainly to financing strain.
  • Some of Constan's figures are unsourced. The 10–15% further fall before policymakers step in, and the stock-issuance examples he gives, are his estimates. The transcript doesn't source them.

Business and practitioner implications

  • The cost of AI capacity depends partly on credit markets.
  • Constan's deciding test is unit economics. Customer ROI, and pricing that pays suppliers, are things buyers and builders can measure themselves.
  • Usage is concentrated. On Ramp, the top 1% of businesses account for about 80% of spending on major AI platforms (Beansprout).
Read the full section
  • The cost of AI capacity depends partly on credit markets. Wider spreads and more cautious lenders, especially toward neoclouds like CoreWeave and Nebius, could change the terms on which compute is financed (Yahoo Finance).
  • Constan's deciding test is unit economics. Customer ROI, and pricing that pays suppliers, are things buyers and builders can measure themselves.
  • Usage is concentrated. On Ramp, the top 1% of businesses account for about 80% of spending on major AI platforms (Beansprout). That suggests broad profitability across customers is still unproven.

This dossier is not investment advice.

FOLLOW THE EVIDENCE

The source trail.

Sources (11)
A LITTLE LESS NOISE. A LOT MORE CONTEXT.

Stay curious.
Follow the evidence.

Independent perspectives, the original sources, and room for the questions that don't have easy answers.

How we build the brief
Connect with us

Find us where you already read.