The AI Bubble Is Now A Debt Disaster

Microsoft

Across the shows: 32 speakers in other episodes, 33 bullish and 4 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Microsoft's voluntary buyouts are a cash-conservation tool to fund AI capex, not a sign of distress.

  Microsoft is offering voluntary buyouts to about 60,000 employees, more than a quarter of its global workforce, after earlier buyouts and 4,800 job cuts. Snider says the company describes this as adapting to a changing landscape, but he sees it as conserving future cash, since payroll is among the few large recurring costs a tech company can control while data center, chip and energy contracts are hard to cut once signed. He stresses Microsoft is not financially distressed, but argues it is being forced to choose, with management prioritizing capital-intensive AI investment over other uses of cash. If internal cash generation falls short, he expects equity issuance to be the next release valve, which would dilute existing shareholders.

  “management is clearly prioritizing capital intensive AI investment over other uses of cash.”at 6:57

Alphabet

Across the shows: 46 speakers in other episodes, 35 bullish and 25 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Generative AI threatens Google's search ad model, making its AI transition extremely expensive and risky.

  Alphabet lost over $500B in market value in its largest one-day decline after disappointing search and advertising sales, while facing more than $1T in future payments for data centers, computing and energy. Snider argues AI answers cost more to compute and may not generate the same ad opportunities as traditional search, so Alphabet must spend heavily to defend a business AI is likely to disrupt while building a replacement revenue stream. He adds that over a third of its backlog came from a joint venture it partly owns, so the demand is less reassuring than outside customers paying profitable prices. Until Google shows it can monetize AI as it monetizes search, each capex forecast looks like a bill to be paid.

  “Generative AI threatens that model because AI answers are more computationally expensive to produce and may not generate the same advertising opportunities as traditional search results.”at 10:08

Anthropic

Across the shows: 78 speakers in other episodes, 94 bullish and 54 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Anthropic's compute debt is getting costlier, with lenders demanding higher rates and stronger default protections.

  Anthropic has to borrow tens of billions to build data centers and compute. Banks, Bank of America, Citigroup, Morgan Stanley and Blackstone committed about $63B to finance Broadcom chips and networking for Anthropic, but some lenders reportedly considered walking away. They said rates had risen, AI ventures were harder to finance and they were not being adequately paid for the risk. The final deal included higher rates and stronger guarantees protecting lenders if Anthropic defaults, and Broadcom supplied residual value support for the first lien. Snider reads this as rising financing costs and tighter terms for the company's build-out, alongside the Texas project delay tied to Anthropic.

  “And the final agreement included higher rates and stronger guarantees protecting lenders if Anthropic does default.”at 14:51

Broadcom

Across the shows: 9 speakers in other episodes, 7 bullish and 4 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Broadcom's residual-value guarantees for Anthropic financing show it is now backing customer chip debt.

  Banks committing around $63B to finance Broadcom chips and networking equipment for Anthropic had cold feet, saying rates were too low for the risk. Broadcom's CEO reportedly intervened, and the final deal had higher rates and stronger lender protections if Anthropic defaults. Broadcom also gave residual value support for the first-lien financing because banks did not want to rely only on the chips as collateral. Snider finds this extraordinary, since the chips are supposed to be scarce assets with nearly unlimited demand. He argues Broadcom is no longer just selling equipment but helping customers finance it and insuring the lenders, which deepens the circularity. Broadcom's CDS also hit a new record high.

  “Broadcom is no longer merely selling its equipment. It's increasingly helping customers finance purchases of that equipment and protecting the lenders making those purchases possible”at 15:36

Amazon.com

Across the shows: 30 speakers in other episodes, 39 bullish and 6 bearish calls, past 30 days

• Jeff Snider · Atlas FinancialBearish

  Amazon's AI spending overwhelmed free cash flow, pushing it toward $80B hardware-backed supplier financing.

  Amazon spent over $77B on property and equipment in one quarter, above expectations, producing a $42.4B free cash outflow despite 12% revenue growth to about $201.9B and sixth straight quarter of AWS acceleration. Snider argues the costs are arriving much faster than the cash return. Amazon has funded its build-out with cash flow and record bond sales and is now turning to supplier financing that could borrow up to $80B against the hardware it buys. That moves the cost across time and adds a lender, and the collateral (fast-depreciating AI hardware) may be worth less than projected if demand disappoints. Snider concludes the math doesn't work, and Amazon's CDS hit a record of about 94bp.

  “Amazon spending overwhelmed its free cash flow and pushed the company toward an 80,000,000,000 supplier financing facility.”at 17:31

AI infrastructure & data centers

• Jeff Snider · Atlas Financial

  Circular financing between chipmakers, AI firms and their customers makes AI demand look stronger than it is.

  Snider describes a loop where a chipmaker or tech company invests in an AI business, which then uses the money to buy the investor's chips, cloud capacity or services. That supports the investor's revenue and strengthens the AI firm's demand projections, which helps it raise more debt. Examples include Nvidia-backed Zenco borrowing about $3.1B to buy Nvidia GPUs, an Nvidia-backed neocloud discussing about $10B of financing to buy Nvidia chips for an Indonesian data center, and SoftBank, an OpenAI investor, helping broker a $40B bridge loan for AI data centers and chips. Individual transactions are legitimate, but the chain can depend on continued financing rather than end-user cash flow. Snider argues circularity looks strong on the way up but spreads losses faster on the way down, since one failed major customer can hit operators, chip suppliers, cloud providers, lenders and investors. Parts of the system may be borrowing against demand created by the financing system itself.

  “circularity makes the system look stronger on the way up, but spreads losses faster on the way down.”at 12:23

• Jeff Snider · Atlas Financial

  Delayed AI data center bridge loans trap bank balance sheets and crowd out other borrowers.

  Banks led by Morgan Stanley arranged about $15B to build an Anthropic-linked data center near Waco, Texas, with a $14.2B one-year bridge loan and each bank committing about $2B. They planned to refinance part of it through $5-7B of high-yield bonds in September, but Texas paused new data center projects to assess grid risks, so the bond sale was shelved and distribution may be spread over the next year. Snider argues nothing has to default for this to become a problem: construction revenue is delayed, interest accrues and banks keep exposure they expected to sell. Banks then become more selective and demand higher rates, collateral and equity, making the next project less profitable and the chain more fragile. AI bridge loans trap capital that could go to commercial real estate, leveraged buyouts and corporate borrowers, so AI crowds out other credit even if it does not collapse.

  “So a temporary distribution delay becomes a mechanism for restricting credit far and wide. Every dollar that gets trapped in an AI bridge loan is a dollar that can't be deployed as easily elsewhere.”at 16:52

AI bubble & tech valuations

• Jeff Snider · Atlas Financial

  The AI bubble will burst before its benefits arrive, with debt destruction and fallout likely around 2027.

  Snider has repeatedly said he hopes the AI bubble bursts sooner rather than later, because AI's benefits will accrue far down the road while a lot of debt destruction happens first. He calls AI a 2030 story but says its potential fallout is likely to hit sometime in the 2020s, maybe even 2027. His argument rests on the credit developments he has described: record CDS prices, Amazon's supplier financing, Alphabet's more than $1T in payment obligations, and banks stuck with data center bridge loans. These show financing that cannot keep pace with capital spending, so he expects the financial strain to surface before the technology pays off.

  “I really do hope the AI bubble bursts sooner rather than later because the benefits to AI, they're gonna accrue way down the road. And in between, there's a lot of debt destruction that's gonna happen before we ever get there.”at 2:17

Credit markets

• Jeff Snider · Atlas Financial

  Record CDS prices on AI giants show investors no longer see AI-linked financing as safe.

  Credit default swap prices on Amazon reached a record of about 94bp after its capex and slowest e-commerce growth in three years. Microsoft and Alphabet's credit risk gauges also hit record wides, Broadcom's protection reached a record high, and Intel's five-year CDS rose above 120bp to a record. Snider argues these prices don't mean these companies will default. They show investors no longer regard AI financing as unquestionably safe. The market only needs to believe leverage, cash flow or financing needs are moving in a less favorable direction. When that happens across a whole sector, risk premiums rise together, and the weakest borrowers are cut off while strong ones face higher cost, less flexibility and tighter covenants, which he calls credit rationing.

  “These prices don't apply that Amazon, Microsoft, or Alphabet are about to default and go out of business. They show that investors no longer regard the financing behind the AI boom as unquestionably safe”at 8:32