AI demand is holding up; the financing chain is the fault line

The market’s clearest geopolitical signal is a split decision. Enterprise-facing software names are stabilizing, while the most debt-dependent AI infrastructure providers remain under pressure. That is not a clean rejection of AI demand. It is a repricing of who can finance capacity, at what cost, and with how much exposure to energy and policy shocks.

On October 26, Datadog (DDOG) closed at $280.24, up 3.93% on the day, while Snowflake (SNOW) closed at $334.95, up 2.93%. CoreWeave (CRWV) also rose, but only 1.76% to $55.19 after a much more severe financing-driven selloff. In after-hours trading, DDOG was $280.85 as of 16:26 ET and CRWV was $55.25 as of 16:27 ET; SNOW was unchanged at $334.95 as of 16:22 ET. These are delayed market snapshots from FMP, not a claim that the underlying risk has disappeared.^[1]

The working hypothesis for this basket was that resilient demand and earnings growth could support DDOG, SNOW, CRWV, AN, RH, WSM, ETH and LZB over the next year. The latest evidence supports only the first half of that thesis: usage-sensitive software is being treated differently from leveraged capacity, while consumer and crypto-linked exposures remain more vulnerable to the macro shock.

The market tell: demand is not the same as funding

The most important distinction is between companies selling software into AI workloads and companies financing the physical compute itself. DDOG and SNOW can benefit if customers continue deploying applications, monitoring systems and moving data, even if the capital markets become less generous. CRWV has a different equation: it must fund chips, data-center capacity and power commitments before those assets generate their contracted cash flows.

CNBC reported that CoreWeave’s unsecured debt was yielding about 15%, its five-year credit-default-swap spread had risen to 1,145 basis points, and the stock was down 36% for October as of the report’s October 23 update. The same report said the company had about $20.7 billion of debt due through 2030 and another $14.9 billion thereafter.^[2] Those figures explain why a modest equity rebound does not equal a repair in financing conditions.

Yahoo Finance’s October 26 account made the same distinction in plainer market language: lenders were becoming selective rather than disappearing entirely, with investors favoring companies that can fund AI exposure from stronger balance sheets. It reported that a basket of eight neoclouds had lost more than 40% since late July, while Microsoft had gained nearly as much.^[3] The signal is a sorting mechanism: AI spending can remain intact while the landlord layer absorbs the stress.

Geopolitics raises the cost of the build-out

The geopolitical channel is not limited to export controls. The current Middle East conflict is keeping pressure on energy markets, while disruption risk around shipping lanes can raise the cost and uncertainty of moving equipment, fuel and industrial inputs. That matters because data centers are not weightless software businesses: they require land, grid connections, cooling systems, power contracts, networking equipment and access to debt.

A market overview published by BlackRock’s Investment Institute described persistent pressure on energy prices from the Iran war and linked the AI cycle to U.S.-China competition, technology sovereignty and a broader move toward energy security and reindustrialization. Its page was not accessible for full verification in this research pass, so the claim is treated as contextual rather than as a primary forecast.^[4] The more directly verifiable market evidence is the financing response: rising yields and political uncertainty are making the cost of capacity harder to ignore.

What the basket is saying

DDOG and SNOW: Their October 26 gains are consistent with investors still assigning value to resilient enterprise usage and AI-linked consumption. That is evidence in favor of the demand thesis, not proof that growth expectations will be met. The next test is whether customer expansion, usage intensity and operating leverage continue to offset valuation sensitivity to rates.

CRWV: The stock’s small rebound sits on top of a far more consequential credit signal. CNBC reported that more than two-thirds of CoreWeave’s prior-year revenue came from Microsoft, and that the company combines infrastructure-backed borrowing, corporate debt, equity and customer prepayments.^[2] That mix may provide flexibility, but it also makes the cost and timing of external capital central to the equity case.

AN, RH, WSM and LZB: The auto-parts and home-furnishings names were down on October 26: AN fell 0.53% to $160.20, RH fell 1.31% to $135.63, WSM fell 0.99% to $249.96 and LZB fell 1.20% to $29.62 at the regular close.^[1] This is consistent with a macro channel in which energy, freight and financing costs pressure discretionary demand and margins. It does not establish a single geopolitical cause for each move.

ETH: ETH was down 3.29% to $22.08 at the regular close.^[1] The symbol’s feed value should be treated cautiously because it does not resemble a conventional spot-crypto quote; the defensible conclusion is only that the supplied market snapshot showed weakness, not that it measures the full global ETH market.

What would confirm or break the hypothesis?

The bullish version requires more than continued AI headlines. It requires evidence that enterprise usage is translating into durable revenue, that software companies can expand margins as workloads rise, and that infrastructure operators can secure power and capital without repeatedly diluting shareholders or refinancing into punitive yields.

The bearish version would gain credibility if three things happen together: software consumption slows, customer concentration rises, and credit spreads remain wide enough to delay or reprice capacity additions. In that scenario, the demand story may still be true in the abstract while the returns to infrastructure owners deteriorate.

The balanced interpretation is therefore conditional. Resilient demand can support the software layer, but it does not immunize the ecosystem from energy shocks, shipping friction, permitting delays, higher rates or debt-market selectivity. The market is asking a narrower question than “is AI real?” It is asking which business models can carry the cost of making AI real.

What to watch next

1. CRWV credit, not just CRWV equity: Watch unsecured yields and CDS spreads. A durable improvement in the cost of insurance would be more informative than a one-day stock bounce. The October 23 report put the five-year CDS spread at 1,145 basis points.^[2]
2. Funding and customer concentration: New borrowing, customer prepayments, equity issuance and the share of revenue tied to a small number of customers will show whether growth is self-funding or balance-sheet dependent.
3. Energy and shipping signals: Track oil, freight, insurance and power-market developments alongside data-center permitting. These are input costs and timing risks, not merely macro background.
4. Software usage versus optimization: DDOG and SNOW need to show that AI workloads are additive rather than simply shifting or compressing existing cloud budgets.
5. The consumer read-through: AN, RH, WSM and LZB will help distinguish a narrow infrastructure credit event from a broader hit to household demand and discretionary margins.

The evidence currently supports a two-speed conclusion: demand-sensitive software is holding up better than leveraged AI capacity, but the geopolitical and rates backdrop is making the infrastructure economics less forgiving. That is a market test of financing quality and execution—not yet a verdict that the demand thesis has failed.