Geopolitical fragmentation is testing the durability of AI demand—and imported-goods margins

The market is separating two ideas that are often bundled together: resilient demand for software and compute, and resilient earnings. The first still has meaningful evidence behind it. The second is more conditional, because export controls, power equipment constraints, shipping disruption and tariffs can raise the cost and delay the delivery of the infrastructure and goods that sit underneath reported growth.

The market tell: usage is holding up, but the cost of delivery is becoming the risk

On October 19, DDOG closed at $260.36, up 5.36% from the prior close, while SNOW closed at $315.35, up 2.77%. CRWV closed at $61.12, down 0.44%; in post-market trading it was $61.25 at 19:59 ET, 0.21% above the 16:00 ET close. ETH was $2,396.29 at 20:27 ET, down 0.50% from the prior close, with a $2,393.84 extended print at the same time. These are snapshots, not proof of a single causal trade, but they show that the market is rewarding software exposure more readily than it is rewarding the capital-intensive infrastructure layer.^[1]

The current geopolitical backdrop makes that distinction important. Recent reporting describes proposed Chinese controls covering 40 key technologies, alongside renewed concern about U.S. restrictions on technology transfers and AI-chip access. A separate weekly technology review identified AI export controls, data-center power rules and chip-supply constraints as active issues in the same period.^[2][3]

DDOG: the strongest evidence for the resilient-demand thesis

Datadog’s latest available call gives the thesis its cleanest operating support. Management said Q2 revenue reached $1.12 billion, up 36% year over year, and that non-AI customer growth accelerated to the high-20% range. It also said more than 6,500 customers were sending data through one or more AI integrations; those customers represented about 80% of ARR.^[4]

That matters because it is not only a story about a small group of AI-native companies. Management described broad adoption across startups and large enterprises, with AI increasing the need to monitor, secure and operate cloud workloads. The counterpoint is valuation and usage sensitivity: observability spend can rise with AI activity, but customers can also scrutinize consumption, consolidate vendors or seek lower-cost architectures if budgets tighten.

For the hypothesis, DDOG is evidence in favor of durable demand. It is not evidence that geopolitical friction is irrelevant. Export controls can change where workloads run, which models customers can access and how much infrastructure is available; those changes may create monitoring demand while also altering the growth mix.

CRWV: demand is visible, but financing and physical capacity are the fault lines

CoreWeave’s call history shows why AI infrastructure should not be treated like ordinary software. In Q2 FY2026, management reported $2.6 billion of revenue, up 112% year over year, and a $104.2 billion revenue backlog, up 246% year over year. It also said margins had inflected higher and that pricing increased across SKUs after July.^[5]

But the same evidence highlights the risk. CoreWeave is building a business that depends on GPUs, power, data-center delivery, financing and customer commitments arriving in the right sequence. Management described the constraint as the speed at which high-performance capacity can be delivered, while the company’s own history includes data-center delivery delays and a large capital program.^[5]

That makes current export-control and power-policy developments economically relevant even if end-demand remains strong. A restriction can reduce available components or slow deployment; a power constraint can postpone revenue recognition; higher financing costs can reduce the return on capacity added ahead of demand. The base case can still be strong, but the range of outcomes is wider than the backlog headline alone suggests.

RH: geopolitical risk is already showing up as a margin and logistics problem

The consumer side of the requested scope provides a useful contrast. RH’s Q2 FY2026 call said revenue was $922.2 million, up 2.6% year over year, and that the company recognized a $55.1 million tariff benefit while expecting $13.9 million more in the second half. Management said it planned to use those proceeds to offset $50 million of unplanned supply-chain costs caused by a sustained spike in oil prices linked to the conflict in the Middle East.^[6]

This is the important distinction: a tariff refund can cushion a quarter, but it does not remove the operational exposure. RH management has previously described repeated tariff announcements, sourcing delays, out-of-stocks and several rounds of price negotiations. The company’s latest commentary also tied energy costs to the regional conflict.^[6]

RH therefore supplies evidence against the strongest version of the thesis. Demand can hold up while earnings quality becomes less clean. AN, WSM and LZB should be read through the same lens—freight, imported product, housing sensitivity and pricing power—but this pass did not produce equally specific, current transcript evidence for each name. It would be a mistake to turn the group into a uniform bullish or bearish basket.

What the broader scope says—and does not say

SNOW’s October 19 close was higher, while DDOG was the clearest positive mover in the supplied snapshot. That is consistent with the market continuing to value software demand and AI exposure. CRWV’s small regular-session decline and modest post-market rebound are consistent with a more cautious view of infrastructure economics, but they do not establish a durable divergence.^[1]

ETH’s decline is a useful reminder that digital-asset demand is not the same thing as enterprise software demand. Its price can be influenced by liquidity, positioning and crypto-specific factors that do not map cleanly onto DDOG, SNOW or CRWV. The available evidence is therefore strongest for a differentiated thesis: recurring software usage may remain resilient, while the physical and financial inputs required to monetize AI remain exposed to policy shocks.

What to watch next

1. Export-control scope and implementation. Watch whether proposed technology controls become specific licensing rules, broad restrictions or negotiating leverage. The market impact will depend on affected components, jurisdictions and compliance timing—not just the headline.
2. Data-center power and delivery milestones. For CRWV and the wider AI infrastructure chain, track energized capacity, deployment delays, financing terms and customer concentration alongside backlog.
3. Usage quality at software vendors. For DDOG and SNOW, the key question is whether AI-driven usage becomes recurring production activity or remains experimentation that customers can trim.
4. Freight, oil and tariff pass-through. For RH, AN, WSM and LZB, monitor inventory availability, lead times, price increases and gross-margin recovery rather than treating tariff relief as recurring operating improvement.
5. The market’s relative response. If software continues to outperform infrastructure while demand indicators remain firm, investors may be pricing execution and financing risk rather than rejecting AI demand outright. If both groups weaken together, the concern would be shifting from delivery friction to a broader spending slowdown.

The balanced conclusion is that the earnings-growth thesis remains plausible, but it is not a single tradeable factor. DDOG’s usage evidence and CRWV’s backlog support the demand side; RH’s tariff and oil commentary shows how quickly geopolitical conditions can convert demand into a margin test. The next confirmation should come from realized capacity, recurring usage and unassisted margin performance—not from a risk-on or risk-off label.