Ukraine expands drone strikes on data centres owned by Russia’s Yandex

Ukrainian drone strikes partly disabled two Yandex data centres in successive attacks, including the Sasovo hub housing two of the company’s three AI-training supercomputers; the extent of damage and restoration prospects remain unclear. Yandex said the strikes hit core operations and disrupted services, while an analyst warned that continued attacks could eliminate the option of shifting workloads to other facilities. Yandex shares fell 3.75% on the Moscow exchange.

Analysis

The market-relevant risk is not a near-term global shortage of AI compute; it is that physical redundancy can fail when attacks are repeated across a concentrated footprint. If Yandex cannot restore workloads or shift them to the remaining sites, the damage could move from temporary service interruptions to lost B2B trust, customer migration, and higher resilience spending. That second-order effect matters more than one day’s outage: businesses may reassess dependence on a single domestic platform, while Russian data-centre operators face pressure to add geographic separation, backup power, and recovery capacity. Those costs could weigh on margins before any replacement capacity produces returns.

For NVDA, the destroyed or impaired A100-based capacity creates only a conditional replacement-demand channel. Procurement restrictions, payment and logistics barriers, and the possibility that equipment is recoverable make any read-through to revenue highly uncertain; the incident is not evidence of a material change to global accelerator demand. GOOG has no clear direct earnings exposure in the facts provided. A broader geopolitical risk premium for infrastructure security is plausible, but not enough here to support a directional trade in either US-listed name.

Near term, watch restoration updates and whether outages spread to business customers. Over 1–3 months, repeated strikes or disclosed customer losses would raise the probability of persistent disruption and defensive capex. Over 6–18 months, continued targeting could make data-centre location and redundancy a larger cost and valuation issue for regional technology firms. The contrarian point: the immediate share-price reaction may understate tail risk if attacks recur, but it may also overstate durable damage if workloads shift successfully and equipment proves recoverable. The unrelated gold-price headline in the article slug is not supported by the article body and provides no trade signal.

Key Decisions for Investors

• No directional GOOG or NVDA trade on this incident alone; the article supplies no evidence of material earnings exposure for either mapped company.
• Treat NVDA replacement demand as a watch item, not a thesis. Reassess only if credible reporting identifies unrecoverable accelerator losses, replacement orders, and a feasible procurement route; recovery of the equipment or continued procurement barriers would falsify the demand angle.
• Monitor Yandex disclosures over the next several weeks for restoration time, service availability, and B2B customer retention. Persistent outages or evidence that remaining sites cannot absorb workloads would strengthen the regional infrastructure-risk thesis; rapid recovery without customer losses would weaken it.
• If attacks broaden to additional data centres or produce sustained outages, review exposure to Russia-linked digital businesses and data-centre operators for resilience-related capex and customer churn risk. Do not infer a broad global AI or cloud earnings impact without evidence of cross-border supply or demand effects.