How resilient are Ukraine's essential digital services to sustained attacks on power, connectivity and computing infrastructure?

A’s strongest point is that cloud migration and hardened banking branches provide genuine operational resilience. Its weakest point is the threshold: avoiding prolonged nationwide collapse does not establish reliable access to essential services. The cited memorandum reports 98.5% availability within POWER BANKING, covering approximately 55% of branches; it does not measure completed customer transactions. Crucially, it describes online banking as available to clients “with internet connectivity.” That qualification is precisely the dependency under dispute—not evidence that the remaining branches failed. (bank.gov.ua)

I accept A’s working definitions: a data center houses computing, storage and networking equipment; failover switches operations to an alternative system; critical infrastructure supports essential societal functions; hybrid conflict combines conventional attacks with cyber and other nonmilitary operations.

My first counter-evidence concerns observed physical disruption. The UN Human Rights Monitoring Mission’s September 19, 2024 assessment, based on site visits and interviews, documented nine attack waves against electricity infrastructure between March 22 and August 31. Some cities subsequently experienced at least 12 hours of blackouts per day during summer. The mission reported disappearing telephone and internet connections and compromised health services. This establishes a physical-attack-to-service-disruption pathway without relying on disputed data-center strike allegations. It does not, however, provide hospital-by-hospital digital recovery times. (ukraine.ohchr.org)

Second, Kyivstar’s July 12, 2024 operational explanation identifies why backup power did not eliminate interruptions. Its base-station batteries offered up to four hours but needed four to six hours to recharge; fixed-network backup units needed eight hours. Repeated outages prevented full charging, and fixed-network backup could last under two hours at peak demand. The operator also explained that customers with electricity could lack internet because neighboring buildings supporting the network were unpowered. Thus, restoring one endpoint does not necessarily restore access to an intact overseas application. That last conclusion is an inference from the documented dependency. (kyivstar.ua)

These are historical limits, not an assertion that investment stopped. Kyivstar’s current resilience page advertises up to ten hours for mobile infrastructure and twelve for fixed internet, while explicitly qualifying endurance by outage frequency, load and battery charge and excluding certain damaged or inaccessible sites. Those improvements strengthen A’s case, but advertised autonomy still cannot substitute for measured service completion during repeated outages. (kyivstar.ua)

European hosting likewise reduces physical exposure rather than abolishing dependencies. Microsoft documented hosting across Europe and Russian intrusion attempts against 128 organizations in 42 countries outside Ukraine, including IT and infrastructure suppliers. This supports investigating cross-border cyber exposure; it does not prove that hosting Ukrainian workloads caused those attacks. (blogs.microsoft.com)

My normative conclusion is to judge resilience by completed payments, functioning hospital workflows and usable records, with geographic coverage and end-to-end recovery times. A has demonstrated meaningful continuity, but not that surviving systems remain reliably accessible throughout sustained backbone disruption.

A’s specific weakness is a mismatch between the question and its evidentiary threshold. Avoiding prolonged nationwide collapse is a weaker test than maintaining essential services during sustained infrastructure disruption. A concedes that annual payment totals cannot establish regional access or recovery after individual strikes, and that adequate hospital, register and industrial recovery evidence remains unavailable. Those concessions do not prove widespread failure; they mean A’s evidence cannot establish the end-to-end reliability under dispute.

The strongest evidence favoring A deserves full weight: the NBU recorded 8,184.8 million cashless card transactions in 2024, while 91.8% of all card transactions occurred domestically. This demonstrates substantial functioning payment infrastructure, not merely surviving backups. But 94.6% is the cashless share of recorded transactions—not the success rate of attempted payments. Annual aggregates cannot distinguish uninterrupted access from transactions deferred until connectivity returns. That is a measurement limitation, not an allegation that the recorded transactions failed. (bank.gov.ua)

A’s De Novo example actually sharpens my concern. Its April 26, 2025 report says maintenance temporarily reduced electrical redundancy; a hidden switch defect then caused complete power loss. Power returned after fourteen minutes, the cloud around 09:45, and the incident’s consequences were eliminated around 13:00. The report attributes the delay to restarting interdependent hardware and software. Thus electrical restoration, cloud restoration and customer-system recovery were distinct outcomes. This operator account establishes neither attack resilience nor independently verified essential-service recovery, but it directly demonstrates why infrastructure uptime understates disruption. (denovo.ua)

Independent evidence establishes the upstream attack pathway. UN monitors documented nine waves of electricity-infrastructure attacks between March 22 and August 31, 2024, affecting facilities in twenty regions. Their assessment describes compromised health services and disappearing telephone and internet connections during outages. These are observed civilian consequences, although they do not quantify digital hospital-workflow recovery. (ukraine.ohchr.org)

Network investment does not remove the measurement problem. Kyivstar now advertises up to ten hours of mobile backup and twelve for fixed internet, but qualifies endurance by outage frequency, duration, load and battery charge; its coverage claims exclude specified damaged or inaccessible infrastructure. These qualifications identify conditions requiring field measurement—not proof that every prolonged blackout exhausts backup power. (kyivstar.ua)

Cross-border hosting remains a genuine protection. Microsoft reports government continuity through European hosting alongside Russian intrusion activity outside Ukraine. My inference is that geographic separation reduces domestic physical exposure while leaving dependence on providers and connectivity; the evidence does not establish that Ukrainian hosting caused attacks on European providers. (blogs.microsoft.com)

My normative conclusion is therefore narrower than predicting national collapse: measure completed essential workflows during disruption, geographic exclusion, and time to usable recovery. Data preservation and substantial annual throughput are achievements, but neither answers whether a disconnected institution can retrieve records or complete payments when needed.

A’s strongest argument is that Ukraine has demonstrated operational resilience, not merely preserved backups: the NBU recorded 8,184.8 million cashless transactions in 2024, and 91.8% of all Ukrainian-issued card transactions occurred domestically. (bank.gov.ua) Microsoft also reported that European cloud hosting sustained government operations. (blogs.microsoft.com) Together, these observations support A’s qualified conclusion that substantial national functionality can survive sustained attacks despite serious local interruptions.

My position remains stronger on the question of essential-service reliability because national continuity and timely local access are different outcomes. A’s proposed falsification threshold—simultaneous failure across most regions for over 72 hours—tests nationwide collapse, not whether redundancy protects essential workflows. I accept the evidence of national functionality; I reject treating it as sufficient evidence of reliable access during disruption. The NBU’s 94.6% cashless share describes recorded transaction composition, not attempted-payment success, and its annual totals do not report strike-specific interruption durations. (bank.gov.ua)

Positive evidence supports this distinction. UN monitors, drawing on site visits and interviews, documented nine electricity-attack waves between March 22 and August 31, 2024. Some cities experienced at least twelve hours of blackouts daily; communications disappeared during outages, and essential services, including health services, were compromised. These findings establish actual consequences of physical-network disruption, although they do not isolate digital hospital failures or measure their recovery times. (ukraine.ohchr.org)

Kyivstar’s July 12, 2024 explanation identifies a concrete mechanism: base-station batteries required four to six hours to recharge, while fixed-network backup units required eight. Repeated outages prevented complete charging; customers could have electricity yet lack internet because equipment serving neighboring buildings was unpowered. These are historical operating observations, not a claim about today’s maximum endurance. My inference is that an intact overseas application can remain unreachable because its access path fails independently. (kyivstar.ua)

European hosting therefore remains a genuine protection, not a complete solution. Microsoft reported both government continuity through European hosting and intrusion attempts against 128 organizations in 42 countries outside Ukraine. This supports investigating provider and cross-border security dependencies, but does not establish that Ukrainian workloads caused those attacks or that overseas hosting increases net risk. (blogs.microsoft.com)

My normative standard is timely completion of essential tasks. I would change my assessment after an independently audited, geographically representative study of multiple verified attack waves showed that payments, hospital workflows, registry access and industrial services consistently met predeclared sector-specific deadlines during actual power and connectivity losses. It must reconcile local attempted and abandoned tasks with server logs, include disconnected institutions rather than dropping them from the sample, and measure battery depletion, generator refueling, failover and end-to-end recovery. Successful offline alternatives should count if they deliver the required service on time. Such results would demonstrate that redundancy protects usable services, not merely surviving systems.