GPU flexibility and DDoS readiness shape the infrastructure proposition
Three Gcore announcements present AI compute and cyber resilience as complementary infrastructure priorities. Gcore says its new GPU Virtual Machines use NVIDIA Hopper GPUs and high-bandwidth NVIDIA Quantum InfiniBand networking, initially from its Sines-3 sovereign AI region in Portugal. Customers can select configurations ranging from one to eight GPUs, while GPU billing pauses when a virtual machine is powered off; storage and IP charges continue. Separately, Gcore reports that its Network Layer DDoS Protection was connected to Armenian telecommunications provider Ucom within one day to protect public live-broadcast infrastructure during Armenia’s 2026 parliamentary elections. According to the companies, the environment remained available and experienced no observed DDoS-related disruption during the engagement. Gcore’s own Radar analysis also claims that recorded attacks rose from 512,000 in Q4 2024 to 1.3 million in Q4 2025, with peak attack volume reaching 12 Tbps. These figures come from Gcore’s data and have not been independently validated here.
A European platform strategy built around sovereignty, elasticity and protection
The emerging strategic direction is an integrated infrastructure proposition rather than a narrow GPU or security play. The Portugal launch suggests that Gcore sees sovereign European AI capacity as a route to customers that need regional infrastructure but do not want long-term bare-metal commitments. Its stop-and-resume VM model is aimed at experimental, burst-oriented and variable workloads, complementing the provider’s existing bare-metal and spot GPU options. At the same time, the Ucom engagement and Radar findings position distributed network protection as an operational prerequisite for cloud-dependent services. The connecting theme is workload readiness: capacity should be available when AI teams need it, while protection should already be in place before a high-profile event or attack. For Western Europe, this combination is relevant to enterprises balancing AI experimentation, cost control, data-location requirements and service continuity.
Claims require workload-level and procurement scrutiny
The announcements leave important questions for prospective customers. Gcore does not provide comparative GPU pricing, measured training or inference performance, utilisation data, availability commitments, capacity levels at Sines-3, or detailed definitions of sovereignty in the selected material. Pausing GPU billing may reduce idle-compute expenditure, but buyers still need to model storage, networking, IP, data-transfer and operational costs, as well as restart times and GPU availability. The Ucom account describes successful preparedness, yet no attack was observed, so it does not demonstrate mitigation performance under a live DDoS event. Likewise, the Radar statistics reflect Gcore’s own visibility and methodology; they should not automatically be treated as representative of the entire threat landscape. Western European buyers should also verify contractual data residency, support arrangements, incident reporting, portability, security controls and dependencies on NVIDIA technology.
Buyer view: broadcast resilience case, GPU VM launch and DDoS findings
For IT decision-makers in Western Europe, Gcore merits evaluation where sovereign European AI infrastructure, burstable GPU access and distributed DDoS protection intersect. The provider’s material indicates a coherent direction: bring flexible AI compute closer to European users and pair cloud services with network-level resilience. That direction is strategically credible, but the supplied evidence is not sufficient for a provider-wide performance or value judgment. A sensible assessment should therefore use a representative proof of concept, total-cost modelling, contractual review and a tested incident-response exercise. Buyers can examine Gcore’s GPU Virtual Machines announcement, its account of the Ucom election-broadcast deployment, and the Gcore Radar DDoS trends release. The practical conclusion is to shortlist Gcore for relevant regional workloads, while requiring verifiable service metrics, transparent economics and precise sovereignty commitments before production adoption.