AI demand reshapes the server and storage stack
Two Selectel articles describe AI infrastructure as a combined compute, memory, power and storage challenge rather than simply a race for faster accelerators. In its processor analysis, Selectel reports AMD’s specifications for EPYC 9006 “Venice,” including up to 256 cores and 512 threads, 16 memory channels, PCIe 6.0 and CXL 3.1. The article also highlights the new SP7 platform, power requirements of up to 600 W and AMD’s planned Verano processor for its Helios GPU racks. Separately, Selectel cites third-party pricing showing a 30 TB enterprise TLC SSD at $22,600 and a same-capacity HDD at $1,216 in the third quarter of 2026. It attributes this gap to AI-driven demand, constrained NAND production and large cloud customers reserving storage supply. These figures and comparisons are claims reproduced or interpreted by Selectel from AMD, manufacturers, analysts and financial disclosures; they are not independent CloudInsights benchmarks.
The emerging priority is infrastructure density, not component speed alone
Taken together, the articles indicate that Selectel is preparing its audience for a more vertically integrated and capital-intensive phase of cloud infrastructure. Its technology direction emphasizes high-core-count CPUs, faster memory fabrics, CXL, dense GPU racks and tiered storage. The strategic trend is clear: AI platforms need rapid data delivery near accelerators, but the expanding body of training data, checkpoints and other retained information keeps capacity-oriented HDD storage relevant. Selectel’s view that Venice initially suits hyperscalers and large AI projects, while many other customers may remain on EPYC 9005 Turin, also implies a pragmatic adoption strategy rather than an immediate wholesale refresh. For Russia and the wider Eastern European context, that focus is regionally relevant because acquisition cost, power density, cooling design, component lead times and long-term access to replacement hardware may matter as much as headline processor specifications.
Power, supply and procurement remain material constraints
The proposed architecture carries several risks. According to Selectel, exploiting Venice’s memory bandwidth requires extensive memory population, while a new socket forces platform replacement and 600 W processors make liquid cooling increasingly relevant. Its storage analysis likewise suggests that neither flash nor high-capacity HDD supply should be assumed to become cheap or abundant quickly; manufacturer road maps for 44 TB, 50 TB and 55 TB drives do not establish mass availability or lower cost per terabyte. There are also important evidence limits: the articles provide no Selectel workload benchmarks, service availability dates, local prices, energy-cost modelling or total-cost comparisons between current and next-generation platforms. Most importantly for a Russia-based provider, they do not address sanctions, export controls, vendor support eligibility, payment routes, firmware access or continuity of spare-parts supply. Eastern European buyers should therefore treat the hardware road maps as planning inputs, not proof that particular configurations can be procured, supported or operated throughout their intended lifecycle.
Decision guidance and sources: Selectel’s AMD EPYC 9006 analysis and Selectel’s enterprise storage-price analysis
For IT decision-makers in Eastern Europe, the practical conclusion is to evaluate AI infrastructure as a complete system. CPU density should be tested against software licensing, memory population, accelerator feeding, rack power and cooling; storage should be separated into performance and capacity tiers rather than standardized on SSDs by default. Selectel’s review of AMD EPYC 9006, Embedded 9006 and Verano supports cautious planning for denser future compute, while its assessment of HDD and SSD economics under AI demand argues for retaining magnetic storage where latency requirements permit. Before selecting Selectel or designing around the cited components, buyers should request workload-specific benchmarks, confirmed local configurations, delivery and replacement commitments, power assumptions and documented compliance with all applicable sanctions and export-control rules. The strongest near-term strategy is staged adoption with contractual supply safeguards, not migration based solely on vendor road maps.