Telecom operators in 2026 face an architectural inflection point: legacy NFV environments aging out, edge-compute demands growing, sovereign-cloud branded products launching across the sector, AI workloads on customer data, and NIS2 essential-entity compliance. The architectural answer is a coherent platform that runs at core data centres, regional sites, and customer edge — under one operations model.
Ænix builds platforms for telecom operators across the EU, DACH, and Central Asia. Same platform, Cozystack, running at core and edge.
Pairs with: Ænix Public Cloud Platform for telcos launching customer-facing sovereign cloud products at multi-region scale; Private Cloud Platform for NIS2-aligned internal cloud and edge platforms.
What telecom operators come to us for
- Sovereign cloud product launches — operators offering customer-facing sovereign cloud (regional telco sovereign cloud products, member-state-specific sovereign products). See Data sovereignty.
- NFV → Kubernetes platform transitions — legacy NFV environments replaced with Kubernetes-native equivalents.
- Edge cloud at scale — central core + regional + customer-edge sites under unified operations.
- Sovereign AI infrastructure — telco-data analytics, customer-care AI, network-ops AI.
- NIS2 compliance — telecom is an essential entity under NIS2.
- Hyperscaler exit — sustained workloads where economics no longer fit.
Why telco architecture is different
- Multi-site operations — core DC + regional sites + edge sites + customer-premises equipment, under one platform abstraction.
- 5G + AI convergence — network slicing, MEC, and AI inference at edge.
- Regulatory complexity — NIS2 essential-entity, sectoral telco regulators, sovereignty mandates per jurisdiction.
- Customer-facing offers — many telcos resell platform capability to enterprise customers; multi-tenant customer-facing model is structural.
- Long depreciation cycles — hardware refresh cycles longer than typical enterprise; platform must work on multiple hardware generations.
The dataplane question
Everything above is true of any distributed estate. What makes telco different is that a share of the workload does not tolerate a generic Kubernetes dataplane, and a platform that cannot say how it handles that is not a telco platform.
- SR-IOV and device passthrough. Packet-processing functions want a VF bound directly to the workload, not a veth pair. Cozystack supports SR-IOV virtual functions and PCI passthrough into both KubeVirt VMs and containers, so a VNF that was certified on a VF keeps its VF. Where a function needs a whole NIC or an accelerator, it gets one.
- DPDK and userspace networking. Functions built on DPDK bypass the kernel network stack entirely; they need hugepages, CPU pinning, isolated cores and a NUMA-aware placement, not a best-effort scheduler slot. Those are node-level and pod-level settings on the platform, applied per workload class rather than cluster-wide, so a DPDK function and an ordinary microservice share a cluster without either compromising.
- CNFs and the VNFs that are not ready to become CNFs. Vendor network functions arrive as VMs with a support matrix, and the transition to containers happens on the vendor’s schedule, not yours. Both run side by side on one platform: KubeVirt for the VM-packaged functions, containers for the cloud-native ones, one API, one lifecycle, one observability stack.
- Cilium and eBPF for everything else. Management, OSS/BSS, customer-facing services and MEC applications run on the eBPF dataplane, with network policy and observability that do not require a sidecar per pod.
The honest boundary: this is a platform for the estate around the network — MEC, edge applications, OSS/BSS, customer-facing cloud, AI inference, and the VNFs that live in a data centre. It is not a claim to replace a purpose-built packet core.
How Ænix engages with telecom operators
Standard Platform Readiness Assessment with telecom-specific workstreams: multi-site architecture, edge readiness, sovereign-cloud product packaging (where applicable), NIS2 controls, AI infrastructure for telco use cases.
Phase 2 implementation typically spans 6-24 months for a multi-site telecom platform.
What runs on Cozystack in telecom
"Running Nuvolos on sovereign bare metal meant we needed hyperscaler-grade multi-tenancy without the hyperscaler. Cozystack delivered exactly that — isolated tenants, shared GPUs, KubeVirt for legacy workloads — and it's why we can promise our customers infrastructure control, not just a workspace."
Alexandru PopescuCEO, Nuvolos
Telco engagements sit in the same NDA cohort as the bank work; naming is permitted from mid-2027. The closest written-up case is a telecom operator that built a corporate AI platform on Cozystack and shipped the same distribution into a state-owned end customer.
Why Ænix specifically for telecom
- Edge-to-core platform. Cozystack runs at scale at core data centres, regional sites, and edge — under unified operations.
- Multi-tenant customer-facing model — Tenant CRD designed for service-provider model, suitable for telco customer-facing cloud products.
- Open-source foundation — Apache 2.0 platform; no per-CPU pricing, no vendor lock-in across hardware refresh cycles.
- Sovereignty-aligned — air-gapped deployments supported, customer-controlled encryption, audit-readiness for NIS2.
- EU + Central Asia teams.
How to start
- Sovereign AI — telecom AI use cases
- Data sovereignty — telco regulatory trigger
- Private cloud consulting — broader scope
- Platform Readiness Assessment
- Cozystack
Ænix is the team behind Cozystack (CNCF Project), and we offer Ænix Platform — our commercial productized offering based on Cozystack, Kubernetes Certified Distribution.