infrastructure hardware · network · compliance

The cloud runs on hardware, not slides

Seven regions across Russia and CIS, our own EPYC and Xeon Platinum hardware, H100 and B300 GPUs, NVMe replication for databases, dual-uplink in every rack. No marketing euphemisms here — what's where, plainly.

7regions · 12 AZsgeography
99.95%compute SLA · 10× creditsavailability
≤ 1 msp99 between AZs inside a regionnetwork
8certifications and attestationscompliance
geography · where your cloud physically sits

5 live regions in Russia and 3 more coming in CIS

Every region has at least 2 availability zones (AZs). Between AZs inside a region — a dedicated dark-fiber network with guaranteed ≤ 1 ms latency. Between regions — our own IP transit with redundancy.

RU-MSK-1RU-MSK-2RU-SPB-1RU-EKB-1RU-NSK-1KZ-AST-1AM-EVN-1RS-BEG-1
RU-MSK-1
Moscow
M9 · IXcellerate · 3 AZ
Available
RU-MSK-2
Moscow-2
DataLine · 2 AZ
Available
RU-SPB-1
Saint Petersburg
Linxdatacenter · 2 AZ
Available
RU-EKB-1
Yekaterinburg
1 AZ · GPU pool
Available
RU-NSK-1
Novosibirsk
1 AZ · low-latency Asia
Available
KZ-AST-1
Astana
Tier III · 1 AZ
Q2 2026
AM-EVN-1
Yerevan
DA-CMC · 1 AZ
Q1 2026
RS-BEG-1
Belgrade
Edge · 1 AZ
Q3 2026
hardware · what's actually spinning in the rack

EPYC, Xeon Platinum, NVMe Gen5. No compromise configs.

We don't build a cloud on 2018-era oversold nodes. Every platform is the current generation, bought through direct channels, rotated on a 3-year cycle.

Compute · Standard

AMD EPYC Genoa 9554

The base line for VMs, K8s nodes, app servers. High clock, large L3.

Cores64C / 128T
Base / Boost3.1 / 3.75 GHz
L3256 MB
RAMDDR5-4800 ECC
Compute · Memory

Intel Xeon Platinum 8480+

For databases, analytics, in-memory workloads. Up to 4 TB DDR5 per node.

Cores56C / 112T
Base / Boost2.0 / 3.8 GHz
L3105 MB
RAMup to 4 TB
Storage · NVMe

Samsung PM1743 · 7.68 TB

U.2 NVMe Gen5 for NVMe Replicated volumes. 13 GB/s read, 1 DWPD endurance over 5 years.

Read13 GB/s
Write6.6 GB/s
IOPS 4K2.5M / 200K
MTBF2M hours
Storage · SSD

Micron 7450 PRO · 3.84 TB

SATA for general-purpose disks and S3-Cold archive pools.

Read6.8 GB/s
Write5.6 GB/s
IOPS 4K1M / 350K
DWPD1
gpu · for inference and training

Four generations of NVIDIA. Including Blackwell B300.

B300from ₽820/hourMIG 141 GB

NVIDIA Blackwell B300

192 GB HBM3e · NVLink-5 · 12 GPU/server

H100from ₽250/hourMIG 40 GB

NVIDIA H100 SXM5

80 GB HBM3 · NVLink-4 · 8 GPU/server

A100from ₽180/hourMIG 20 GB

NVIDIA A100 80GB

80 GB HBM2e · 8 GPU/server

L40Sfrom ₽95/hourFull card

NVIDIA L40S

48 GB GDDR6 · for inference and rendering

network · how traffic is wired

Spine-leaf topology with dual uplink in every rack

Network design is a modern Clos fabric. No legacy core/aggregation/access from 2009. AS announced to 12 exchange points, dual-stack IPv4/IPv6 at every layer.

outer border
Edge / Internet peering
12 IX points · 4 transit providers · L3/L4 DDoS protection included
≥ 2 × 400G
spine
Spine Layer · Arista 7800R3
L3 underlay, EVPN-VXLAN overlay, BGP EVPN. 25.6 Tbps line rate per switch.
100G / 400G
leaf
Leaf · Arista 7050X3 / Mellanox SN3700
2× uplink to spine. Dual-homing via mLAG. Node uplinks at 25/100 Gbit.
25G / 100G
compute / storage
Hosts · Bonded NICs (LACP)
All nodes: bond0 in 802.3ad mode, active-active. SR-IOV for VMs, RDMA for GPU/storage.
2× 25G
Latency
≤ 1 ms p99
between any two AZs inside a region
DDoS
800 Gbit/s
L3/L4 protection on for every customer
Anycast
AS208924
our own autonomous system, peering in RIPE
compliance · certifications and attestations

8 certifications — so the lawyers can sleep

Every certificate and attestation is real, current, and we'll send copies on request. Happy to take part in your audits: checklists, policies, evidence — all ready.

152

FZ-152

Personal data storage in Russia. Levels UZ-2, UZ-3, UZ-4.

ISPDn

ISPDn attestation

Infrastructure ready for information-system attestation.

PCI

PCI DSS 4.0

CDE segment, isolated subnets, audit logs.

ISO

ISO 27001 / 27017

Information security management and cloud controls. Confirmed.

FSTEC

FSTEC / FSB

Certified cryptographic and integrated protection systems. Confirmed class K2.

GOST

GOST R 57580

Financial sector. Compliance level 0.85.

242

FZ-242

Data localization for Russian citizens on Russian soil.

GDPR

GDPR-ready

For AM/RS regions — DPA, SCC, DPO.

SLA · availability guarantees

Not '99.9% on average somewhere' — concrete numbers per service

Compute (VM, K8s nodes)99.95%Multi-AZ deployments. Credit: 10× downtime.
Block storage (SSD/NVMe)99.99%3-way replication. Auto-failover ≤ 60 sec.
Object storage (S3)99.99%11 nines of durability. Multi-region optional.
Managed Kubernetes99.95%Control plane + node auto-recovery. P1 response ≤ 5 min.
Network (intra-region)99.99%Any two AZs ≤ 1 ms p99. Bandwidth guaranteed.
Public ingress (LB L7)99.95%Multi-AZ. TLS termination. WAF optional.
status · live from monitoring

What our monitoring is showing right now

Compute · MSK-1uptime 99.997% · 90 days
Compute · MSK-2uptime 99.999% · 90 days
Compute · SPB-1uptime 99.991% · 90 days
Object Storage (S3)uptime 100% · 90 days
Managed Kubernetesuptime 99.998% · 90 days
GPU pool · EKB-1uptime 99.94% · 1 degradation on Oct 22
Public networkuptime 99.999% · 90 days

Want a detailed infrastructure briefing under NDA?

Network architecture diagrams, storage topology, certificate copies, pentest reports. Sent after NDA signing — usually within the hour.

Request NDA package →
Infrastructure — h3llo cloud