MegaCore
Home / Products / NVIDIA AI Infrastructure
NVIDIA AI Infrastructure · The Balkans

NVIDIA AI hardware for the Balkans. The right channel for your configuration.

The full NVIDIA AI portfolio — workstation GPUs, server-integrated accelerators, datacenter-class systems, and supporting NVIDIA networking fabric — sourced through authorized distribution channels and delivered DDP to your data center in the Balkans. New and sealed, with full provenance documentation, warranty position verified per shipment, and export-control workflow handled in-house.

New, sealed, full provenance trail
Lead time in weeks
DDP across the Balkans
Request a quote How we source
Sourcing path
1
Source matching
Configuration-appropriate authorized channel — through Gamma Group's 15-year distribution network
2
Verified procurement
Provenance documentation, ECCN classification, warranty position confirmed before contract
3
DDP delivery
Multi-jurisdiction customs, duties, VAT pre-financing — all in-house, all in your contract
Coverage
Serbia, BiH, North Macedonia (other countries on request)
Terms
DDP · EUR or RSD · 30% prepayment standard
Quote response
Within 24 hours
Hardware status
New, sealed, original packaging, with serial numbers documented
NEWLY AVAILABLE

Next-generation rack-scale AI, now sourceable

Two flagship NVIDIA rack-scale platforms are now open for order through MegaCore for Balkan deployments — sourced via Gamma Group's procurement channels, delivered DDP with full provenance and export documentation.

Grace Blackwell Ultra

NVIDIA GB300 NVL72

On request

A single liquid-cooled rack that fuses 72 Blackwell Ultra GPUs and 36 Grace CPUs into one NVLink domain — built for the largest reasoning-model inference and trillion-parameter training. NVIDIA rates it at roughly 1.1 exaFLOPS of FP4 inference and about 1.5× the AI throughput of GB200 NVL72.

Compute
72× Blackwell Ultra + 36× Grace
Memory
~20 TB HBM3e · ~40 TB total
Interconnect
5th-gen NVLink · 130 TB/s
Networking
ConnectX-8 · Quantum-X800
Check availability →
Rubin platform · 2H 2026

NVIDIA Vera Rubin NVL72

Pre-order

NVIDIA's next-generation rack, pairing 72 Rubin GPUs on HBM4 memory with 36 Vera CPUs. Targets roughly 3.6 exaFLOPS FP4 inference and 1.2 exaFLOPS FP8 training per rack over 6th-generation NVLink. Reserve allocation now for delivery in the second half of 2026.

Compute
72× Rubin + 36× Vera
Memory
HBM4 (next-gen)
Interconnect
6th-gen NVLink
Peak performance
~3.6 EF FP4 · ~1.2 EF FP8
Check availability →

All figures are NVIDIA's announced specifications and are subject to change and to allocation. NVIDIA also markets the Rubin rack as "Vera Rubin NVL144" (counting GPU dies); "72" refers to GPU packages. These are advanced-compute, export-controlled systems — lead time, licensing and eligibility are confirmed per project before contract.

WHAT WE SUPPLY

NVIDIA AI hardware across four product tiers

From workstation GPUs to rack-scale systems. Availability and lead time depend on the tier, the configuration, and the source channel for each request. Check the page for indicative status, then send a request for a binding quote.

GPU accelerators

8 models
H100 SXM5
Available
80 GB HBM3 · NVLink
Training, large-scale inference
Check availability →
H100 PCIe
Available
80 GB HBM3 · PCIe Gen5
Inference, mixed workloads
Check availability →
H200 SXM5
Available
141 GB HBM3e · NVLink
Training, memory-bound LLM workloads
Check availability →
H200 NVL
Available
141 GB HBM3e · NVLink Bridge
Inference, fine-tuning
Check availability →
B200
Pre-order
192 GB HBM3e · NVLink
Frontier training, next-gen inference
Check availability →
A100 80 GB
Available
80 GB HBM2e · NVLink / PCIe
Established workloads, research
Check availability →
L40S
Available
48 GB GDDR6 · PCIe Gen4
Inference, graphics-AI hybrid, virtualization
Check availability →
RTX PRO 6000 Blackwell
On request
96 GB GDDR7 · PCIe Gen5
Workstation, single-node training
Check availability →

DGX & HGX systems

7 configurations
DGX H100
Available
8× H100 SXM5 · 640 GB HBM3 · 900 GB/s NVLink
Production AI workhorse
Check availability →
DGX H200
Available
8× H200 SXM5 · 1128 GB HBM3e · 900 GB/s NVLink
Memory-intensive training
Check availability →
DGX B200
Pre-order
8× B200 · 1440 GB HBM3e · 5th-gen NVLink
Frontier training
Check availability →
DGX GB200 NVL72
On request
72× B200 + 36× Grace CPU, rack-scale · Rack-scale NVLink
Trillion-parameter training, exascale inference
Check availability →
GB300 NVL72
On request
72× Blackwell Ultra + 36× Grace CPU · rack-scale · 5th-gen NVLink · 130 TB/s
Grace Blackwell Ultra — exascale reasoning inference & trillion-parameter training
Check availability →
Vera Rubin NVL72
Pre-order
72× Rubin GPU + 36× Vera CPU · HBM4 · rack-scale · 6th-gen NVLink
Next-generation flagship — reserve allocation for 2H 2026
Check availability →
HGX H100 / H200 / B200 baseboards
Available / On request
8-GPU OEM-integrated platforms · Supermicro · Dell · HPE · Lenovo
For customers who prefer their own server vendor
Check availability →

Networking fabric

5 options

AI clusters are bottlenecked by the network. We supply the full NVIDIA / Mellanox fabric, not just the GPUs.

ConnectX-7 / ConnectX-8 NICs
Available
400 / 800 Gb/s · InfiniBand & Ethernet
Cluster interconnect
Check availability →
Quantum-2 InfiniBand switches
Available
NDR 400 Gb/s · Up to 64-port leaf/spine
Production AI fabric
Check availability →
Quantum-3 InfiniBand switches
On request
XDR 800 Gb/s · Next-gen fabric
For B200/GB200 deployments
Check availability →
Spectrum-X Ethernet switches
Available
800 GbE AI-optimized · Ethernet
Alternative to IB for some workloads
Check availability →
BlueField-3 DPU
Available
400 Gb/s · Integrated security / storage offload
Multi-tenant AI cloud, GPUaaS
Check availability →

Reference configurations

3 starting points

Three starting points. Real configurations get tailored on a sizing call.

Config 01

Inference cluster · entry

8× L40S in 2× 4U servers · 25 GbE fabric · suitable for production inference up to ~10k RPS depending on model size

Lead time 6–8 weeks
Discuss this configuration
Config 02

Training cluster · mid-size

1× DGX H200 + 4× HGX H200 nodes · NDR InfiniBand fabric (Quantum-2) · suitable for 70B–200B parameter fine-tuning and from-scratch training of mid-size models

Lead time 6–10 weeks
Discuss this configuration
Config 03

Production SuperPOD-class deployment

32–256× H200 or B200 GPUs · full Quantum-2/-3 IB fabric · NVIDIA reference architecture · sized to NVIDIA SuperPOD specifications

Lead time By quotation
Discuss this configuration
HOW WE SOURCE

Three principles for NVIDIA supply across the Balkans

NVIDIA's product portfolio spans from consumer-grade GPUs to datacenter-class AI accelerators, and each tier has its own distribution structure, allocation pattern, and warranty model. We don't pretend that one supply story fits all of it. Here's how we actually work.

1

Channel matching across NVIDIA's tiers

NVIDIA hardware moves through different channels depending on what it is and who's buying it. We work with multiple distribution partners — through Gamma Group's 15-year regional network — and we match each request to the appropriate path:

Workstation and edge tier
RTX PRO 6000 Blackwell, RTX 6000 Ada, RTX 5000 Ada, A40 — readily available. Standard procurement, no special structure required, lead times measured in weeks. This is where the supply story is simplest.
Server-integrated accelerators
L40S, L40, A40 inside Dell, HPE, Supermicro, Lenovo platforms — sourced as part of the integrated server, through authorized server vendors. The accelerator inherits the server's warranty and support contract structure.
Datacenter accelerators & DGX systems
A100, H100, H200, B-series — subject to NVIDIA's regional allocation, US export controls (ECCN 3A090 for the silicon, additional classifications for complete systems), and per-deployment registration. We don't quote these generically. Every request goes through configuration discussion, source path, registration logistics, warranty position, and realistic timeline before any commercial proposal.

Across all tiers, units ship new and sealed in original NVIDIA or OEM packaging, with manufacturer serial numbers intact. Provenance documentation — distributor of record, country of dispatch, serial numbers, export-control classification — is part of the standard contract package.

2

Warranty — verified per shipment, not assumed

NVIDIA warranty terms vary significantly across the product portfolio. We verify the applicable warranty path before contract signature, so the warranty position is explicit upfront — not discovered after delivery.

Workstation GPUs
RTX PRO 6000 Blackwell, RTX 6000 Ada, RTX 5000 Ada — NVIDIA's standard warranty applies on global terms for authentic, channel-sourced units. Validation through NVIDIA's standard channels is straightforward.
Server-integrated accelerators
L40S, L40 inside server platforms — warranty inherits from the server vendor's support contract (Dell ProSupport, HPE Foundation Care, Supermicro Service). We sell the appropriate regional contract alongside the hardware.
Datacenter accelerators & DGX systems
NVIDIA Enterprise Support is structured per deployment, registered to a specific country and entity, and tied to the source channel through which the unit was originally sold. For these products, we document the applicable warranty terms — including any limitations relative to standard regional support — before the commercial agreement. If full NVIDIA Enterprise Support cannot be secured for a specific configuration, we state that explicitly and propose alternatives.
Specialty vendors
xFusion, Huawei, Kioxia, Dapu — manufacturer warranty per their standard terms; we facilitate registration and support coordination.

For RMA across all tiers, MegaCore is your single point of contact. You raise the case with us; we coordinate with NVIDIA or the relevant OEM, ship replacements where needed, and handle re-import logistics.

3

DDP across the Balkans

We deliver on Delivered Duty Paid terms across Serbia, Bosnia and Herzegovina, and North Macedonia. The contract price covers the hardware, international transport, insurance, multi-jurisdiction customs clearance, import duties, VAT pre-financing, and the full export-control documentation chain.

For NVIDIA datacenter accelerators (ECCN 3A090) and DGX systems, the export-control workflow is more involved than for workstation tier. We complete end-user statements, end-use declarations, and destination country verification with the source distributor before shipment. The destination country is named on documentation from the start — we do not structure shipments through declared intermediate destinations to obscure the actual deployment country.

You sign a contract. We deliver to your loading bay. Your procurement and compliance teams receive the audit trail they need.

MegaCore is an independent supplier registered in Belgrade, Serbia. We source through authorized distribution channels in their respective regions. NVIDIA, DGX, HGX, ConnectX, Quantum, Spectrum, BlueField and related marks are trademarks of NVIDIA Corporation, used here for nominative descriptive purposes only.

FAQ

Questions we hear most often

Direct answers on sourcing, warranty, payment and export-control.

Sourcing & supply

Where does the hardware physically come from?
Source channel depends on the product category. For tier-1 enterprise vendors (Dell, HPE, NVIDIA, AMD), we source through authorized distribution — European, Asian, or direct from the manufacturer's regional partners — selecting the path that delivers the right configuration on the right timeline. For specialty vendors with limited Balkan distribution (xFusion, Huawei, Kioxia, Dapu), we source directly through Gamma Group's 15-year procurement network in Asia. Every shipment ships with full provenance documentation as part of the standard contract: distributor of record, country of dispatch, manufacturer serial numbers, and applicable export-control classification. Not optional, not on request, not under NDA.
Is the hardware new?
Yes. New, sealed, in original manufacturer or OEM packaging, with manufacturer serial numbers intact. We do not deal in refurbished, pulled-from-cluster, re-flashed, or re-marked hardware. If you have specific verification requirements — independent inspection, photo verification of seals before shipment, or serial validation through your own vendor channels — we accommodate them as part of the contract.
Can I verify the serial number is genuine?
Yes. Manufacturer serial numbers are provided ahead of delivery for your independent validation. Validation procedures vary by vendor and product class — for some products (most workstation and server hardware) you can verify directly with the manufacturer; for others (NVIDIA datacenter accelerators, DGX systems) verification typically goes through the manufacturer's authorized channel partner. We document the appropriate verification path for each shipment before contract signature.
Why is this faster than the official European channel?
Two reasons, both depending on the product: For specialty vendors (xFusion, Huawei, Kioxia, Dapu): there's no European channel to compete with. These manufacturers have limited or no authorized Balkan distribution, so direct procurement through our Asian network is often the only practical supply path — and the fastest by definition. For tier-1 enterprise products: lead times and allocation cycles vary significantly between regional distribution channels. European authorized distribution prioritizes volume customers and quarterly allocation cycles; Asian authorized distribution sometimes has different inventory positions for specific configurations. We coordinate across both, plus direct manufacturer paths where available, and select what works for your project. If your established distributor can deliver the configuration you need on your timeline, use them. We engage where the standard channel reaches its structural limits — not as a quota-bypass, but as a complementary supply path.

Warranty & support

Who honors the warranty?
Manufacturer warranty applies, with terms varying by product category: • Workstation and edge GPUs (RTX 6000 Ada, L40S, A40): standard manufacturer warranty applies on global terms for authentic, channel-sourced units. • Enterprise servers (Dell PowerEdge, HPE ProLiant, Supermicro): warranty and support are tied to the support contract sold alongside the hardware (Dell ProSupport, HPE Foundation Care, equivalent). We sell the appropriate contract for your delivery region. • NVIDIA datacenter accelerators (A100, H100, H200, B-series): warranty and NVIDIA Enterprise Support availability depend on source channel and registration path. We verify and document the applicable warranty terms per shipment, before contract signature, so there are no surprises post-delivery. • Specialty vendors (xFusion, Huawei, Kioxia, Dapu): manufacturer warranty per their standard terms; we facilitate registration and support coordination. Across all categories, MegaCore acts as your single point of contact for RMA logistics — you raise the case with us, we manage the process with the manufacturer and handle re-import where needed.
What about DGX systems specifically?
DGX warranty and NVIDIA Enterprise Support are structured differently from card-level warranty — they're tied to a regional support contract that NVIDIA registers per deployment. This is more involved than standard procurement and requires explicit alignment between source channel, registration path, and deployment country. We don't quote DGX systems generically. Every DGX inquiry goes through a sizing call where we discuss support contract structure, registration logistics, and realistic timeline before any commercial terms are proposed.

Commercial & legal

What are typical payment terms?
Standard terms: 30% on contract signature, 70% before shipment. Contracts in EUR or RSD. Other terms — milestone-based payment for multi-shipment deployments, L/C or escrow for first-time large orders, extended terms for established customers — are negotiable. We invoice from MegaCore with full Serbian VAT documentation.
What's the typical lead time?
Lead time depends on product category and configuration: • Specialty vendors (xFusion, Huawei, Kioxia, Dapu): 4–8 weeks DDP for standard configurations. • Tier-1 enterprise servers (Dell, HPE, Supermicro): 6–10 weeks DDP, depending on configuration and source channel. • NVIDIA workstation/edge tier (RTX 6000 Ada, L40S): 6–10 weeks DDP for in-stock SKUs. • NVIDIA datacenter tier (A100, H100, H200): 12–20 weeks, subject to allocation and configuration. We don't quote a generic timeline — every request goes through a sizing call. • DGX systems and rack-scale deployments: by quotation only. We confirm the realistic timeline at the feasibility stage, before contract signature. Zero overpromising — if a configuration cannot meet your timeline through any of our channels, we say so upfront.
Do you handle US export-control compliance (BIS / ECCN)?
Yes. Several products in our portfolio are subject to US export controls — primarily NVIDIA datacenter accelerators (A100, H100, H200, B-series), which fall under ECCN 3A090. For these products we complete the full export-control documentation chain: end-user statements, end-use declarations, destination country verification, and re-export documentation where applicable. The Balkan destination country is named on the documentation from the start — there is no "declared destination Asia, actual destination Balkans" structure. Import procedures into Serbia, BiH, and North Macedonia are handled by our customs team. We provide the audit trail your procurement and compliance teams need.
Can I get an invoice on a non-Serbian Balkan entity?
Yes. MegaCore invoices in EUR or RSD to registered legal entities in Serbia, Bosnia and Herzegovina, and North Macedonia, with VAT handling appropriate to the buyer's jurisdiction and the cross-border transaction structure. Other Balkan countries on request — let us know the destination during the quote stage.
Is MegaCore an authorized NVIDIA partner?
No. MegaCore is an independent supplier. We source through authorized distribution channels and direct manufacturer relationships, but we don't represent ourselves as part of any vendor's official partner program. We say this clearly because being upfront about the supply model is part of how we work. If your procurement policy requires an authorized partner relationship, we're not the right fit — use your established distributor. If you need a configuration or timeline your authorized partner cannot deliver, that's the gap we exist to fill.
CONTACT

Tell us what you need

Send the model, the quantity, and the delivery country. We come back within 24 hours with availability, lead time, and indicative pricing.