Private Paths to Every Cloud, Data Center, and GPU Cluster

Explore Our 4 Dedicated Cloud Connectivity & AI Fabric Practice Areas

Because enterprise buyers evaluate private cloud on ramps, bare metal AI GPU clusters, zero egress storage, and metro data center interconnect as distinct layers of the architecture, Obsidian X maintains four specialized engineering guides under our Cloud Connectivity hub:

1. Direct Cloud Interconnect, Megaport MCR & Virtual Edge: Private Layer 2 and Layer 3 on ramps to AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect, and Oracle FastConnect, plus Megaport Cloud Router (MCR) up to 100 Gbps and Megaport Virtual Edge (MVE) for SD-WAN branch aggregation.

2. Bare Metal GPU Cloud & Latitude.sh AI Compute Fabric: Single tenant bare metal CPU servers and NVIDIA Blackwell B300, H200, and H100 GPU clusters across 24 data centers in 26 countries via Latitude.sh on the Megaport private fabric, solving 6 to 8 month OEM server supply chain backlogs.

3. Zero Egress Cloud Storage, VAST Data NVMe & NAT Gateway: Fabric native Standard S3 Object Storage (Wasabi with Object Lock immutability) and High Performance NVMe Object, Block, and File Storage (VAST Data up to 1 Tbps with NFS, RDMA, and GPUDirect) with zero egress, zero API, and zero retrieval fees, plus Megaport NAT Gateway cutting outbound NAT bills by 70 percent or more.

4. Data Center Interconnect (DCI), Metro Wavelengths & Texas On Ramps: High capacity 1G, 10G, 100G, and 400G Layer 2 interconnects benchmarked across Megaport, Equinix Fabric, PacketFabric, and carrier wavelengths across 634 global and 266 US facilities, including 16 in Dallas Fort Worth and 8 in Houston.

Direct Connect, ExpressRoute & Megaport NaaSBare Metal GPU, Latitude.sh & AI Cloud FabricZero Egress Storage, NVMe AI & NAT GatewayData Center Interconnect (DCI) & Texas On Ramps

Why Legacy Telco Circuits and Public Internet Fail Modern Cloud and AI Workloads

Enterprise cloud architectures have outgrown the two connectivity models most IT teams still rely on. Traditional carrier point to point and MPLS circuits require 30 to 120 days to provision, lock buyers into rigid multi year capacity commitments, and charge punitive early termination fees when workloads move. On the other side, routing production database replication, storage backups, or AI inference traffic over public internet VPN tunnels exposes critical operations to BGP hijacks, packet loss, and unpredictable jitter.

Financial exposure compounds the operational risk. Hyperscalers charge steep data egress fees every time data leaves their environment over the public internet, turning multi cloud synchronization and AI model training into runaway monthly line items. When an enterprise connects through private cloud on ramps such as AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect, and Oracle FastConnect, hyperscalers apply discounted private data transfer rates while delivering deterministic throughput and lower latency.

Through a software defined Network as a Service (NaaS) fabric, physical cross connects are installed once inside your colocation facility or regional point of presence, with the Letter of Authorization generated in minutes and the facility cross connect completing in hours to days. From that moment forward, your network team provisions Virtual Cross Connects (VXCs), adjusts bandwidth up or down, and turns up new cloud destinations on demand in under 60 seconds without waiting on another carrier truck roll.

The Complete Megaport Software Defined Fabric: 6 Core Building Blocks

Obsidian X scopes, benchmarks, and manages the entire Megaport software defined network ecosystem alongside competing fabrics such as Equinix Fabric, PacketFabric, and Console Connect. Rather than piecing together disparate carrier contracts, engineering teams gain a unified global fabric spanning 1,000+ enabled locations worldwide (634 locations across 266 US facilities, including 16 in Dallas Fort Worth and 8 in Houston, in the October 14, 2026 public locations API), 300+ service providers, and 318+ cloud on ramps across 26 countries.

1. Physical Ports (1 Gbps, 10 Gbps, 100 Gbps, and 400 Gbps): High capacity physical Ethernet interfaces installed in Megaport enabled data centers. Once the physical cross connect is active (typically hours to days depending on the facility), a single port can host dozens of independent virtual circuits to cloud providers, partner data centers, and internet exchanges.

2. Virtual Cross Connects (VXCs): Private point to point Layer 2 Ethernet circuits provisioned via portal, API, or Terraform in under 60 seconds once your port is live. Bandwidth scales elastically from 1 Mbps to 100 Gbps on flexible month to month or 12, 24, and 36 month terms.

3. Megaport Cloud Router (MCR): Managed virtual Layer 3 BGP routing instances deployed in key metro zones without buying or racking physical routers. MCR enables direct cloud to cloud routing between AWS, Azure, Google Cloud, and Oracle at speeds from 1 Gbps to 100 Gbps, keeping multi cloud traffic off your headquarters WAN.

4. Megaport Virtual Edge (MVE): An on demand NFV hosting platform (available in 2, 4, 8, and 12 vCPU tiers) that runs virtual SD-WAN gateways and next generation firewalls from Cisco, Fortinet, Versa, Aruba SilverPeak, VMware VeloCloud, Palo Alto Networks, Meraki, Aviatrix, and Arista directly on Megaport global points of presence. Branch offices connect over local internet or SD-WAN within roughly 20 milliseconds of an MVE hub and hop immediately onto Megaport's private backbone.

5. Megaport Internet and Managed NAT Gateway: On demand, DDoS protected Dedicated Internet Access (DIA) scalable from 20 Mbps to 10 Gbps in 1 Mbps increments across two diversity zones delivered over your existing Megaport Port or MVE, paired with a carrier grade Managed NAT Gateway (RFC 6888 and RFC 4787 compliant) that reduces cloud egress and outbound NAT costs by 70 percent or more (with typical workload savings ranging from 47 percent to 93 percent) while eliminating RFC 1918 IPv4 address exhaustion.

6. Global Internet Exchange (MegaIX and AMS IX): Direct Layer 2 peering fabric access across major domestic and international peering points, including DFW, LAX, SFO, ORD, NYC, and MIA, cutting transit hops to SaaS platforms and content networks.

Bare Metal CPU, On Demand NVIDIA GPU Compute, and Zero Egress AI Storage

Following Megaport's acquisition of Latitude.sh (announced November 17, 2025) and the 2026 launch of Megaport Storage, the same software defined fabric that routes your cloud traffic now provisions dedicated, single tenant bare metal CPU and NVIDIA GPU compute plus zero egress storage connected directly to your private Layer 2 and Layer 3 network. For engineering teams evaluating colocation versus public cloud for AI and automation workloads, bare metal on a global NaaS fabric eliminates the noisy neighbor contention and virtualization tax of hyperscaler instances while bridging 6 to 8 month OEM server supply chain backlogs.

On demand single tenant bare metal CPU instances spin up immediately across 24 data centers in 26 countries, while dedicated multi node NVIDIA GPU clusters (including NVIDIA Blackwell B300, Hopper H200, H100, and L40S architectures) are scoped per workload with Megaport engineering. In addition, Megaport's Layer 2 AI Inference Cloud (marked Planned on Slide 22 of the October 2026 partner architecture brief) is designed to extend low latency inference endpoints directly onto the private fabric. Because compute sits directly on the private backbone, training datasets and inference payloads move between your data center, public cloud VPCs, and GPU nodes over private VXCs.

To feed those clusters without incurring hyperscaler data transfer penalties, Megaport Storage provides two fabric native tiers with zero egress fees, zero API request fees, and zero retrieval fees plus free in country private networking: Standard S3 Compatible Object Storage (powered by Wasabi, launched July 21, 2026, with Object Lock immutability for backups and cyber resilience) and High Performance NVMe Object, Block, and File Storage (powered by VAST Data, launched June 4, 2026, delivering up to 1 Tbps sustained throughput with NFS, NFS + RDMA + GPUDirect, and SMB).

How Megaport Compares to Carrier MPLS, Equinix Fabric, and Direct Peering

No single interconnect provider wins every route or every facility. A company with racks exclusively inside Equinix IBX buildings may evaluate Equinix Fabric alongside Megaport, while an enterprise running steady 100 Gbps point to point metro transport between two owned facilities may compare dark fiber or carrier wavelengths. Obsidian X models the exact port, VXC, cross connect, and cloud egress math across all available providers before you sign.

Real World Reference Architectures We Design and Source

Hybrid Multicloud Core (AWS + Azure + Colocation): Many mid market enterprises run core ERP or regulated databases in Dallas Fort Worth colocation or cloud migration environments while running analytics in Google BigQuery and application tiers in AWS and Azure. Rather than backhauling traffic across public internet VPNs, we deploy an MCR between the clouds and a single Megaport Port in your data center, cutting inter cloud latency and qualifying every flow for private egress tiers.

SD-WAN Branch Aggregation via Megaport Virtual Edge (MVE): For distributed operators running 10 to 200+ locations on managed networks or SD-WAN, hairpinning cloud traffic through a corporate data center adds unnecessary latency. Deploying your existing Cisco, Fortinet, Versa, Aruba, VMware, or Palo Alto virtual edge on MVE lets branch sites transition from local broadband onto a private Layer 2 highway within milliseconds of the branch.

AI Training and Zero Egress Storage Pipeline: When engineering teams train models or run inference on sensitive enterprise datasets, keeping data out of public cloud lock in is critical. We architect private Layer 2 loops connecting enterprise colocation cages, Wasabi or VAST Data storage arrays, and Latitude.sh bare metal GPU clusters over dedicated 10 Gbps or 100 Gbps VXCs.