Parinita AI EdgePARINITA AI EDGENINE PLANES. ONE FABRIC.

Architecture

Nine planes. One fabric. Replicated at every AI Edge Hub.

A Hub is not a generic colocation cage. It is a fixed, repeatable stack of nine specialized planes, so workloads land on hardware chosen for them rather than whatever happens to be free. Identical builds mean identical unit economics and one operational playbook across all 101 sites.

9
Compute planes

Per Hub, identical build

5,858
Servers per Hub

Plane 9 is network fabric

16.1M
Seats at capacity

Across the full fabric

$175
Per seat / month

Hybrid seat, per location

The nine planes

Each workload class gets its own plane

Operated, serviced and refreshed independently.

  1. P1AI Inference
  2. P2Heterogeneous AI Compute
  3. P3Sovereign CPU Compute
  4. P4Vector / RAG / AlmanacAMD EPYC 9965
  5. P5Distributed NVMe Storage
  6. P6Media / Streaming
  7. P7Real-Time Edge AI
  8. P8Orchestration / ControlAMD EPYC 9965
  9. P9Sovereign AI NetworkingCisco + Arista + NVIDIA BlueField-3 + Palo Alto
Nine stacked infrastructure planes forming the Parinita AI Edge fabric

Underwriting rationale

Why the plane model matters

Placement is deterministic

The orchestration plane routes each workload to the plane built for it, so utilization is a function of demand mix rather than luck of scheduling.

Capex is a known quantity

Because every Hub is the same build, the bill of materials is fixed and tiering changes only unit count — not architecture.

Latency is structural, not optimized

Compute sits inside the metro it serves. Distance to the user is a property of the network map, not something to tune later.

One playbook, 101 times

Identical installs mean commissioning, staffing and vendor support scale linearly instead of being negotiated per site.

Sourcing

Planned and quoted vendor roles

  • AMD

    Instinct MI350P / MI350X accelerators, EPYC Turin and 9965 CPUs

  • NVIDIA

    B300 accelerators, RTX PRO 6000, BlueField-3 DPUs

  • Cisco / Arista

    Leaf-spine switching across the sovereign networking fabric

  • Palo Alto Networks

    Perimeter enforcement and segmentation at every Hub

Vendor references describe planned, quoted or negotiated sourcing roles. They do not imply partnership, exclusivity or endorsement. The Manager retains substitution rights, so the deployed configuration may differ.

Detailed per-plane server counts and the full bill of materials are available to verified investors in the data room.