Sovereign edge infrastructure, engineered per deployment. The GCS Kernel inside is a black box — proprietary, undisclosed, and the reason the node is leased and never sold. Every Hummingbird is built for one facility. The data inside never leaves.
Every Hummingbird node is engineered for a specific deployment — scoped, commissioned, and certified for the licensee's exact data environment. The GCS Kernel inside is proprietary and undisclosed. The node stays on our balance sheet. You operate it. We own it. That arrangement is not incidental — it is the IP protection mechanism.
The GCS Kernel is the core IP. Its internal architecture is not disclosed — to licensees, to end clients, or in this document. What it enables: zero-copy data transfer between the GPU and neuromorphic layers, bypassing the CPU bottleneck entirely. This is why Hummingbird achieves sub-60μs detection at 60W. GCS is not a chip reseller. The current silicon is the best available for this application at this time. When GCS proprietary silicon matures, it replaces it.
The Hummingbird architecture was validated against three classes of high-volume, high-velocity public data — each chosen for a specific property, not a specific application. The data streams were the instruments. The architecture is the product.
Founded 1978. Commercial real estate and insolvency boutique. Approximately 30 years of un-indexed commercial Agreements of Purchase and Sale — the Iron Vault corpus for the legal vertical. The Hummingbird node processes this archive locally. Client data is PII-redacted via Microsoft Presidio before entering the AI layer. Solicitor-client privilege preserved by architecture — no data touches foreign infrastructure.
Hardware-agnostic. Every Hummingbird deployment is scoped and configured for the specific licensee environment. These specifications reflect the current reference architecture. As GCS's proprietary silicon programme matures, these will evolve. The kernel performance targets do not change.
| Component | Current Implementation | Notes |
|---|---|---|
| GPU Layer | NVIDIA Jetson AGX Orin | 32GB or 64GB unified memory. Architecture-agnostic kernel — not coupled to NVIDIA silicon. |
| Neuromorphic Layer | BrainChip Akida AKD1000 | Current reference NPU. SNN inference. GCS is not a BrainChip reseller — Akida is the best available neuromorphic silicon for this application at this time. |
| GCS Kernel | Proprietary | Architecture undisclosed. Black box by design. NRC security validation in progress: TB Digital Transformation Policy, PIPEDA, Protected B and Higher. |
| Proprietary Silicon | In Development | GCS is developing proprietary silicon with embedded LLM architecture. Replaces the current GPU+NPU reference implementation. Kernel transfers. |
| Storage | NVMe — encrypted | Hardware-encrypted. Capacity and configuration specified per deployment. ChromaDB Iron Vault partition sized to domain. |
| Network | Physical ethernet only | No WiFi. No Bluetooth. Air-gapped by hardware — not policy. |
| Power | Max 60W | Standard wall outlet. No data centre infrastructure required. |
| Form Factor | Configured per deployment | 1U rack, desktop tower, ruggedised tactical enclosure. Specified at engagement scoping. |
| Latency | <60μs | Validated across three data stream classes. SNN detection layer. |
| Ownership | GCS — lease only | Never sold. All Hummingbird nodes remain GCS property. Kernel updates and silicon transitions are GCS's responsibility throughout the lease term. |
Every Hummingbird deployment is a bespoke engineering engagement. GCS scopes, architects, commissions, and certifies each node for its specific deployment. The node stays on our balance sheet. The kernel stays in our custody. The lease is the IP enforcement mechanism.
This is not a contact form. It is a demonstration of the GCS privacy posture. Your enquiry is encrypted client-side using RSA-OAEP public key cryptography before it goes anywhere. Only GCS can decrypt it. No plaintext message touches a server.