Direct answers.
No hedging.

Five things, all built on the same vertically integrated stack:

• Modular data centres. Bare-metal capacity from a Tier III pod, sold by the rack or by the module.
• Turnkey systems. Pre-configured compute, storage, and networking shipped as a single, deployable unit.
• Private sovereign cloud. Orchestrated GPU, HPC and storage on top of that capacity, with governance built in.
• Cybersecurity. Quantum-safe communications and data-protection layers from the foundation up.
• Utility. The DCIM and operating layer that customers and Navon ops both work through.

Most customers start with one and grow into the others.

Three differences that matter:

• Physical residency. Your data sits in a facility we own and operate, in your jurisdiction. Not a regional zone of a US hyperscaler.
• Energy pedigree. Power is sourced from local renewable baseload — geothermal, hydro — not bought as offsets against a coal grid.
• Right-sized economics. Modular capacity priced for growth-market enterprises, not a one-size hyperscaler price book.

For workloads where sovereignty, cost, or proximity matters, the math changes. For workloads that don't care, hyperscalers remain a fine choice.

Hells Gate goes live in Q3 2026 with the first 400 kW Tier III module. Customers signed before commissioning are placed first; the typical onboarding cycle from contract to live workload is 6–10 weeks after the facility opens.

If you have a workload that's date-sensitive, talk to us early — we can sequence module commissioning around committed demand.

• Sovereign AI — training, fine-tuning, inference where data jurisdiction is non-negotiable.
• HPC simulation — climate, genomics, CFD, financial risk on dedicated GPU/CPU clusters.
• Government & public-sector platforms — digital ID, health informatics, sector platforms with compliance baked in.
• Enterprise sovereign cloud — banking, telecom, energy workloads that can't sit in a foreign jurisdiction.

If your workload is latency-insensitive, low-density, and has no sovereignty requirement, a hyperscaler is probably cheaper.

Yes. Navon's GPU pods are Ollama-ready for self-hosted open weights — Llama, Mistral, Qwen, DeepSeek — with vLLM, TGI, and TensorRT-LLM available for production serving.

An OpenAI-compatible endpoint sits in front, so SDKs and tools you've already integrated — LangChain, LiteLLM, OpenRouter, LlamaIndex, Cursor — work without code changes. Bring your stack; we provide what's underneath.

Routing platforms are software gateways over third-party model APIs. They aggregate access; they don't own the inference. The compute, the model weights, and the request traffic all live with someone else — usually a hyperscaler outside the jurisdiction.

Navon is the silicon layer. We own the racks, the cooling, and the geothermal baseload. When a model runs on Navon, the weights, the prompts, and the outputs stay in country.

Routing platforms are useful — and many of them will eventually run on top of Navon.

By physics first, contracts second:

• The hardware sits in a facility we own and operate inside your jurisdiction. No cross-border replication unless you ask for it.
• Encryption keys are held in-country. Sovereign HSM by default.
• Compliance is mapped to local statute (e.g. Kenya DPA 2019) and aligned with GDPR where useful, not bolted on as a checkbox.
• Operators on the ground are local. Audits happen in person.

"Residency by physical fact, not policy promise" is the test we hold ourselves to.

Three commercial models, depending on how you want to consume:

• Co-location — per-rack pricing, monthly, with power and cross-connects bundled.
• Bare metal & managed — dedicated GPU/CPU clusters with our orchestration layer on top, billed monthly.
• Sovereign cloud — consumption-based for compute, storage, and managed services. Billable in USD or local currency where supported.

We publish indicative ranges to qualified customers under NDA. Public price lists come once we're commercially live.