Three configurations built for different workloads and budgets. Every system ships pre-configured, burn-tested, and ready to run.
Available by special order with confirmed deposit. GPU sourcing timeline: 2-4 weeks depending on market availability.
InquireAvailability pending used H200 market development. V4-Flash requires 282GB VRAM - dual H200 141GB GPUs are the minimum hardware to run it at full quality. This tier does not currently ship.
Join Waitlist| Specification | NVIDIA H100 80GB | NVIDIA H200 141GB |
|---|---|---|
| VRAM | 80GB HBM3 | 141GB HBM3e |
| Memory Bandwidth | 3.35 TB/s | 4.8 TB/s |
| FP16 Performance | 989 TFLOPS | 989 TFLOPS |
| Interconnect | PCIe Gen 5 | PCIe Gen 5 |
| Combined VRAM (2x) | 160GB | 282GB |
| 70B Model Fit | Yes (FP16) | Yes (FP16, fastest) |
| V4-Flash (284B, full quality) | Quantized only | Yes (282GB required) |
| Island Mountain Tier | Summit Base ($75-85K) | Summit Pinnacle ($350-400K) |
Power Requirement: All systems require a 208V/30A dedicated power circuit (NEMA L6-30R). This is standard in server rooms and data closets. Residential or office installation by a licensed electrician typically costs $500-$2,000. The PSU operates at 200-240V only.
The A100 and H100 GPUs in Island Mountain systems are not the same hardware you find in gaming PCs or consumer AI builds. The performance difference matters, but the documentation chain matters more for regulated buyers.
Every GPU ships with full procurement documentation: purchase receipts from authorized NVIDIA channels, serial number registry, and RMA history. If your compliance team, external auditor, or federal contracting officer needs to verify the provenance of the hardware processing your sensitive data, that chain exists and is complete.
Consumer GPUs - RTX 4090s, 5090s, and their equivalents - deliver strong inference performance per dollar. For personal projects and unregulated workloads, they are the right choice. But they are purchased through retail channels that do not provide supply chain documentation. They cannot be independently verified for ITAR, DFARS, or CMMC supply chain requirements. And when they fail, the consumer RMA process does not maintain the documentation continuity that regulated environments require.
This is not a performance argument. It is a compliance architecture argument. The GPU provenance chain is part of the system, not an afterthought.
As newer models demand more VRAM or faster memory bandwidth, Island Mountain offers a GPU upgrade path. This is a hardware swap - your existing GPUs are removed, credited at current secondary market value, and replaced with the next-generation cards.
This is not a firmware update or a software toggle. It's a physical component replacement that requires shipping the system or scheduling an on-site visit. Upgrade pricing depends on the current market price of the replacement GPUs minus your trade-in credit.
We disclose this upfront because "upgrade-ready" language in this industry often hides the reality: getting to the next tier of performance means buying new silicon. We'd rather you know that going in.
Yes. Your existing H100 GPUs are credited at current secondary market value and replaced with higher-performance GPUs if available. This is a physical hardware swap, not a software update. The chassis, CPU, RAM, and storage remain the same. Upgrade pricing depends on the replacement GPU market price minus your trade-in credit. Turnaround is typically 2-4 weeks from order.
The Summit Base tier (160GB total VRAM) runs any model that fits within 160GB at FP16 precision, including all 70B-parameter models and Mixtral 8x22B. It can also run DeepSeek V4-Flash in quantized form. It cannot run V4-Flash at full FP16 quality (requires 282GB VRAM). The Summit Pinnacle tier with H200 GPUs is the only tier that runs V4-Flash at full quality.
Summit Base and Summit Ridge tiers use verified refurbished datacenter GPUs sourced from enterprise resellers with documented provenance. This keeps pricing at $75,000-$85,000 for Summit Base and $150,000-$160,000 for Summit Ridge instead of new GPU costs. Every GPU is individually tested before assembly and then undergoes 72 hours of continuous burn-in testing. GPUs showing any instability are rejected.
Personal and small-team AI servers for professionals who handle sensitive data daily.
The Summit Series serves organizations with enterprise compliance requirements and institutional budgets. But sovereign AI is not only an institutional need.
Attorneys handling privileged client documents at home. Physicians reviewing patient records on personal devices. Emergency managers processing sensitive incident data from a home office. Researchers running models on proprietary datasets outside the lab.
The Landfall Series brings Island Mountain's build quality, burn-testing, and pre-configured software stack to individual professionals and small teams at a fraction of the institutional price point.
Anticipated: $4,500 - $7,500
Personal AI server for individual professionals. Consumer-tier GPU, full software stack, 72-hour burn-in, direct builder support.
Anticipated: $9,000 - $14,000
Small team AI server for 3-8 concurrent users. Dual consumer GPUs, same burn-testing and configuration standard.
Hardware sourcing and margin analysis is underway. If you want to be notified when the Landfall Series is available, let us know.
Tell us what you're running, how many users need access, and what models matter most. We'll spec the right system.
Or call directly: 1-801-609-1130