Sovereign Compute — dedicated single-tenant AI infrastructure for ITAR, CUI & CMMC workloads.
The architectural answer for buyers where 'our cloud provider's policies' is not a sufficient compliance answer. ITAR. CUI. CMMC L2/L3. Pre-disclosure biotech IP. Material non-public information. Federal data residency.
Most AI vendors rent compute. We own it.
OpenAI does not own GPUs. Anthropic does not own GPUs. Mistral, Cohere, Perplexity, Harvey — they all rent compute from a hyperscaler. This is a sensible engineering decision for a company whose product is the model and the user experience around it. It also forecloses true sovereignty as an architecture. When they say "your data stays in this region," the floor of that promise is whatever Azure, AWS, or GCP can guarantee about that region. The customer's contract is with the AI vendor; the physical answer is the cloud provider's.
Backplain runs the inverse shape. Multi-tenant SaaS for buyers who want SaaS economics and SaaS compliance — and dedicated bare metal in a SOC 2 Type II certified Tier 3 colocation facility in San Diego for buyers who specifically need the architecture the SaaS shape cannot provide. Customers who need the SaaS shape get it. Customers who need the sovereign shape get it. Nobody is paying for an architecture they do not need.
For Sovereign Compute customers, the data physically lives where you say it lives. The asset tag is yours. The serial number is on file in your contract. Your auditor can visit the address.
Backplain owns the server. We do not own the data center — we colocate in a recognized, SOC 2 Type II certified Tier 3 facility. That distinction matters: the facility's compliance attestations are real, auditable, and independent of Backplain.
Three locations × three service levels.
Sovereign Compute is priced per dedicated server (one Lambda Scaler) in a 3×3 matrix: three locations crossed with three service levels. Every line below is a starting number; the floor moves up with custom networking, storage tiers, or BYOK demands.
| Location | Bare Metal | + DevOps | + Full Managed Ops |
|---|---|---|---|
| Backplain SD (Tier 3 colo, dedicated server, default) | $9,500/mo | $11,500/mo | $14,500/mo |
| Customer site (your DC, your SCIF) | $11,000/mo + travel | $13,000/mo + travel | $16,500/mo + travel |
| Customer-chosen colo | $10,000/mo + transit | $12,000/mo + transit | $15,500/mo + transit |
Sovereign software tier ($15,000/mo + setup) sits on top and includes 40 hrs/mo of embedded Pro Services. AI Lab tier ($25,000/mo) bundles one server at the Carlsbad Lab plus one ML engineer and one data scientist. AI Lab Fractional ($12,500/mo) gives a half-time named engineer for smaller engagements.
Production in San Diego. R&D in Carlsbad.
Dedicated bare metal in a SOC 2 Type II certified Tier 3 colocation facility, San Diego
Backplain owns and operates a dedicated single-tenant Lambda Scaler server, housed in a SOC 2 Type II certified Tier 3 colocation data center in San Diego. 100,000 sq ft total · 56,342 sq ft of whitespace · 2.9 MW critical IT load · 10 MW generator capacity · N+1 UPS · 13 on-net carriers (Lumen, NTT, Spectrum, Verizon, Zayo, others). Customer Sovereign Compute workloads run here by default.
Facility attestations: SOC 1/2/3 Type II · ISO 20000-1 · ISO 22301 · ISO 27001 · HITRUST CSF · HIPAA · PCI DSS · NIST 800-53 · NIST 800-171 (the CMMC L2 baseline).
The Backplain AI Lab — Carlsbad
The Backplain AI Lab is a leased facility in Carlsbad, California — in the Southern California aerospace, biotech, and defense corridor, neighbors include General Atomics, ViaSat, Illumina, Ionis Pharmaceuticals, and Thermo Fisher.
The Lab connects to the San Diego colocation server over SD-WAN. Clients can lease a desk, a private room, or the full facility. Engagements range from setup-only to fully staffed — our bench includes ML engineers, data scientists, DevOps engineers, and project managers who can work alongside your team or own an entire workstream. For clients requiring a fully air-gapped deployment, the server can be relocated to the Carlsbad Lab on request.
Lambda Scaler — single-tenant, dedicated.
| Component | Specification |
|---|---|
| GPUs | 8× NVIDIA L40S |
| VRAM | 384 GB total |
| CPU | Dual Intel Xeon Gold 6448Y (64 cores) |
| RAM | 1,024 GB DDR5 |
| Storage | 1.92 TB NVMe (system) + 7.68 TB NVMe (data) |
| Networking | Dual 10 Gbps · isolated VLAN per customer · customer-controlled VPN |
| Out-of-band mgmt | Dedicated management network · iDRAC behind hardware-MFA jump host |
| OS / Stack | Ubuntu + Lambda Stack · vLLM / TGI for inference · PyTorch / HF for training |
The Backplain stack is open-source-first underneath. The Backplain-specific layer (AI Firewall, multi-model orchestration, audit fabric) is the product. Customers can audit it. Customers cannot be locked into it.
Who does what — the shared-responsibility shape.
| Facility operator | Backplain | Customer |
|---|---|---|
| Building, power, cooling, physical security | Cage internals, hardware, network gear | Workload code · policies |
| Tier 3 facility compliance attestations | Hardware lifecycle · OS · Backplain stack | Identity provider integration |
| Carrier-neutral transit options | Customer VLAN configuration | Encryption keys (BYOK option) |
| Smart hands for physical interventions | 24×7 monitoring · SecOps · audit fabric | Audit log review · retention policy decisions |
Four kinds of buyers. Four different reasons.
Defense contractors
Run AI workloads on hardware that has never touched a public cloud. Default deployment: our private San Diego Tier 3 DC with Full Managed Ops, or customer site for SCIF-adjacent requirements. The Carlsbad proximity matters in this segment.
Late-stage biotechs
Fine-tune on a proprietary corpus at the Carlsbad AI Lab. Run production at our private San Diego Tier 3 DC on a dedicated server. Compound codes, trial identifiers, and IND content stay inside Backplain infrastructure end-to-end.
Regulated finance & healthcare
Sovereign Compute deployed at our private San Diego Tier 3 DC with Full Managed Ops, paired with the Sovereign software tier. Audit-trail and BAA postures are why.
Government & gov-adjacent
Default deployment in our private San Diego Tier 3 DC. For the strictest cases, on-site air-gapped on customer hardware. (Backplain is pursuing FedRAMP certification — not yet listed on the FedRAMP Marketplace.)
Sovereign is for the data class, not the whole stack.
A Sovereign Compute deployment is not "rip and replace." A contractor running classified-adjacent workloads on a Lambda Scaler at our San Diego Tier 3 DC can still have Claude Enterprise for marketing copy, ChatGPT Business for engineering brainstorming, and Microsoft Copilot for Office integration. Sovereign is the architecture for the workloads that cannot leave a controlled boundary; the rest of the AI stack continues to live wherever the rest of your enterprise IT lives.
That framing matters in the procurement conversation. The buyer does not have to defend a wholesale tooling change — only the deployment of dedicated infrastructure for the specific data class that requires it. That is a much shorter conversation with the CFO.
Start a Sovereign Compute conversation.
Inquiries route directly to the founder. Site visit by arrangement. Spec sheet, SOC 2 reports, and BAA templates available under standard NDA.