Fragmented orchestration
Schedulers, cluster managers, network domains, and facility systems can expose separate operating views across accelerator types and locations.
DOKU SOVEREIGN · JAPAN
Accepting Japan design partnersDoku designs and delivers an operating control layer for new sovereign AI-cloud buildouts. It brings GPU-capacity orchestration, tenant isolation and governance policy, reliability automation, usage metering, and the path from commissioning to production operations together under the infrastructure operator’s authority.
Moves to a concise non-confidential inquiry form. Email remains available as a fallback.Founder experience across
FluidstackDesign-partner platform scope
Doku Sovereign is being designed as a modular operating control layer above operator-owned infrastructure and existing systems. Software scope, integration method, and production sequencing are agreed for each engagement.
Organize queues, quotas, topology, workload placement, and operating rules across heterogeneous accelerators.
Structure health signals, draining, recovery, failure domains, and maintenance-capacity workflows.
Define customer-controlled deployment, identity and policy enforcement, audit evidence, tenant isolation, and in-country data boundaries.
Define evidence and interface points for usage, allocation, service catalog, billing, and reconciliation.
Structure infrastructure validation, burn-in, acceptance criteria, evidence collection, and production-readiness decisions.
Connect utilization, available capacity, revenue-capacity, and operating risk to operator decisions.
Reference architecture
Doku is conceived as a neutral layer above the physical estate and below the service catalog, coordinating acceptance, placement, reliability, governance, and metering workflows.
Power · cooling · networking · servers · accelerators
Logical coordination for acceptance, policy, reliability, metering, and economics
Clusters · containers · inference endpoints · training jobs · APIs
Reliable capacity · tenant services · utilization · billing evidence
JAPAN-SPECIFIC OPERATING MODE
A sovereign cloud in Japan may need to coordinate capacity distributed across facilities, GPU generations, and operating organizations without consolidating ownership. Doku’s reference model is intended to coordinate capacity discovery, policy- and topology-aware placement, maintenance and failure routing, and utilization evidence through a neutral layer while each participant retains its authority.
Integration with the existing estate
Doku Sovereign is not a monolith intended to replace schedulers, cluster managers, identity, telemetry, facility controls, billing, or the service catalog. It is designed to preserve chosen systems and connect agreed operating workflows and evidence modularly. The labels below are neutral interface domains, not claims of certified integration.
Deployment model
Doku is not a GPU reseller or a requirement to depend on a foreign public cloud. These are reference options for design discussion; delivery scope, operating responsibility, and support access are defined before engagement.
Map hardware, facilities, networking, workloads, governance requirements, and the commercial model.
Agree infrastructure validation, acceptance criteria, control-layer configuration, and the initial operating workflow.
Use reliability, utilization, allocation, and tenant-operating evidence to plan phased expansion.
Software deployment targeted to operator-controlled infrastructure in Japan
A dedicated configuration isolated for the operator
Software modules with scoped integration and commissioning support
OPERATOR CONTROL BOUNDARY
This table identifies the dimensions reviewed in a Japan reference design. These are design targets to define in technical review and contract, not pre-existing guarantees.
Defined per engagement, with operator-controlled Japanese infrastructure as the reference target
Design targetWorkload and tenant data remains in operator-selected systems; data paths are agreed during design
Design targetOperator ownership is the reference target; connection to existing identity and key management is confirmed per stack
Design targetThe operator owns policy and final authority; the Doku layer coordinates agreed enforcement workflows
Responsibility boundaryLocation, retention period, and access are defined against operator requirements
Design targetAn approved, time-bounded access model is the reference target; the runbook is defined per engagement
Design targetDesigned to follow the operator’s change-control and approval workflow
Design targetRetained by the operator or its nominated infrastructure owner
Operator controlledPricing, contracts, and customer relationships remain with the operator
Operator controlledDesign-partner work product
Initial work is organized into implementation-oriented technical outputs tied to the operator’s existing estate and launch plan.
Initial conversations are technical fit reviews scoped to the operator’s infrastructure, governance requirements, existing stack, and launch plan. No confidential information is required.
Request a Japan architecture reviewFOUNDER-LED
Jorg Doku leads Doku Sovereign design conversations directly, drawing on work across production GPU-cloud operations, frontier AI, and compiler and accelerator infrastructure.
GPU-cloud deployment and operating experience at RunPod and Fluidstack.
Research and engineering experience at Google Brain and Meta AI.
Technical work spanning compilers, accelerators, and orchestration.
A practical focus on reliability, commissioning, capacity operations, and infrastructure economics.
NON-CONFIDENTIAL INQUIRY
Tell us briefly about your current environment and planning stage using non-confidential information only. We use the submission solely to assess fit and respond.