Sovereign AI Cloud — MEDIA HOLDING INTELLIGENCE ECOSYSTEM
Sovereign AI inference for businesses: open-weight models on dedicated instances, EU jurisdiction, zero retention of your content — built and operated by Media Holding.
What the platform is
Sovereign AI — inference
MeasuredThree open-weight models — Qwen3.5 4B and 9B, Gemma 4 12B — served by llama.cpp through an OpenAI-compatible API format. Measured on 2026-10-12: 24.6 tokens per second single-stream on the 9B model, a three-hour soak of 1,617 requests with zero server errors. No OpenAI services are involved.
OpenSovereign AI Machine — managed execution
BuiltThe execution component that routes every request to Sovereign AI with no external AI supplier in the path. It exists with 151 unit tests and 14 named acceptance checks; the copy that will serve customer tenants is being adapted.
OpenDedicated enterprise instances
PlannedA GPU node reserved for one customer in an EU data centre: your models resident, your own limits, your own audit trail. The first commercial offer of the platform; cloud operation starts after the infrastructure decision.
OpenCloud cabinet and API keys
BuiltOrganisations, projects, API keys whose secrets are stored only as a keyed one-way hash (HMAC-SHA-256), usage by project, budgets and limits, invoices. Served by the control plane under /cloud/.
OpenZero retention of content
BuiltThe inference gateway writes one audit line per request — request id, model, status, token counts, duration — and never the prompt or the completion. The platform default is no retention of request content.
OpenStatus labels — MEASURED — exists and was measured on the stated date by the stated method. BUILT — exists and works, not yet measured under load. PLANNED — designed, not yet built. A line without a label is not a claim of this platform.
Why sovereign
Sovereignty here is concrete, not a slogan. The models are open weights under the Apache 2.0 licence. The inference runs on infrastructure Media Holding controls. The jurisdiction is Estonia and the European Union. The content of your requests is not stored. There is no third-party model provider behind the API and no reseller between you and the machine.
What it is not
Sovereign AI Cloud does not compete on the lowest token price. Per-token APIs for 9B-class open models are cheaper elsewhere, and we say so. What you pay for is a reserved machine under EU jurisdiction, a written data-processing agreement, a sub-processor list you can audit, and an operator you can reach.
OpenAI-compatible API format. No OpenAI services are involved.
Three parts, one platform
Sovereign AI
The inference provider: the models, the llama.cpp runtime, the OpenAI-compatible API format, the governor that keeps one machine honest.
OpenSovereign AI Machine
The execution component: takes a request from the control plane, routes it to Sovereign AI, records metadata, returns the answer. Offered as Tier B — managed execution.
OpenSovereign AI Cloud
The control plane: accounts, organisations, projects, API keys, metering, limits, budgets, billing, the cabinet and this site.
OpenHow to start
Read the models and the measured numbers
Every model carries its parameters, context window, licence and measured speed with the date and the method. Nothing is claimed without a label.
Create an account in the cabinet
Business customers only: company name, registration number and VAT number are asked for before a paid service is activated.
Order a dedicated instance or prepaid credits
A dedicated instance is a reserved GPU node with a signed data-processing agreement and a service level agreement; prepaid credits are consumed by usage and expire after twelve months.
Point your client at the API host
Set
base_urlto the API host and use your project key. The request and response format is OpenAI-compatible, so your client library works unchanged. No OpenAI services are involved.
Steps 2–4 run today in the isolated build: the control plane passes its tests and is labelled BUILT. They open to customers with publication, which is a separate decision.
Part of
Sovereign AI Cloud is a product of MEDIA HOLDING INTELLIGENCE ECOSYSTEM — the group founded by Pavel Fjodorov in Tallinn, Estonia. It is not an independent company and not a separate brand: the same identity, the same design language and the same standard of honesty as every other product of the group.