Offer a European AI API from your own regions.
Your customers' compliance teams want an AI API that stays in Europe. Their developers want the APIs they already code against. ARK Core gives you both: open-weight models behind OpenAI- and Anthropic-compatible endpoints, running in your data centres, under your brand and on your price list.
Europe's sovereign cloud spending is forecast to grow 83% in 2026, to $12.6 billion, and is forecast to overtake North America's in 2027.
Gartner, 9 February 2026 (sovereign cloud IaaS). Worldwide: $80.4 billion in 2026.
ARK Cloud runs the same software you would license. Measured in October 2026:
Qwen3.5-122B-A10B on four RTX 5090, 60 concurrent requests, through ARK Cloud's public API.
Qwen3.6-35B-A3B on four RTX 4070 Ti SUPER, 48 concurrent requests, public API.
Per million tokens on that same deployment with 16 or more users, about 0.025 Wh per request.
Why your customers haven't bought AI from you yet.
The AI has to come with the workload
Customers moving to European clouds expect AI services in the same region, not a detour back to a hyperscaler's API.
In July 2026 Airbus chose Scaleway as its sovereign cloud provider after a competitive tender. Press reports put the first phase at about 70 applications now on a US hyperscaler. — Scaleway; The Register
Sovereignty is being graded
The Commission's sovereign cloud framework scores providers against sovereignty levels, and the draft Cloud and AI Development Act proposes four. “Sovereign” will be checked, not claimed.
Power is the constraint
European data-centre power is expensive, and new grid connections take years. Every token has to earn its watt-hours.
An API is a product, not a server
Developers expect OpenAI-style endpoints, current open models, streaming, tool calling and long context on day one. Serving, metering and keeping models current take a full-time platform team.
Your AI service, ready to sell.
Everything between your GPUs and your customers' code, running in your regions.
An API under your name
OpenAI- and Anthropic-compatible endpoints on your own domains: chat, vision, embeddings and speech to text, with streaming, tools and reasoning.
Identity and billing your way
Keycloak, or your own identity and billing systems through HTTP hooks. Every request reports its token usage, cached and reasoning tokens included.
Open models from any maker
69 official models in nine families, plus more than 29,000 fine-tunes, with 19 chat models benchmarked on every release. It is your own catalogue, not one model vendor's API resold in your region.
Inside your operations
The whole inference stack runs in your data centres and is operated by your team, with no third-party API in the data path. Your customers' assurance questions get your answers.
More tokens per watt
ARK shares every GPU step across all requests in flight, so energy per token falls as traffic grows: 1.8 to 4.7 times less with 16 or more users than with one.
Elastic and resilient
ARK runs on every NVIDIA generation from Turing to Blackwell, data centre to GeForce, and one model can span GPUs of different generations, sizes and counts across machines over standard Ethernet. ARK Cloud runs Qwen3.5-122B-A10B on four RTX 5090 in two machines, with its image encoder on an RTX 4060 Ti in a third. Servers join or leave without a restart, and with a second copy of each part a failed server's requests are replayed on another path.
Your infrastructure, your customers, your pricing.
You own the customer relationship, the regions and the price list. ARK Core is licensed per GPU, with no per-token fee: your brand, your billing. Sovereign capacity already sells at a premium: where AI vendors offer EU or regional endpoints, they price them about 10% above their global rates.
ActuIA review of 18 price lines from six AI vendors, October 2026.
How you bring ARK into your regions.
ARK Core Your service
The platform and its models, with documentation and release packages, licensed per GPU. Your team runs it as your own service, under your brand. Our engineers help size and integrate your first region.
ARK CorePrototype firstARK Cloud
Build and test your offer against ARK Cloud's API before your own regions are live. Same API and models, so nothing changes when you switch.
ARK CloudPlan your launch with us.
Tell us your regions, the GPUs you have or plan to buy, and the models your customers ask for. We reply with a deployment plan, benchmarks on comparable hardware and the partner terms, including deal registration and engineering support on your first customer projects.