# Ekim AI > Ekim AI builds AI factories in Australia beside renewable generation and runs them itself. Beside the factory sits a floor that builds agentic products, runs them on its own company first and then deploys them inside the client's organisation. Operated by Neo Holdings Pty Ltd, ABN 13 703 040 064. The AI factory sells GPU capacity on a date: sovereign Australian GPU racks at a wholesale rate, agreed before signature and held for the term, with the power behind the meter. The first phase is in design and capacity is contracted before construction starts. No capacity is live yet. ## Pages - [Ekim AI, the AI factory](https://ekim.ai/): the four layers, the sovereignty position, power at the source, the operating model, the partner bench, the contract rules, the products and the capacity request, on one page. - [Agentic Innovation Factory](https://ekim.ai/agentic-innovation-factory): the floor beside the AI factory, where agentic products are built, run on our own company first and then deployed inside the client's organisation. Agentic means agents do the work and named people sign it. Includes the in-house platform record and the people who sign. - [Dark Factory](https://ekim.ai/dark-factory): an agent build team for governed software engineering. It works specify-first with enforced quality gates, ships in the client's repositories and leaves the code with the client. Delivered as forward-deployed missions, a hosted MCP server or a private VPC licence. - [Tommy & Spartan](https://ekim.ai/tommy-spartan): a marketing department built as software, with two named partners signing every recommendation. Offered as the SIGNAL Assessment, a 30-day foundation and a monthly retainer. - [Graphene](https://ekim.ai/graphene): the scheduling layer for GPU fleets. The Margin Router places each inference request on the lowest-cost node that meets the latency target, with failover and OpenAI-compatible and MCP endpoints. The API is pre-GA. - [Agentic Engineering](https://ekim.ai/agentic-engineering): a site suitability verdict for AI factory and data centre sites, read by engineering desks in parallel and closed by a readiness gate of GO, CONDITIONAL GO or NO-GO. Begins with a First Read. - [Site map](https://ekim.ai/sitemap): every page and section of the site, with the machine-readable files. ## The AI factory: four layers, one operator, one site - The building: a powered shell beside wind or solar generation we hold, behind the meter, firmed by battery storage, with a grid connection brought in over time. 25 MW IT per block. - The machines: NVIDIA GB300 NVL72 in the first phase, then Vera Rubin as it arrives, on NVIDIA's reference design. Whole racks, single tenant, at a rate fixed for the term. - The compute: three tiers on Graphene. Standard (by the hour), Reserved (committed term, priority on the queue) and Sovereign (Australian-owned and operated, engineered for assessment under IRAP at OFFICIAL: Sensitive and PROTECTED). - The platform: managed services between a bare GPU and a model in production, with nothing to pay to move your data out. ## Contract shapes - Sovereign Reserved (5 years), Wholesale Capacity (5 years), Enterprise Committed (3 years) and Premium Inference API (2 to 3 years). Two years is the shortest commitment we contract. Rates, service levels and the full specification follow once an NDA is in place. ## Reserve capacity - Form: https://ekim.ai/#reserve, with first name, surname, work email, phone, company, role, GPU type, GPU quantity, term and the window the capacity is needed live. - Requests reach mike@ekim.ai. ## Contact Mike Doria, CEO and Founder: mike@ekim.ai