# Neo Cloud & Inference

> Neo cloud and inference from Boltcliq: GPU capacity and model serving closer to where African data is created, with local data centre partners, data residency, low latency and predictable cost.

Canonical URL: https://www.boltcliq.com/solutions/neo-cloud-and-inference
Part of: https://www.boltcliq.com/llms.txt

Sovereign compute and inference, delivered with local partners.
- Africa should not have to rent all of its intelligence from other continents. We help bring GPU capacity and inference closer to where data is created, with local partners and local control.
- Local partners: Data centre, telecom and infrastructure operators
- Low latency: Serving closer to users and data
- Resident: Architecture that keeps sensitive data in-country
- Predictable cost: Capacity and unit costs known up front, not discovered
### What we do
- Today, many African organizations run AI on compute that sits in another continent. That works until it does not. Capacity is scarce and often rationed. Prices are set in foreign currency. Round trips add latency to every request. And some data cannot leave the country, or should not. A neo cloud is a newer kind of provider, built around GPU capacity and AI workloads, and it can be built locally.
- We do not own data centres, and we do not pretend to. We work with the partners who do: colocation and data centre operators, telecoms with fibre and edge sites, national research and education networks, power and renewable energy developers, and public cloud providers with regional presence. Our role is to design the architecture, help you choose and integrate partners, and operate the software layer that turns raw capacity into a usable inference platform.
- The work covers the whole path. We assess your workloads and what they need. We compare options for capacity, location and cost. We design serving for the models you use, with routing, caching, batching and fall-back between sites. We put in place the identity, metering and observability that a shared platform needs. And we build the operating model so your team, or a partner, can run it.
### Why location matters
- Latency is physical. A request that travels to a far region and back is slower than one served in the same city, and the gap matters for voice agents, fraud checks, clinical tools and anything that sits inside a customer interaction. Bandwidth also costs money, and moving large documents or images across continents is wasteful.
- Location also shapes risk. A model served in-country can keep prompts and results inside the same legal boundary as the data they concern. That does not remove the need for legal review, but it widens the options. Which rules apply depends on your sector and countries, so confirm with counsel.
- Lower latency for interactive AI
- Less cross-border traffic and cost
- Data and prompts kept within a chosen jurisdiction
### Capabilities
**Capacity strategy for scarce GPUs**
- We help you plan around GPU scarcity. That means measuring real demand, separating workloads that need top-end accelerators from those that run well on smaller ones, and using reservations, partner capacity and burst options together. We build a capacity plan with lead times, so you are not caught by a supply gap.
- Workload profiling and right-sizing
- Reserved, partner and burst capacity mix
- Lead-time and supply risk planning
- Scarcity punishes organizations that plan late.
**Local partner integration**
- We help you assess and integrate local infrastructure partners across categories: colocation and data centre operators, telecoms with edge and fibre assets, research and education networks and renewable energy developers. We review power, cooling, connectivity, security and commercial terms, and design a contract and technical structure that avoids lock-in. We describe partner categories here and do not imply any deals.
- Partner assessment criteria
- Power, cooling and connectivity review
- Contract and interface design
- Local capacity is built with local operators.
**Inference platforms**
- We design serving platforms for language, vision and speech models, open or licensed. The platform includes routing, batching, caching, quantisation options, autoscaling and fall-back to other sites. We test on your own prompts and loads and report latency and throughput honestly.
- Routing, batching and caching
- Autoscaling and multi-site fall-back
- Load tests on your own workloads
- Serving cost is mostly design, not hardware.
**Data residency by design**
- We classify workloads by data sensitivity and place them accordingly. Sensitive inference runs in-country or on-premises, less sensitive work can use wider capacity. Keys stay under your control, and logs and prompts follow the same placement rules.
- Placement by data class
- Customer-held encryption keys
- Logs and prompts governed like data
- Residency is easier to design than to retrofit.
**Latency and edge serving**
- We place small and mid-size models at edge sites near users and larger models at regional hubs, and route requests by need. For voice and interactive use, we measure end-to-end latency from the customer device, including network and mobile conditions, not only server time.
- Edge and regional tiering
- End-to-end latency measurement
- Support for low-bandwidth clients
- Customers feel latency, not benchmarks.
**Cost, metering and operations**
- We build metering by tenant and workload, unit cost reporting and budgets, so internal teams or paying customers see what they use. We set up observability, on-call and capacity reviews. Cost per thousand requests or per million tokens becomes a number you track, with currency exposure visible.
- Metering and chargeback per tenant
- Unit cost dashboards
- Capacity review and on-call process
- You cannot manage cost you cannot see.
### Where it applies
- Banks and fintechs
- A bank wants customer-service and fraud models but customer data must stay in the country, and foreign-hosted APIs raise compliance questions.
- We design in-country inference with a local partner, with keys held by the bank and logs inside the same boundary.
- The risk and compliance teams decide which workloads qualify. Counsel confirms the legal basis.
- Telecommunications
- An operator has fibre, power and edge sites and customers asking for AI services, but lacks a serving platform.
- We design an inference platform on the operator sites, with metering so enterprise customers can be charged by use.
- The operator decides on products, pricing and capacity investment.
- Government and public sector
- Agencies want AI services but do not want sensitive citizen data processed abroad.
- A shared in-country inference platform serves agencies under common policy, hosted with a local operator.
- Government decision makers and counsel decide what data may be used and for what purpose.
- Healthcare
- Hospitals want imaging and documentation tools but have limited bandwidth and strict privacy duties.
- We place models at regional sites or on premises, with tiering so only the necessary data travels.
- Clinical governance boards decide which tools to use and how.
### Built on
**Argus**
- Provides the portable runtime, storage and environments that sit above partner infrastructure.
- Boltcliq product
- Workloads can move between sites and providers.
**Odin**
- Handles identity, tenants and access for people, services and agents using the platform.
- Boltcliq product
- Shared capacity with separate, provable access.
**Cyclopes**
- Meters usage and cost, measures latency and quality and watches the health of sites.
- Boltcliq product
- You can run it like a service.
### What we do and do not do
- Design architecture and capacity strategy
- Assess and integrate local partner infrastructure
- Build and operate the serving and metering software layer
- Design residency and key control with your counsel
- Train your operators
- Own or build data centres
- Guarantee GPU supply or specific chips
- Claim partnerships we have not signed
- Give legal advice on data law
- Promise latency or cost before measuring on your workload
### Questions we hear
- Do you own data centres or GPUs?
- No. We work with data centre operators, telecoms and other infrastructure partners. We design the architecture, integrate partners and operate the software layer.
- Which partners do you work with?
- We work across categories: colocation and data centre operators, telecoms with edge sites, research and education networks, energy developers and public cloud providers. We will not name partners on this page unless an agreement is in place.
- Can you guarantee GPU availability?
- No. Supply is constrained and changes quickly. We plan capacity with lead times, a mix of sources and fall-back options so that you are less exposed.
- How do you handle data residency?
- We classify workloads and place them by data sensitivity, keep keys under your control and apply the same rules to logs and prompts. Which laws apply depends on your sector and countries, so confirm with counsel.
- How much will it cost?
- It depends on workloads, location and capacity source. We model unit costs, including currency exposure, before you commit. We agree scope and fees up front per engagement.