What is a neocloud? A practical definition for modern AI workloads

What is a neocloud?

A neocloud is a cloud platform engineered from the ground up to serve AI and data-centric workloads while keeping data within the customer’s jurisdiction. It bundles native AI services, serverless compute, and sovereign data hosting into a single, cohesive experience. This eliminates the need to stitch together separate AI, database, and storage providers, and it removes the hidden cost of always-on virtual machines that dominate traditional cloud bills. In practice, a neocloud lets you deploy an inference service, a stateless API, and a database that all stay in the same region, all while paying only for the compute that actually processes requests.

The core ingredients of a neocloud

The compute layer runs on a scale-to-zero model. Functions, event-driven microservices, and container workloads pause when traffic stops, so an idle service costs nothing rather than accruing the metered hours an always-on machine would. Wake-up happens on the first request, which is what makes the model usable for the bursty traffic most AI applications actually see. Above that sits the AI surface itself: inference endpoints and the surrounding tooling, integrated with the compute fabric rather than bolted on through a separate vendor account.

Where the data lives is the part that separates a neocloud from a generic AI platform. A database, an object store, and the audit trail all sit in the jurisdiction you choose, with encryption at rest and access controls handled as platform primitives rather than as a compliance service you buy separately. Data leaves that jurisdiction only when you explicitly say so. One dashboard, one API, and one CLI cover the whole surface, so provisioning and scaling do not turn into a project of stitching separate consoles together.

How it compares to the old model

In a traditional cloud, you might spin up a VM for your inference service, a separate managed database, and a storage bucket for training data. Each of those resources lives in its own namespace, has its own billing, and requires separate access controls. The bill is a patchwork of compute, storage, and network charges, and the customer often ends up paying for idle compute that never processes a request. In a neocloud, the same workloads are co-located, share a single billing stream, and the platform automatically scales resources to zero when idle. This eliminates the hidden cost of a VM that runs 24/7 but serves traffic only 10% of the time.

Real-world metrics

• In 2025, 73 % of our customers used the neocloud for at least one AI inference workload (Scalix internal survey).
• The average first-request latency for serverless functions in the neocloud is 120 ms, compared to 650 ms for a traditional VM that must wake from sleep (Scalix performance benchmark).
• The platform’s autoscaling engine can spin up a new inference instance in under 200 ms, ensuring that traffic spikes are handled without manual intervention (Scalix real-time monitoring data).

A deeper dive into the components

AI-native inference

The inference service is built on a custom runtime that can run models in TensorFlow, PyTorch, or ONNX format. It exposes a REST API that automatically scales based on request volume. Because the runtime runs on the same hardware that hosts the database, the network latency between model and data is minimal. This is a key advantage over traditional clouds where inference services often sit in a separate region from the data store.

Serverless compute

Functions in the neocloud are billed per second of active execution, with a minimum of one second. The platform keeps a pool of pre-warmed containers that can be reused for subsequent invocations, reducing cold-start latency. When traffic drops to zero, the containers are automatically de-allocated, and the cost for that period is zero. This is the same scale-to-zero model that powers our managed database, which suspends to zero when idle.

Sovereign data hosting

All data is stored in an object store that is physically located in the customer’s chosen country. The platform enforces encryption at rest using keys that the customer controls. Audit logs capture every read, write, or delete operation, and the logs themselves are stored in a separate, tamper-evident bucket. This architecture gives regulated buyers the substrate their regime asks for, without bolting on a separate compliance service.

Unified management

The dashboard lets you view all resources in one pane, see real-time usage, and set up alerts. The CLI and API expose the same set of commands, so you can automate provisioning with scripts or Terraform. The platform also supports role-based access control, so you can grant developers read-only access to the inference API while restricting database credentials.

How to get started

If you’re curious about how a neocloud can simplify your AI workflows, start by exploring our documentation. The docs walk you through creating a new project, deploying a model, and connecting it to a sovereign database. Once you’re comfortable, you can sign up for the free tier to experiment with a small inference service and see the cost savings for yourself.

Explore the documentation → /docs/neocloud-intro

Final thoughts

A neocloud is more than a marketing buzzword. It is a concrete architecture that brings together the most common pain points of AI-centric development: high-performance inference, cost-efficient compute, and data sovereignty. By building these capabilities into a single platform, we remove the friction of juggling multiple services and let you focus on building the next generation of intelligent applications. We look forward to seeing what you build.