Fundraising Fox

Unikraft High-performance cloud infrastructure

3 known investors

Unikraft provides a compute platform that transforms Dockerfiles into minimal Linux-based unikernel VM images, offering VM-grade isolation with sub-10ms cold starts and high density (100K+ instances per server). It targets developers and enterprises running sandboxes, AI agents, and serverless workloads, with deployment options including on-prem, BYOC, dedicated hosts, and integration into Kubernetes clusters.

Also known as Unikraft Cloud Β· unikraft.com Β· unikraft.io

Investors Β· 3

Company profile

researched Aug 2026

Unikraft operates Unikraft Cloud, a compute platform that runs application workloads as lightweight, specialised microVMs rather than containers or language-runtime isolates. The platform advertises cold starts of under 10 milliseconds for all workloads, scale-to-zero with sub-10ms stateful resume, and density of over 100,000 microVMs on a standard server, which the company frames as 10-100x better unit economics than conventional approaches. Workloads are defined from Dockerfiles and the platform integrates with existing tooling such as Docker, Kubernetes and Prometheus.

The product targets use cases including agent and code sandboxes, headless browsers (Chromium, Firefox), full remote desktop environments, build and test environments, serverless functions, serverless databases (PostgreSQL, MongoDB, Redis) and game servers. Platform capabilities include forking of live VMs and their children, reusable instance templates with code injected via "ROMs", checkpointing and rewind, live migration between servers, persistent storage shared across clusters, and autoscaling that can bring up additional hosts in under 20 seconds. Checkpointing is offered under limited access requiring a licence, and instance branching β€” creating an independent byte-for-byte copy of a running instance β€” is a preview feature not yet available on the public Unikraft Cloud.

The platform is operated through a REST API (Instances API v1) and a CLI (the current "unikraft" CLI and a legacy "kraft" CLI), with instances organised into service groups and addressed via per-metro domains. Instances move through states including stopped, starting, running, draining, stopping and standby (scaled-to-zero), and the API exposes detailed diagnostics such as boot and network setup times, stop reason bitmasks, application exit codes and guest-kernel stop codes encoding fault reasons and Linux errno values.

Business model

Unikraft sells access to a managed compute platform, with deployment options spanning the public Unikraft Cloud service, bring-your-own-cloud, on-premises and dedicated hosts. Some advanced capabilities, such as checkpointing, are gated behind a licence and limited-access programme.

Latest developments

The company announced Release 12, codenamed "Thebe", which adds GPU support, expanded image controls and a dashboard update. Documentation also describes two newer capabilities: checkpointing, available under limited access with a licence, and instance branching, a preview feature not yet available on the public Unikraft Cloud. A newer "unikraft" CLI has become the standard, superseding the legacy "kraft" CLI.

β–ΈFull profile β€” market position, technology, go-to-market, geography

Market position

Positioning rests on combining hardware-level VM isolation with startup and resume latencies normally associated with containers or language runtimes: sub-10ms cold start and scale-to-zero resume, 100K+ microVMs per server, and Dockerfile-based workload definition with compatibility with Docker, Kubernetes and Prometheus. Snapshot-centric features β€” millisecond forking, templates, checkpoints, branching and live migration β€” are presented as capabilities not typically available in container or serverless platforms.

Technology

The platform builds minimal, single-process microVMs that boot in milliseconds and run under a hypervisor, which the company contrasts with containers and language-runtime isolation on trusted-computing-base size: roughly 200-250k lines of code for a KVM-based VM boundary versus around 40M lines for the Linux kernel underpinning containers and about 47M lines for Linux plus a V8/Node runtime. The company cites a series of published container and runtime escape CVEs from 2024-2025 (including CVE-2024-21626, CVE-2024-0132, CVE-2025-9074, CVE-2025-23266, CVE-2025-31133, CVE-2025-38617 and CVE-2025-52565) as evidence that a smaller TCB reduces escape paths. Core platform technology centres on snapshotting: forking running VMs and their descendants, instance templates with ROM-based code injection, incremental checkpoints with history and point-in-time restore, live migration between hosts, and shared persistent cluster storage. Application-level integration is also supported, for example a pg_ukc_scaletozero PostgreSQL module that suspends scale-to-zero during query processing.

Go-to-market

The website directs prospects to talk to an engineer and states the product is in production with customers. Adoption is supported by public documentation, a CLI, an API and a GitHub examples repository (unikraft-cloud/examples), with deployment available as a managed service, BYOC, on-premises or dedicated hosts.

Developers and engineering teams running sandboxes, AI agents and sub-agents, headless browsers, remote desktop environments, CI build and test environments, serverless functions, serverless databases and game servers, including SaaS platforms with spiky or low-utilisation workloads.

Geography

Unikraft Cloud is organised into regional metros; documentation uses Frankfurt, Germany ("fra") as the example metro, with API endpoints at api.fra.unikraft.cloud and instance domains under fra.unikraft.app.

Compiled by commissioned research from 5 cited public sources β€” announcements, filings, and press listed under research sources below.

Key figures

latest reported
Autoscale host provisioning timeJan 2026new fleet of hosts launched in <20 seconds
Cold start timeJan 2026<10ms for all workloads
Infrastructure cost reductionJan 2026up to 99% infra cost cuts
Instance density per serverJan 2026100,000 microVMs per server
Scale-to-zero resume timeJan 2026<10ms
Trusted computing base size (containers, Linux kernel)Jan 2026~40M lines of code
Trusted computing base size (KVM virtual machine)Jan 2026~200-250k lines of code
Trusted computing base size (runtime isolation, Linux + V8/Node)Jan 2026~47M lines of code
Unit economics improvementJan 202610-100x better unit economics

Company-reported or press-reported figures, each dated to when it was claimed β€” not independently audited.

Timeline Β· 3

launches, deals, and filings
Jan 2026
Checkpointing available in limited access

Checkpointing, which captures the complete memory and volume state of a running microVM without interrupting execution and supports incremental, named restore points, is offered as a limited-access feature requiring a valid licence.

source β†—

Jan 2026
Instance branching released as preview feature

Instance branching, which creates an independent copy of a running instance with identical state in milliseconds, is a preview feature not yet available on the public Unikraft Cloud.

source β†—

Jan 2026
Release 12 "Thebe" adds GPU support, image controls and dashboard update

Unikraft announced Release 12, codenamed "Thebe", introducing GPU support, improved image controls and a major dashboard update.

source β†—

Dated company events from announcements, filings, and press; legal rows summarize public dockets and regulator releases.

β–ΈResearch sources Β· 5

primary sources listed

5 public sources were cited for this profile; the first-party ones are listed here.

Frequently asked questions

What does Unikraft High-performance cloud infrastructure do?
Unikraft Cloud runs microVM-based workloads that cold-start in under 10ms with VM-grade isolation and high per-server density.
Who are Unikraft High-performance cloud infrastructure's investors?
Unikraft High-performance cloud infrastructure's investors include Fly Ventures, First Momentum, Mango Capital.