Valohai

What Is Valohai?

Reproducible machine learning training at scale is a problem that predates the current LLM boom by several years. It is the one a small Finnish team led by CEO Eero Laaksonen set out to solve, registering Valohai Oy in Turku in October 2016 under business ID 2786205-7. Before "MLOps" was a common term, data science teams were already losing days to the same failure mode. A model would train successfully on one engineer's laptop and then prove impossible to reproduce six months later, on a different cloud, by a different team member.

Valohai's answer was to build an orchestration layer that treats every training run as a versioned, reproducible unit — inputs, code, environment, metrics, and outputs all captured automatically. That layer was designed to work identically regardless of which cloud, or no cloud, the compute actually runs on. This cloud-agnostic bet has aged well: the platform now runs on AWS, Azure, Google Cloud Platform, OpenStack, Scaleway, and Kubernetes. It can also be deployed fully on-premises or air-gapped for organisations that cannot let training data touch the public internet at all.

Valohai has stayed small and Finnish throughout. It raised a $1.8 million seed round in 2018, led by Nordic investor Superhero Capital with participation from Reaktor Ventures and the Finnish government's Business Finland agency, bringing total disclosed funding to around $2.47 million. That is a modest sum for an infrastructure company competing conceptually with venture-scale rivals like Databricks and Weights & Biases, and it shows up in the company's size: roughly 29 employees as of 2026.

Key Features

Pipeline Orchestration Built for Reproducibility

Every Valohai execution — a single training run or a multi-step pipeline — is versioned automatically. Metrics, metadata, logs, and the exact environment used are captured without manual instrumentation, and any past run can be reproduced with one click. Teams that have ever tried explaining to an auditor why a production model can no longer be recreated exactly will recognise the value here. It is the feature that justifies adopting a dedicated orchestration layer instead of ad hoc scripts.

Genuine Multi-Cloud Flexibility

Many "multi-cloud" ML platforms really mean "runs on AWS, with some support for others." Valohai instead treats every supported target — AWS, Azure, GCP, OpenStack, Scaleway, Kubernetes, or bare on-premises hardware — as a first-class citizen. The company extended this further with an OVHcloud partnership announced in August 2024, part of OVHcloud's Open Trusted Cloud programme, plus a listing on Oracle Cloud Marketplace from September 2025. European teams wary of AWS/Azure/GCP concentration get a genuine EU-cloud execution path that most MLOps vendors simply don't offer.

Air-Gapped and On-Premises Deployment

Valohai can run fully disconnected from the internet. The company has demonstrated this in real deployments involving highly sensitive training material, where even the ML engineers building the models could not directly access the underlying data. For defence, government, or safety-critical industrial customers, this is not a checkbox feature — it is frequently the deciding factor in whether a platform can be used at all.

Framework and Language Agnosticism

Because every execution runs inside a Docker container, Valohai does not care whether a team works in Python, R, or a niche in-house framework. Existing MLflow, SageMaker, or Kubeflow users can migrate incrementally, keeping their training code largely unchanged while Valohai takes over orchestration, versioning, and scheduling. Jupyter notebooks work alongside plain scripts, and integrations with Snowflake, Redshift, BigQuery, and Labelbox cover the data side of a typical pipeline.

Model Registry and Lineage

A built-in model registry tracks lineage from raw data through every intermediate pipeline step to the final trained model. This supports the kind of audit trail that regulated industries increasingly require for any model touching customer data or safety-relevant decisions.

Pricing

Valohai does not publish a price list. The company uses a per-user monthly licence that includes unlimited projects, experiments, pipelines, and deployments — you pay for seats, not for usage volume — but the actual number requires a sales conversation. A 14-day free trial gives prospective customers hands-on access before that conversation happens, and there is no permanent free or open-source tier.

That opacity is a genuine trade-off worth naming plainly. Feature-store peer Hopsworks, also an EU MLOps-adjacent product, ships a fully open-source community edition with no artificial feature gating. That is a meaningfully different value proposition for teams that want to start free and only pay once they scale. Valohai's all-custom pricing means smaller teams can't self-serve a quote, and larger teams should budget time for procurement before assuming Valohai fits their cost envelope.

Third-party software directories cite ballpark starting prices anywhere from roughly $350 to $560 per user per month. None of those figures come from Valohai directly, and the company's own pricing page displays no numbers at all. Prospective buyers should treat any third-party estimate as a rough planning guide rather than a quote, and confirm actual cost through Valohai's sales team or the 14-day trial process.

EU Compliance & Privacy

Valohai Oy is a Finnish limited company under full EU jurisdiction, confirmed active in Finland's official business register with no foreign parent or holding structure on file. That EU incorporation matters less for data processing specifically than Valohai's deployment flexibility does. Because the platform can run fully on-premises or air-gapped, an organisation's most sensitive training data never has to leave infrastructure it directly controls. The OVHcloud partnership adds a concrete EU-cloud execution option for teams that want managed infrastructure without relying on a US hyperscaler.

Valohai has not published ISO 27001 or SOC 2 certifications on its own site. That is worth flagging for procurement teams running a formal certification checklist, even though the underlying deployment model — on-premises or air-gapped — can substitute for some of what those certifications are meant to assure.

Who It's Best For

Regulated industries running sensitive ML training — finance, healthcare, defence — get real value from Valohai's on-premises and air-gapped deployment options, which few MLOps platforms offer with the same maturity.

Teams already committed to multi-cloud or hybrid infrastructure benefit from Valohai's genuine cloud-agnosticism, avoiding the lock-in that comes with AWS SageMaker or Azure ML's native tooling.

European organisations wanting an EU-cloud execution path can use the OVHcloud partnership as a concrete alternative to routing training workloads through a US hyperscaler by default.

Teams building primarily LLM or agent applications should look elsewhere first. Valohai is classical ML pipeline orchestration and training infrastructure, not an LLM framework. It has no relationship to tools like LangChain or LlamaIndex, so organisations whose core workload is prompt engineering or RAG rather than model training will find little here built for that use case.

The Verdict

This is a small company solving a real problem — reproducible, cloud-agnostic ML training — with a level of deployment flexibility that larger, better-funded rivals often can't match for on-premises and air-gapped use cases. Its Finnish base, EU jurisdiction, and now OVHcloud partnership give it a genuine European infrastructure story. Set against that: quote-only pricing, no free tier, roughly 29 employees and under $2.5 million raised since 2016, and a product built for classical ML pipelines rather than the LLM-and-agent tooling most of the market now prioritises. Teams whose problem is exactly "reproducible training across clouds" should take Valohai seriously; teams building LLM applications need a different category of tool entirely.

Frequently Asked Questions

Is Valohai GDPR compliant?

Yes. Valohai Oy is registered in Turku, Finland (business ID 2786205-7) and is fully subject to EU data protection law. Because Valohai can be deployed on-premises or fully air-gapped, organisations can also keep ML training data entirely inside their own infrastructure with no external data flows at all.

Which clouds does Valohai support?

Valohai runs on AWS, Microsoft Azure, Google Cloud Platform, OpenStack, Scaleway, and Kubernetes, plus on-premises and air-gapped installations. It has also partnered with OVHcloud (since August 2024) and is listed on Oracle Cloud Marketplace (since September 2025), giving European teams more EU-based deployment options than most US-centric MLOps platforms.

How much does Valohai cost?

Valohai uses a per-user monthly licence covering unlimited projects, experiments, pipelines, and deployments, but does not publish a list price. Every quote requires contacting the sales team. A 14-day free trial is available to test the platform before committing.

Is Valohai an LLM framework like LangChain?

No. Valohai is ML pipeline orchestration and training infrastructure — it schedules, versions, and reproduces training runs across clouds. It is not an application framework for building LLM-powered products, and teams should not expect LangChain- or LlamaIndex-style agent-building features from it.

How big is the company behind Valohai?

Valohai is a small company: around 29 employees, founded in 2016 in Turku, Finland, with roughly $2.47 million raised in total, including a $1.8 million seed round led by Superhero Capital with Reaktor Ventures and Business Finland. It has stayed independent, with no acquisition on record.