Artificial intelligence
H2O.ai, known for its open source AutoML engine, offers an Enterprise AutoML and GenAI-oriented platform. It combines data preparation and modeling tools, generative models (h2oGPTe) and advanced governance capabilities. An H2O.ai release (September 2025) describes the addition of RAG support, integrations with Amazon Bedrock and secure hybrid deployments, as well as role-based access controls and enhanced PII safeguards. A BusinessWire announcement states that H2O.ai has integrated NVIDIA Nemotron models and offers NIM microservices to improve performance and reduce costs.
h2oGPTe with RAG: the latest version of H2O.ai’s generative model includes retrieval-augmented generation (RAG), which improves the relevance of answers by combining information generation and retrieval.
Cloud integrations: the platform can be deployed on-premise, in the public cloud or in hybrid mode. It integrates with Amazon Bedrock, Dell AI Factory and other environments, enabling great deployment flexibility.
NIM microservices and NVIDIA models: H2O.ai has integrated NVIDIA’s Nemotron models via microservices, offering fast performance and cost-efficiency for inference.
Security and governance: the platform offers role-based access controls, enhanced safeguards to detect sensitive information (PII), and monitoring tools for compliance. It also supports model lifecycle management.
AutoML and feature engineering: H2O.ai automates data preparation, model selection and evaluation, reducing the time needed to create high-performance models.
The H2O.ai platform is made up of several modules. h2oGPTe is its next-generation generative engine, featuring retrieval-augmented generation (RAG ) to enrich responses with up-to-date information from databases and documents. The 2025 updates introduce Amazon Bedrock integration to connect the H2O engine to models hosted in a private VPC, as well as a hybrid mode that offers companies the option of deploying agents on-premise, in the cloud or in a mixed environment. H2O.ai has also integrated NIM microservices based on NVIDIA’s Nemotron models, improving performance and reducing inference costs. These microservices provide additional flexibility: companies can run specialized models (vision, text, multimodal) independently, and combine them in larger workflows.
In addition to h2oGPTe, the H2O.ai suite includesAutoML tools (Driverless AI), a labeling assistant for data annotation, feature preparation and engineering tools, and a low-code application framework called H2O Wave, which enables user interfaces to be created around models. The platform emphasizes data sovereignty: companies retain control over their models, datasets and parameters, and can apply granular access controls and safeguards to protect sensitive data.
H2O.ai boasts a customer base of over 20,000 organizations and numerous industry partners. Companies such as Commonwealth Bank of Australia have reduced scam losses by 70% thanks to real-time predictive models, whileAT&T has cut its call center costs by 90% by integrating H2O.ai solutions. Other notable partners include Singtel, Chipotle, Workday, Progressive Insurance and the National Institutes of Health (NIH). H2O.ai is also working with Dell to integrate h2oGPTe into the Dell AI Factory, a hardware and software infrastructure that combines Dell’s resources with NVIDIA GPUs to deploy sovereign agents on a large scale.
The use cases covered go beyond conversational chats. Companies are using H2O.ai to detect fraud, predict risk, analyze customer sentiment, summarize medical documents and generate regulatory reports. RAG support and the ability to control the deployment environment make the platform particularly attractive for sensitive sectors. Nemotron models provide multimodal capabilities (text and images) useful for optical character recognition, image classification and the comprehension of complex documents.
H2O.ai promotes the concept of Sovereign AI, which aims to guarantee companies complete control over their intellectual property. Integrations with Amazon Bedrock and partnerships with Dell offer hybrid deployment: customers can decide to host models and data in their own infrastructure, in the cloud or in a mixed configuration to meet regulatory or performance requirements. The platform provides role-based access controls, enhanced safeguards to detect personally identifiable information (PII) and monitoring tools that alert to any drift, ensuring compliance with regulations such as RGPD or HIPAA.
The 2025 updates emphasize the importance of ergonomics: H2O.ai has introduced a unified interface to simplify the use of h2oGPTe and harmonize the user experience between modules. Improved visualization of workflows and performance metrics helps data scientists and business teams to collaborate and monitor the impact of models. At the same time, integration with Amazon Bedrock enables H2O models to be used within a secure VPC, combining the flexibility of the cloud with data sovereignty.
H2O.ai is positioned as an alternative to conversation-centric agent platforms, focusing ondata analysis, AutoML and augmented generation. NVIDIA microservices integration and a hybrid deployment option make it a suitable choice for organizations that want to leverage their in-house data without relying on a single vendor. New safeguards and role-based access controls address growing security and compliance concerns.
However, the platform requires a certain amount of expertise in data science and model management, which can be a hindrance for teams lacking these skills. What’s more, although H2O.ai has a broad ecosystem, its focus on AutoML and sovereignty means that companies primarily looking for out-of-the-box conversational agents might prefer solutions like OpenAI Frontier or Vertex AI. Despite these limitations, H2O.ai’s approach brings unique value to organizations wishing to combine predictive and generative intelligence while retaining control of their data.
Data orientation and AutoML: unlike conversational agent-centric platforms, H2O.ai focuses on data analysis, AutoML and augmented generation. This suits companies looking to create predictive models and generative applications on their own data.
Deployment flexibility: the platform can be deployed in different environments (on-premise, public cloud, hybrid), which is ideal for organizations concerned with data sovereignty.
Integrations and performance: compatibility with Amazon Bedrock, Dell AI Factory, NVIDIA models and NIM microservices offers high-performance, cost-effective options.
Less focused on conversational agents: although it offers a generative model (h2oGPTe), H2O.ai is more oriented towards AutoML and data analysis. Companies wishing to develop complete conversational agents will need to integrate other tools.
Complexity: using the platform may require data science expertise to adjust AutoML parameters, supervise models and interpret results.
Smaller ecosystem: compared with AWS or Google, the catalog of partners and ancillary services is smaller, although the announced integrations partly compensate for this.
| Solution | Strengths | Weaknesses |
|---|---|---|
| AWS Bedrock/AgentCore | Access to various models, strict governance, Amazon integration | Technical complexity and dependence on AWS |
| Google Vertex AI | Integrated observability, evaluation and RAG | Dependence on Google Cloud |
| OpenAI Frontier | Open agents with context sharing | Limited availability and undisclosed pricing |
| IBM Watsonx | Strong governance, industry-specific solutions | High implementation costs |
What is the H2O.ai platform?
– An AutoML and GenAI oriented solution combining a generative model (h2oGPTe) with RAG, data preparation and modeling tools, and enhanced governance control.
Benefits?
– Flexible deployment (on-premise, cloud or hybrid), integration with Amazon Bedrock and Dell AI Factory, NIM microservices and NVIDIA models for high performance.
Limits?
– Less focused on conversational agents, need for data science skills, smaller ecosystem.
For whom?
– Companies who want to develop predictive models and generative applications while retaining control over their data, particularly in sectors such as finance and insurance.