Cloudera is expanding its enterprise AI platform through partnerships with Mistral AI, Neo4j, Cartesia and VAST Data, as it tries to make it easier for businesses to build AI systems using their own data, without having to rely entirely on public cloud services.
The announcement, made at the company’s EVOLVE26 event in New York on 27 October, builds on its 2024 and 2025 enterprise AI partnerships. The idea is to bring together technologies that businesses would otherwise have to integrate themselves, from the models that power AI applications to the infrastructure needed to run them.
Cloudera calls these environments AI factories, an increasingly popular term for systems that can continuously process information, run AI models and carry out tasks. The ambition is to move beyond AI assistants that answer questions towards software agents that can actually do things, such as analysing transactions, investigating problems or helping employees complete complicated workflows.
For South African businesses, there’s a practical reason to consider this approach. Many organisations have spent years moving their systems into the cloud, but the growth of AI is making them reconsider where some of those workloads should run.
In August, research commissioned by Cloudera found that 83% of South African organisations surveyed had moved at least some AI workloads from public cloud environments into private cloud or on-premises infrastructure. That was considerably higher than the global average of 66%.
Those figures come from research commissioned by a company that sells hybrid data infrastructure, so they’re hardly independent of Cloudera’s commercial interests. But they do point towards a problem businesses are having to deal with as AI becomes part of their everyday operations.
Running these systems can be expensive, particularly when large amounts of information have to move between databases, AI models and cloud services. There are also questions about who can access sensitive information and where that information should be processed.
Cloudera’s latest announcement is aimed at companies trying to work through those decisions.
Four partners, different parts of the same problem
One of the challenges with deploying AI inside a large organisation is that the information it needs is rarely stored in one place.
A bank, for example, might have customer records in one database, transaction histories in another and internal policies stored across several document management systems. An AI assistant that can search these systems might be able to answer a straightforward question, but giving it enough context to investigate suspicious activity or make a recommendation is considerably more complicated.
That’s where Neo4j comes in.
The company specialises in graph databases, which are designed to represent relationships between different pieces of information. Rather than simply retrieving a customer record, an AI system could use a knowledge graph to understand how that customer connects to particular accounts, transactions, businesses or other customers.
In a fraud investigation, those connections could help an AI agent identify patterns that wouldn’t necessarily be obvious from looking at individual records.
Cloudera plans to combine Neo4j’s technology with its existing data platform, giving AI agents access to relationships across information stored in different environments.
According to Sudhir Hasbe, Neo4j’s president and chief product officer, combining the company’s Graph Intelligence Platform with Cloudera’s Open Data Lakehouse will allow businesses to query large datasets and identify relationships across their information.
It’s a potentially helpful addition, although the quality of those relationships depends on the underlying data. A knowledge graph won’t automatically correct inaccurate customer records or resolve conflicting information across different systems.
Cartesia addresses another part of the experience, allowing businesses to build AI agents that interact with people through speech.
Its technology supports speech generation and transcription, which could be used for customer service, internal support systems or other applications where employees and customers speak to an AI assistant instead of typing.
Cartesia claims its speech-generation technology can respond with latency below 90 milliseconds. Its Sonic-3.6 model also ranked first on both Artificial Analysis speech arena leaderboards at launch, according to Cloudera.
That 90-millisecond figure refers specifically to speech generation, though, rather than the time it takes an entire AI system to understand a question, retrieve information and produce an answer.
Cloudera says the partnership will allow businesses to deploy these voice capabilities in customer-controlled environments, which could appeal to organisations that handle confidential conversations.
There’s an obvious application for banks, healthcare providers and insurers, where a voice assistant might need to access sensitive information while responding to a customer.
But keeping a system on infrastructure controlled by an organisation doesn’t automatically make those interactions secure. Businesses would still need appropriate access controls, auditing and safeguards against agents disclosing information they shouldn’t.
Mistral brings the AI models
Cloudera’s partnership with French AI company Mistral was announced in September, so it isn’t an entirely new arrangement.
The collaboration is intended to let businesses build, customise and deploy AI models using their own information, with the option of running those models in environments they control.
Mistral’s portfolio includes its announced Large 4 model, with 1.05 trillion parameters, alongside more specialised systems such as Timestral for forecasting and Robostral for robotics-related applications.
The number of parameters in a model can indicate its size, but it doesn’t necessarily tell you how well that model will perform a particular task. A smaller, specialised model may be better suited to a business application, particularly when processing costs and response times matter.
Cloudera wants customers to be able to choose models according to those requirements, rather than commit to a single provider for every application.
That flexibility could be important as companies begin deploying more AI agents. A business might use one model to extract information from documents, another to analyse financial data and a separate speech model for customer interactions.
The challenge is getting those systems to work together reliably while controlling what information each one can access.
Mistral’s involvement also supports Cloudera’s focus on sovereign AI, where organisations have greater control over the infrastructure and data used to run their systems.
For South African companies, that may be appealing when dealing with personal information, commercially sensitive records or internal systems that can’t easily be moved into public cloud environments.
However, data sovereignty is more complicated than choosing where to host a model. Organisations still need to consider how information is processed, who can access it and whether the infrastructure meets their legal and operational requirements.
VAST Data tackles the infrastructure problem
The fourth company involved is VAST Data, which specialises in the storage and processing infrastructure used by large AI systems.
Its previously announced partnership with Cloudera is intended to improve the movement of information between storage systems and the GPUs responsible for running AI workloads.
That might sound like a fairly ordinary infrastructure improvement, but it can make a considerable difference when companies are spending large amounts of money on expensive AI hardware.
GPUs need a constant supply of data to work efficiently. If a storage system can’t deliver that information quickly enough, the processors can spend time waiting instead of doing useful work.
The industry sometimes calls this GPU starvation, and it’s one of the problems VAST Data is trying to address.
Cloudera says combining VAST’s storage architecture with its containerised data services should improve GPU utilisation and the performance of AI workloads across different environments.
There are potential financial benefits here, particularly for businesses that run their own AI infrastructure. Better utilisation could mean getting more work out of existing hardware rather than having to buy additional processors.
However, Cloudera hasn’t provided independent performance results showing how much the integration improves processing speeds or reduces costs in real-world deployments.
That makes it difficult to judge the financial advantage of the partnership, particularly when businesses would also have to consider integration, licensing and infrastructure costs.
What does this mean for South African businesses?
Cloudera’s announcement comes as South Africa’s AI infrastructure market is attracting more investment, with businesses increasingly considering whether they should run sensitive AI workloads locally.
But there’s another complication. Running AI infrastructure in South Africa means dealing with electricity availability, cooling requirements and the cost of keeping systems operational.
As explored in the growing pressure on South Africa’s electricity grid from AI infrastructure, bringing computing workloads back into the country could solve certain problems while creating others.
Cloudera’s approach offers businesses the option of distributing workloads between public cloud services, private infrastructure and their own data centres.
That could allow an organisation to keep sensitive information within a controlled environment while using external computing resources for other tasks.
It’s an appealing proposition, especially for companies that already operate complicated IT systems across several environments. But achieving that flexibility isn’t necessarily straightforward.
Moving AI workloads between platforms can involve differences in networking, storage, security and software compatibility. An application designed to work with one provider’s proprietary services may not run elsewhere without considerable changes.
Cloudera says its expanded ecosystem will provide joint solutions and validated reference architectures to reduce the work involved in bringing these systems together. It hasn’t, however, announced specific South African deployments arising from the four partnerships.
There’s also the question of how much autonomy businesses should give their AI agents.
Allowing an AI system to retrieve information is relatively easy to control compared with allowing it to make changes to customer records, approve transactions or initiate other business processes.
The consequences of a mistake become considerably more serious when the software can act on the information it retrieves.
Companies would need to decide which actions require human approval, how decisions are recorded and what happens when an agent produces an incorrect result.
Those are questions that remain regardless of which models or infrastructure a business chooses.
Cloudera’s expanded ecosystem gives companies more options for building these systems, particularly if they want to retain control over their data and avoid depending on a single cloud provider.
Whether the approach makes enterprise AI easier or cheaper to deploy will depend on how well the individual technologies work together.
For now, the company has outlined the partnerships and their intended roles, but hasn’t disclosed pricing for the combined solutions, a detailed rollout schedule or independently verified savings.