AI market is ‘frothy not bubbly’, says Cloudian CEO

The artificial intelligence (AI) market is “frothy” rather than in a bubble, according to Cloudian CEO Michael Tso, who argued “there’s a mountain underneath this thing”.

Most of the investments in AI so far has gone into training large language models (LLMs). It costs around $10bn to train a foundation model, he observed, and this is an ongoing effort. However, the real money is to be made from inference.

“There are probably just 25 companies in the world that can afford the cost of training, but inference fits within most organisations’ budgets,” said Tso.

Applying LLMs to an organisation’s own data has been a complicated task, he added, because foundation models lack specific knowledge. To solve this, organisations need an easy-to-deploy package that can combine generic and in-house data.

This is where Cloudian’s new AI data platform comes in. Early releases are already with select customers and, according to Tso, “people are understanding it very quickly”.

By building AI into the storage platform, the models can process and understand the data stored within it. For example, Cloudian stored its own user manual on the system, which could then respond correctly to questions such as, “How do I set up a firewall?”. Tso noted that the process also respects any existing access controls on the data.

Cloudian’s partnership with Nvidia “is very natural”, he said. Although the graphics processing unit (GPU) giant has open-sourced some of its models, it does not want to sell to or directly support the enterprise customers that want to use them. Cloudian is first to market with this on-premises integration concept, he claimed.

The platform provides a path for organisations that do not want to run AI workloads in public clouds. The developers of large AI models are generally circumspect about where their training data comes from, so “people have to take care of their data, or there’s no limit to where it will go”.

“No company wants to volunteer their data,” said Tso, adding that many enterprises are hesitant to spend on cloud-based AI because they do not want to risk exposing proprietary information. Some firms, such as Samsung, have gone as far as blocking employee access to all public AI chatbots.

In the Asia-Pacific region, Cloudian is working in partnership with sovereign AI provider ResetData to offer AI infrastructure as a service with physically separated servers, in preparation for an expected increase in Australian AI spending in 2026.

“The key to our approach is that it is your data,” stored and processed in an environment that provides sovereignty and security,” said James Wright, Cloudian managing director for Asia-Pacific and Japan, calling it a compelling value proposition for prospective customers.

Tso believes these security and sovereignty considerations are likely to lead to a resurgence in datacentre activity, whether in private facilities or in spaces provided by large datacentre co-location operators.

While there is significant enterprise interest in AI, Tso suggested the technology is still waiting for its “Lotus 1-2-3 moment” – a killer application that cements its value. “Once you have a use case like that, it kills everything,” he said.

That moment, he noted, could happen when all of an organisation’s data is securely available to an AI – a feat he claims Cloudian’s platform makes possible.