Interview: Cloudera says open-ended AI spending is over

Cloudera says large enterprises are becoming more disciplined about artificial intelligence spending, demanding measurable business outcomes, clearer deployment plans and lower infrastructure costs as the company expands its private and sovereign AI strategy.

"The experimentation without regard to cost is over because people were surprised. Many companies were surprised by how quickly they could consume compute," said Charles Sansbury, Chief Executive Officer, Cloudera.

The shift means deployment architecture and cost are entering discussions earlier. Enterprises are considering where workloads will run, the computing resources they will require and whether projects can deliver sufficient value before committing to wider production deployments.

Cloudera is responding with a greater focus on business value and return on investment, alongside technology and services designed to run AI applications across private, public and hybrid environments.

Cost discipline

Past experimentation sometimes left companies spending beyond their AI budgets and reallocating money from elsewhere to cover the additional costs.

"So companies overspent their budgets, and they basically went around door to door and scraped money out of discretionary marketing, other IT projects that could be slowed down, discretionary hiring. When they overspent, they went all around the entire organisation and scraped dollars to try to cover that overage. But it was a very short and very unprecedented window of 'go spend what you want'," added Sansbury.

ROI is now entering initial customer discussions rather than appearing later in a deployment. That marks a change from earlier conversations in which speed to market could outweigh cost.

Technical decisions are becoming more complex too. Enterprises must decide whether applications should run in public clouds or on-premises, whether CPUs are sufficient or GPUs are required, and how data should be prepared and governed for production AI.

Poor data quality remains another barrier. Large organisations have often accumulated fragmented data estates through acquisitions and geographic expansion, leaving information spread across multiple systems and environments.

Cloudera is positioning Anywhere Cloud as a way to make distributed data available to analytics and AI workloads without requiring enterprises to move everything into a central repository.

"The data problem is a lot bigger than people realise because almost every company has kicked the can down the road in terms of data quality, data optimisation projects, and now it's coming back to bite them because that's the food that feeds quality AI models," added Sansbury.

Sovereign push

Sovereign and private AI are becoming larger parts of Cloudera's go-to-market strategy as governments and regulated industries place greater emphasis on control over data, infrastructure and workloads.

Cloudera accelerated its work in this area through its acquisition of Taikun, whose technology included a containerisation framework and control platform. The acquisition contributed to an architecture intended to let customers manage workloads, governance and security consistently across computing platforms.

The company sees the model as applicable to both private AI environments and sovereign cloud deployments.

Concerns about data sovereignty have increased particularly in Europe, the Middle East and Asia, driven by regulatory and geopolitical considerations. In most markets, customers are not necessarily demanding that every technology component originate domestically. The priority is retaining operational and governance control over the systems they use.

"We're willing to use those components, but we don't want some situation where they have a control option. Rather, they're going to sell us components, and we're going to have our team implement, manage, use, govern those. I think that's more of a governance question than it is a componentry question," added Sansbury.

China is an exception because of the breadth of its domestic technology ecosystem. Cloudera has large customers in the country, but domestic vendors are more likely to present credible competition there than in other markets.

Cloudera is also launching an AI marketplace that gives users access to technologies including query engines, graph databases and semantic layers. Partners are expected to support deployment across both private and public cloud environments, reflecting Cloudera's customer base in highly regulated industries.

"We do have an AI marketplace, as was shared on the Anywhere Cloud presentation, that gives practitioners access to use a specific tool: query engine, graph database, semantic layer, whichever capability they like. And we will continue to add as we go forward," said Abhas Ricky, Chief Business Officer and GM of Applied AI, Cloudera.

Open source will remain a foundation of Cloudera's platform. The company continues to support and contribute to dozens of open-source projects, while customers increasingly expect user interfaces, observability, security, governance and bespoke functionality on top of those underlying components.

Applied AI

Cloudera has formed an Applied AI organisation to help enterprise customers move projects from initial design into production. The team includes forward-deployed engineers and Applied AI Scientists who can work directly with customers on workflows and applications.

Its initial priorities include helping customers create enterprise AI applications in private environments and developing reusable blueprints, solution accelerators and industry-specific prototypes.

"There are three core things we would like to do and achieve. Number one, we would like to get as many customers as possible to build at least one, if not more, enterprise AI application in a private environment. That's the primary goal of the business, and that's how we're helping large enterprises," added Ricky.

The group will also make selected investments with generative AI partners, an area Cloudera views as requiring specialist skills distinct from its broader go-to-market organisation.

Demand for those resources is already coming from customers, although Cloudera has not set out a target headcount. Coverage will initially be allocated according to customer concentration, geography and industry.

Forward-deployed engineers will have a different remit from traditional customer success teams. They are expected to operate as full-stack machine learning engineers, running workshops, capturing requirements, building workflows and taking initial prototypes into production inside customer environments.

Customer Success teams will retain responsibility for the broader vendor relationship, identifying requirements and coordinating the appropriate internal resources.

Sansbury acknowledged that adding more technical roles to an account could create confusion unless responsibilities are clearly defined.

"If we have three different technical people showing up at an account, it's very important that we and they define their roles and what they do or what they don't do. The challenge is making sure that we have, both for ourselves and for the customers, articulated roles and responsibilities, because then the question becomes: if something goes wrong, who does the customer call?" added Sansbury.

The Customer Success function is expected to remain the main point of coordination, directing requests to specialists where needed.

Developer gains

Cloudera is also applying AI development tools internally to accelerate its product roadmap. The company has used tools including GitHub Copilot and Cursor, while developers have increasingly adopted Anthropic's coding technology.

The effect varies considerably between engineers. Strong developers can achieve substantial gains, while weaker developers can simply produce poor-quality code faster.

Cloudera has a development organisation of roughly 1,000 people, and Sansbury linked coding assistants to its ability to deliver more product work over the past year. He estimated that stronger developers were achieving productivity improvements of around 30% to 40% on average, with some reaching multiples of their previous output.

No large acquisition is currently seen as necessary to support the next 12 months of product development. Cloudera believes it has the technology required for its existing roadmap, although smaller acquisitions remain possible where buying a capability could accelerate development more effectively than building it internally.

The company is also examining longer-term opportunities in physical AI and edge inference. Its Data in Motion portfolio includes NiFi and MiNiFi, which can be deployed on devices, and Cloudera is exploring how those technologies could serve as runtimes for inference at the edge. Automotive, logistics and industrial customers could create additional use cases as robotics and industrial AI deployments grow.

"Majority of the companies who were doing IoT and IIoT will either shift to or have shifted to a strategy around physical AI, and it is in an early stage as a category. It is not as mature and hardened as generative AI has become in the last three to five years. But yes, that will provide an opportunity for us in the future," added Ricky.

"For some people, it's multiple times. My guess is, on average, we're probably getting 30 or 40% more productivity out of our really good developers. It's one of the reasons why we were able to deliver as much of the roadmap as we did: 12 months instead of 18 months," added Sansbury.