Cloudera recently showcased Anywhere Cloud at its EVOLVE event in Singapore. It puts Cloudera’s data and AI software onto the infrastructure that the customer already runs. That can be a public cloud, a private data center, hardware that must remain within one country’s borders, or a network cut off from the internet entirely. A customer runs all of it from one place, which Cloudera calls a single control plane. Services come from a marketplace, picked individually rather than installed as a whole platform.
Cloudera argues that enterprise AI has outgrown the public cloud-only model. The company has been making that case for years, and the market has been moving toward it. Cloudera’s own June survey covered 1,500 architecture and infrastructure leaders. Two-thirds had already moved AI workloads off public cloud and back to private cloud or on-premises. Only a quarter said they plan to make hybrid their main approach over the next two years. Buyers are bringing work home one job at a time without redrawing their architecture to match, and a platform sold in pieces suits that buyer exactly.
Modular Services Let Buyers Skip the Rest of the Platform
The Anywhere Cloud Marketplace works the way the name suggests. It stocks the engines Cloudera packages, meaning the software that does the actual data processing, including Spark, Kafka, and Trino. All three are open-source projects, and a customer can bring any other engine they prefer instead. They browse the marketplace, pick what they need, and deploy it into whichever environment they choose. Cloudera says deploying off a pre-approved blueprint takes minutes. Alongside its own engines, the marketplace carries partner engines Cloudera has certified. PuppyGraph for graph queries and a range of vector databases are among them, and an SDK lets a customer add one that isn’t there. Cloudera claims write-once portability, so a customer builds once and runs it anywhere.
A team that needs streaming and nothing else can deploy streaming and nothing else. Most will take several services, each one deployed on its own. Committing to a data platform has usually meant committing to its engines too, and Anywhere Cloud drops that requirement. Platform decisions are among the largest and slowest purchases in enterprise IT, and breaking one into service-sized pieces makes both entering and leaving cheaper. Portability is worth more on the second decision than the first. A team picks an environment now, and what matters is the bill for moving to a different one a year later.
Cloudera Put Its Governance Work into an Open Catalog
Two open pieces sit under Anywhere Cloud. Apache Iceberg is a table format, the agreed-upon way to store a table so that different engines can all read it. Apache Polaris is a catalog, an index that records which tables exist and who is allowed to read them. Cloudera runs both and publishes shared APIs, so an engine from another vendor can connect without custom integration work.
Cloudera adopted Polaris in June and gave something back. It contributed a plugin, now in beta in Polaris 1.5, that connects Polaris to Apache Ranger. Ranger is the open-source software that Cloudera’s security has run on for years, and it holds the rules about who can see what. Routing those rules through the open catalog means the table remains under the same permissions, regardless of which engine accesses it. Openness is where Cloudera is strongest, and the Polaris work is a concrete example of it.
Cloudera says that security rules, audit records, and lineage stay in one place across all environments. Lineage is the trail showing where a piece of data came from and what happened to it along the way, which is what an auditor asks for. Cloudera also says that workloads move between environments without copying the underlying data. Cloudera names ExxonMobil, IQVIA, and Mastercard as customer design partners, meaning they worked alongside Cloudera while it built the product.
Cloudera made governance the foundation. Each new environment and engine adds another way for access to drift, but the rules within Ranger move with the catalog, so a buyer can trust what comes back when an engine queries the data.
The Copilot Points at Operations and the Agent Question Stays Open
Cloudera says Anywhere Cloud includes an agentic copilot that turns a plain-language request into either an automated data workflow or an infrastructure change. Asking a database a question in English is everywhere now. Pointing a copilot at setting up systems and keeping them running is more interesting to me. Running a mix of clouds and data centers is an operations problem long before it becomes an analytics one.
An agent can’t work from rows and joins alone, and most of the market has tried to fix that by searching through documents. Reading a graph straight off the governed tables gives an agent the entities and relationships instead. The data is already there, and there is no second system to buy and secure.
Cloudera calls the architecture agent-first. With the copilot serving as an interface for executing automations, the assumption would be that it hands the work off to something that runs it. I’d want Cloudera to walk through where agents actually execute within the platform and how they remain subject to the same permission rules while running.
GPU Acceleration for Spark Takes on the Cost of Scarce Hardware
At EVOLVE Singapore, Cloudera and NVIDIA said GPU acceleration for Apache Spark 4.1 will arrive in Cloudera Data Engineering as part of Anywhere Cloud. Spark does the heavy data processing behind most analytics and AI work. Running it on GPUs rather than ordinary processors is meant to make the same jobs finish faster. Cloudera puts that at up to four times. Cloudera says existing code runs unchanged, down to not touching a line of PySpark or SQL, and that decides whether a project like this ever happens. Rewriting a working pipeline to chase a speedup is the cost that kills the idea before anyone starts.
GPUs also connect to something I’ve been tracking this year, and that’s the fact that companies can’t get their hands on GPUs on the schedule they want them. So teams start where capacity is available and move on-premises once their own hardware arrives. That move is the second decision I mentioned earlier, and portability determines whether it’s a redeploy or a rewrite. That looks to me like the reason those workloads moved back, though the survey measured the movement rather than the cause. If you can’t buy more GPUs, the remaining option is getting more work out of the ones you already have.
Anywhere Cloud Fits the Buyer the Market Is Producing Right Now
Regulated and sovereign deployments have been Cloudera’s ground for years. The company held that position through a stretch when most of the market said data belonged in a public cloud. Nearly everyone sells into that ground now, and for most of them, it means offering a second place to run the same product. Cloudera is selling something a customer assembles in pieces, with one set of rules across all of it. Opening a data center in another country doesn’t answer that. Specifically on regulated and sovereign ground, Anywhere Cloud is the strongest position Cloudera has held against cloud-native competitors.
Existing customers will want to know whether moving to Anywhere Cloud is an upgrade or a migration. Those are very different conversations for a bank running Cloudera today. Cloudera has committed to supporting existing installations through 2032 on version 7.3.2, so nobody has to rush. Spelling out the path would still help. Pricing is the second open question, since deploying services separately raises the question of whether to buy them separately.
Cloudera has argued for years that serious enterprise data would stay where it is. Most of the market disagreed then and has since come around. Anywhere Cloud arrives with that argument already won.
