Acceldata CEO: AI Is Breaking The Cloud Centralization Model

Acceldata CEO Rohit Choudhary says the rise of AI agents is breaking the cloud centralization model and pushing enterprises beyond the lakehouse era. In an exclusive meeting with CRN, he explains why data must stay distributed, why costs are set to soar, and how Acceldata’s new xLake platform aims to respond.

As enterprises race to deploy AI across their operations, the underlying data architecture must change just as dramatically, according to Rohit Choudhary, co-founder and CEO of Acceldata.

Choudhary, in an exclusive conversation with CRN, laid out a sweeping vision for the next era of enterprise computing, one in which AI agents, not just humans, become the dominant users of data platforms, and in which businesses can no longer rely on centralized cloud architectures to support analytics and AI at scale.

“Operational systems are producing data, and third-party data partners are depositing data in different islands, locations, and infrastructure sources,†he said. “And not all that data can be moved into a central location. Therefore, the cloud centralization model is breaking.â€

[Related: The 40 Coolest Data Recovery /Observability /Resiliency Vendors: The 2026 Storage 100]

Choudhary argues that the traditional data lakehouse model is nearing its limits. Enterprise data now lives across multiple clouds, on-premises systems, and sovereign environments, while AI workloads increasingly require a mix of CPUs, GPUs, and other specialized infrastructure. At the same time, he said, enterprises face mounting financial pressure as AI-driven data access threatens to multiply cloud costs. In his view, that combination of technical sprawl and budget strain is forcing a rethink of how data is managed and where compute should run.

That thesis is at the heart of Acceldata’s xLake platform, which the company began unveiling in 2025 and which is officially being released this week as part of its broader push into agentic data management. Choudhary describes xLake as a response to a world where data can no longer be centralized and where enterprises need a software layer that can run analytical, machine learning, and AI workloads consistently across cloud, virtual private cloud, and on-premises environments.

“AI agents will have to run in an environment which is operable, governable, explainable, cost efficient, and reliable,†he said. “Unless you have those capabilities built into the platform for this multi-technology, multi-cloud world, it’s impossible to run agents reliably, because somebody has to take accountability for the outcomes that these agents will produce in the future.â€

There’s a lot going on with Acceldata and how to manage data in an increasingly AI-focused world. What follows is CRN’s complete conversation with Choudhary, which has been lightly edited for clarity.

How do you define Acceldata?

It’s the next-generation data and AI platform which is built for running enterprise data which is primarily analytical machine learning workloads, but also AI workloads, because I think a couple of things are going to happen. The number of agents that are going to come in production will be probably 100x more than the number of humans who are accessing these data platforms, and we’re building the next generation of the current evolution of the data platform.

You said there will be 100 times more agents than humans accessing those data platforms?

Yeah, because every person who is accessing databases today is going to have at least 10 more assistants or agents that they will themselves create, or somebody will create for them, and those agents will then be working on our behalf. That’s the new world of agentic AI. We’ll do things by ourselves, but we’ll also have a lot of assistants doing things for us.

What does Acceldata do to prepare businesses for the agentic world that differentiates from your competitors?

What is really happening is that the centralization model that the cloud tried to create in the last decade is changing. AI is actually not waiting for all the enterprise data to be centralized. Data has to be computed and processed wherever it is, and in most of the cases, what you end up finding is that the Fortune 500 and the Global 2000 generally have multi-technology, multi-cloud setups. What it essentially means is that the operational systems are producing data, and third-party data partners are depositing data in different islands, locations, and infrastructure sources. And not all that data can be moved into a central location. Therefore, the cloud centralization model is breaking.

The second thing that’s happening is that people will have to deal with an xPU architecture, or an architecture which includes CPUs and GPUs, and potentially ASICs in the future.

The third thing that’s happening with that flexibility of architecture is flexibility of models. Enterprises will choose different models to do different things on different infrastructures.

The fourth thing that’s happening is, just imagine a world in which data was going to get accessed 100x more than the current utilization. The cost of the cloud is going to break the bank and break the IT budget from a public shareholder perspective, meaning the EPS (earnings per share) will start going down. Therefore, what will end up happening is customers will have to figure out a method and a way to go and process all of this data for analytical and AI purposes very differently.

What did Acceldata do to address this?

What we’ve come up with is called the xLake architecture. The xLake is essentially a fundamental acknowledgement of several facts. Number one, data will be in different places. Number two, data will have to be processed with different models, different processing frameworks. Number three, enterprises will have to proactively reduce their operational expenses that they are paying on the cloud. And when you put all of that together, it essentially signals the end of the lakehouse era, which was kind of centralizing all of that data, processing it all together. That’s not going to happen in the future, which is why this is the right time for a platform like ours to come into the market. AI agents will have to run in an environment which is operable, governable, explainable, cost efficient, and reliable. Unless you have those capabilities built into the platform for this multi-technology, multi-cloud world, it’s impossible to run agents reliably, because somebody has to take accountability for the outcomes that these agents will produce in the future, especially as it pertains to critical things like data analytics, machine learning, all the things that allow people to make business decisions.

Acceldata has a little bit of news. What’s going on?

We’re just launching the xLake platform this week. We’ve already worked with a bunch of customers in the last couple of years, and now we now feel this is the best time for the xLake platform to come into the world as companies are thinking about deploying hundreds of agents in the next couple of years. We find it very interesting that consumer agents and consumer AI have taken off. Prosumer AI has taken off. But enterprise AI is slightly behind because, as we already talked about, data is in different silos. There’s no uniform, interoperable mechanism for computing all that data. And the right guardrails, explainability, observability, and identity is not yet uniform. So we’re pretty excited about this, because as anybody who’s following the news knows, the build time for applications is completely disrupted. But somebody still has to run it in a way that the company can stand behind it and say, ‘Look, I’m taking responsibility and accountability for the outcomes that both these agents and humans will take.†The human accountability is already in the supply chain. Companies have contracts, and will stand behind them, but there isn’t enough of a guardrail for how agents should run. So the run time is also an important factor in our recent news.

Prior to the xLake platform, how did Acceldata do this?

We’ve always been sort of hybrid in our approach and behavior, unlike many companies which are either completely on the cloud or completely on-premises. We always had this as the architecture. What we found out was that the applicability of the architecture is much wider in the last couple of years, especially post the ChatGPT era, because the more you think about the situation, you find out that data is growing exponentially, and that data growth cannot be centralized for all the use cases. Also, as far as AI use cases are concerned, there is massive concern around security and privacy, and companies are not very comfortable in pushing and moving all of that data into a proprietary cloud data store and then sending all of that information into a model that they don’t have complete control over. We’ve always had the architecture to do this, and the market is now recognizing that this is the architecture that they needed. Needless to say, the cloud builds from both ISVs and hyperscalers are getting to a place where CFOs are taking a very keen interest in trying to bring the OpEx down.

The storage industry has a lot of companies talking about bringing AI capabilities, the ability to use data with AI apps across multiple clouds, and so on. How is Acceldata different? What’s your secret sauce here?

There’s a couple of things, and the distinction is really important. The storage industry itself has undergone quite a bit of change from the traditional storage of data to what they’ve eventually now become, whether it’s newer companies like Vast Data and Everpure, or whether you look at more traditional companies like NetApp and IBM storage, everybody has started to provide object store. The reason that cloud became extremely attractive for people was because of two things: it provided the elasticity for on-demand compute, and object store, which meant that you didn’t have to learn how to store a video from a file. The object store took care of that. That same technology, about five or seven years ago, started appearing on the horizon in on-premises technologies as well. And what also happened in parallel, because of Google open sourcing Kubernetes about 10 or 12 years ago, that ecosystem has matured. So the benefits of the cloud are now available to the on-premises, hybrid, and VPC (virtual private cloud) customers. So you can now compute your data uniformly, whether you’re storing it on the cloud, on hyperscaler storage, on a VPC, on a commercial storage provider which has object store, or in your traditional ways.

So what we do is provide the compute layer on top. We don’t provide the storage, but we give the interoperability of the software stack so you can run your analytical machine learning computation and AI agents across the cloud, on premise, and in VPC scenarios, regardless of what your storage format is. Obviously this is moment of transition to object store, which will take some time. Enterprises take some time to move or make these big shifts. We are probably the only computer platform which can interoperate across all these three different environments, let alone what you are doing in each of these. So even when you’re on cloud, your operational databases may be on Azure whereas your analytical databases may be on Google. But how do you compute all of that uniformly without causing software drift?

You use the word ‘compute’ to describe Acceldata. Is Acceldata considered within the storage industry? Sometimes it’s a little gray area in terms of what exactly is storage versus compute, but how do you classify your company?

I think we should be looked at in the same light as Databricks and Snowflake, which essentially are computational platforms, which allow you to process massive amounts of data. As a company, we are also probably the second or third largest contributor to open-source technologies. Those technologies are related to aspects of data ingestion and acquisition, data storage, and data processing, but we are on the software layer. We’re not on the storage layer.

So Acceldata is not migrating storage or storing data. Instead, it’s moving the compute to the data.

Correct.

Is Acceldata actually providing the compute technology part of it, or are customers using it with, say, Amazon’s or Google’s compute capabilities?

Today, all the enterprises and other player teams have a lot of choice. I think the choice that enterprises are trying to make is to have uniform computational capabilities and a uniform data stack across their different environments. So the problem is one of manageability. While you’re on Google for analytical purposes, you’re computing your data in a certain way. But you’re unable to retain that same method, the same software, the same code, when you’re processing similar data across your on-premise environment or your hybrid environment or in another country. For example, a multinational corporation is computing similar data for customers in different countries. But because of sovereignty requirements, they cannot actually bring that data back into the cloud. Now what you’ll see is more and more data islands will start forming, and in order for customers to manage that uniformly, they’ll still need an interoperable data stack, and that is what we provide today.

When was Acceldata founded?

In 2018. We’re a private company with some really good investors. We are in U.S., Canada, U.K., and India, with sales across the world. We’ve raised

over $100 million.

Are you profitable yet?

We don’t talk about that.

Fair enough. What are Acceldata’s primary channels to market?

We have a lot of direct sales, but we also have some amazing partners, both in the storage arena and the hyperscalers, and the latest and greatest ISVs, so we partner really nicely. In addition to that, a lot of consulting companies, global service providers or integrators, and regional service integrators take our products to the enterprises, because these are enterprise problems. We share a lot of accounts with many of these channels, so we’ve developed extensive channel partnerships for the last several years. We just this month unveiled a partnership with ServiceNow with their AI Control Tower initiatives, and we are working very closely with the ServiceNow customer base now.

Does Acceldata have a certification program for channel partnersâ€

Yeah, we have an extensive certification program, deal registration, which is the bread and butter of the channel, and also extensive training programs for all the customers and partners. So far, we have trained about 10,000 analysts, software engineers, and data practitioners across the industry in the last three to five years.

Is there anything else we need to know about Acceldata?

I think the most important thing is that this is a moment of displacement, where most of the enterprise leaders, along with their finance leaders, are starting to think and sharpen the pencils on what is the right amount they should spend, and obviously that has a consequential impact on infrastructure. I think what is going to happen is that enterprises are going to follow the hyperscalers in spending on infrastructure to make sure that they are able to satisfy their enterprise compute needs while keeping a very keen eye on what they want to spend in the next decade. I think both of those will be tremendous factors, and therefore we are signaling, from what we’ve heard from all our enterprise customers, that the data lakehouse era is probably coming to a close.