Acceldata xLake architecture offers compute, governance, observability (and quality)

End of data lakehouses?

The company asserts that the arrival of its Autonomous Data & AI Platform (caps left on to denote product status) signals the end of the data lakehouse era. But why?

According to Rohit Choudhary, founder and CEO, Acceldata… previously, enterprises had to prioritise migrating and centralising data, when AI agents will have to operate on distributed datasets across the enterprise. Enterprise AI adoption is therefore held hostage to an expensive and incomplete migration process. At the same time, companies have to consider where their data is stored for sovereignty and data privacy.

“The lakehouse architecture was built for human access. It broke in the agentic era,” said Choudhary. “We started Acceldata with the conviction that enterprise data would never consolidate, that hybrid would be the durable reality and that the data and AI platforms must evolve to support it. In Europe, data sovereignty mandates are accelerating this shift, making hybrid-native, jurisdiction-aware architectures a board-level imperative, not a future consideration.”

Acceldata is explaining how (in its view) the next era of data belongs to autonomous, hybrid-native, cross-lake (xLake) platforms. Acceldata core offering is an xLake compute platform where analytics and agents operate.

• The platform is hybrid by default, operates autonomously, routes workloads to the right infrastructure, augments data quality, optimises operational cost and enforces governance at machine speed. 
• This architecture is hybrid by default, agent-native by design and operationally autonomous, scaling governance and observability to thousands of agents across hundreds of data sources. 
• Enterprise agents will have access to richer context across the data supply chain to automate business processes and workflows with predictability.

New independent research conducted by GLG on behalf of Acceldata suggested that nearly 80 percent of enterprises had hybrid data operations. The study, which surveyed C-level executives at Fortune 1000 and Global 2000 companies, also found that over 40 percent reported governance fragmentation as the biggest challenge in running cross-platform data environments. 

Governance fragmentation 

The company points to study results which may suggest that governance is fracturing how firms operate their data i.e. governance fragmentation (40 percent) is the biggest challenge for companies running cross-platform data environments, where applying the right regulation and data management approaches consistently and accurately was a challenge. 

Alongside this, 20 percent of companies saw a challenge around data duplication, while 10 percent saw identity fragmentation and data lineage gaps as challenges.

Acceldata also thinks AI initiatives are stalling at the data layer. Supporting AI initiatives is the leading source of board-level pressure on data infrastructure (33 percent), but enterprises report AI/ML operationalization friction, AI/ML integration gaps and skills shortages as top pain points.

These issues show how enterprises are running an AI mandate on a data architecture that wasn’t designed for hybrid reality, agent-scale governance, or the cost discipline AI economics now demands. 

xLake platform

The xLake platform is built to offer the following:

• Petabyte-scale compute for enterprise analytics and AI in a hybrid native environment with automated routing to cost-efficient infrastructure locations
• A secure and governed runtime with autonomous identification of governance boundaries and data availability 
• Agentic runtime for business applications to solve problems across front, back and middle offices with secure access to all enterprise data

Deep dive, agentic deconstruction

The Computer Weekly Developer Network (CWDN) spoke to Mahesh Sharma, CMO at Acceldata for more.

CWDN: How does the data management side affect agentic AI?

Sharma: The following assumptions baked into traditional enterprise data management breaks under agentic AI – batch refresh cycles, human-designed dashboards and periodic BI workloads were designed for a slow, predictable, human consumer. 

Agents are the opposite. They’re high-frequency, non-predictable, multi-modal, and they consume context continuously rather than on a fixed query. When you point an agent at a stack that was architected for quarterly board decks and weekly reports, you get silent failure, decisions made on stale data, and missed expectations.

What agents actually need is three things that traditional data management cannot deliver: runtime governance – policy that travels with the data as the agent uses it, not policy stapled on at the catalog layer; continuous observability –  knowing in real time whether the data an agent is reasoning over is fresh, lineage-verified, and within drift tolerance; and federated access across the heterogeneous estate enterprises actually have, not the centralized monolith vendors keep promising. That’s the architectural shift we announced at Autonomous 26 – the move from bolt-on data management to an Autonomous Data & AI Platform that operates at the speed of agents themselves.

Breaking assumptions

CWDN: What are the big assumptions that enterprises make around AI now, and how does that hold up teams from success?

Sharma: There are four assumptions that hold up teams from AI success.

Assumption one: AI ROI is a model problem. It isn’t. Model costs are collapsing – the unit economics of AI is deflationary. The variable that actually determines AI ROI is what we call the Data Efficiency Ratio: how much trusted, governed, contextual data you can deliver per dollar of infrastructure spend.

Assumption two: We can centralize our way out of this. The lakehouse era was built on the premise that if you just consolidate everything into one platform, AI gets easy. That premise is broken in an agentic world. Real enterprises run on heterogeneous estates – multiple clouds, multiple warehouses, lakehouses, operational systems, SaaS data, edge data. The migration costs of centralization now exceed the value. The winning architecture is federated, not centralized. This is why we said that the lakehouse era is ending.

Assumption three: We can bolt AI onto the existing data stack. Every layer of the modern data stack — ingestion, catalogue, quality, governance — was designed for human-paced, query-driven consumption.

Assumption four: Pilots will scale. There’s a big jump in infrastructure and governance requirements when going from pilots to production. Enterprises that don’t recognise the fundamental shifts needed will find that AI isn’t working or is too slow to take hold in their enterprise.