Two Years In: How Dell and Starburst are Removing the Data Bottleneck for Enterprise AI

• Enterprise AI is failing at the data layer, not the model layer — over 40% of agentic AI projects are projected to be canceled by the end of 2027 due to inadequate data foundations.
• The Dell AI Data Platform, powered by Starburst, delivers governed, federated data access at inference speed — without centralizing or duplicating data.
• At DTW 2026, we announced up to 6x faster query performance on NVIDIA Blackwell GPUs, 3x faster throughput on NVIDIA Vera, and 12x faster vector indexing — all targeted for GA in 2026–2027.

The data problem that’s quietly killing AI projects

Enterprises in 2026 are not struggling to build AI agents. They’re struggling to give those agents governed access to the data that informs every answer.

The evidence is hard to ignore. As Info-Tech’s CIO Priorities 2026 puts it, “AI ambitions are outpacing the data foundations needed to support them.” Gartner reinforces the point, projecting that more than 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value and inadequate risk controls. In most of those failures, the model works fine. The data foundation does not.

That is the exact problem Dell Technologies and Starburst set out to solve when we launched our partnership in 2024, with the Dell Data Lakehouse reaching general availability that spring.

The Bet We Made in 2024 — and Why It Was the Right One

In 2024, the dominant AI narrative was still about model size and GPU counts. The real problem was data sprawl: siloed datasets scattered across on-premises systems, multiple clouds, and the edge. Strict sovereignty rules made a simple cloud migration a non-starter. Yet the industry treated all of this as an infrastructure afterthought.

We took a different approach. Dell and Starburst co-engineered a deeply integrated solution that is rigorously validated and built for the production realities of enterprise IT — a platform where data doesn’t have to move to be useful.    

Today, that bet is showing up in production: CTBC Bank is detecting fraud in 0.03 seconds, Dell is modernizing petabyte-scale analytics in-house, and NTT DATA is helping enterprise clients process massive data volumes in minutes instead of hours.

Two years of joint innovation

Over the last two years, the rapid pace of joint development has delivered the platform enterprises need for the agentic era. Each milestone built on the last — moving from federated lakehouse to a GPU-accelerated data layer purpose-built for AI agents:

Dell Data Lakehouse launch in 2024: Dell Data Lakehouse reaches general availability with Starburst-powered federated query capabilities. 

DTW 2025: Dell and Starburst deepen platform integration for enterprise-scale AI and analytics

DTW 2026: Dell AI Data Platform adds GPU-accelerated data processing and agentic AI capabilities.

What we shipped at Dell Technologies World 2026

An anniversary is only worth marking if it ships something. At DTW 2026, the Dell AI Data Platform, powered by Starburst, moved from a fast federated query layer to a GPU-accelerated data foundation built for agentic AI. Here’s what that means in numbers:

Availability and performance claims are based on Dell internal testing and/or partner-published benchmarks; final GA timelines are subject to change.

While competing platforms require data to be centralized into a proprietary warehouse before it can power AI, this architecture delivers governed, federated access at inference speed — directly where the data already lives.

The independent validation backs this up. A GigaOm benchmark found the Dell Data Analytics Engine delivers 3x faster time to insight at roughly half the cost of comparable technologies. A separate Enterprise Strategy Group study found Starburst’s platform delivered a 45% lower three-year total cost of ownership and a 414% three-year ROI versus alternative platforms.

At Dell Technologies World 2026, Michael Dell put the data foundation at the center of Dell’s AI story, and NVIDIA CEO Jensen Huang echoed the same message at GTC Taiwan 2026.

What two years of co-engineering actually built

The Dell AI Data Platform is built on a simple principle: no single layer should try to do everything, and each layer should evolve on its own cadence — storage, compute, processing, query and governance, each moving independently.

A production-grade data processing engine built for real enterprises

One of the most consequential things Dell and Starburst built together is an enterprise-grade Data Processing Engine based on Apache Spark, accelerated by the NVIDIA RAPIDS library. It’s designed for the reality of enterprise production environments: secure by design, built to meet strict data sovereignty requirements, integrated with fine-grained access controls, and tuned for the performance demands of large-scale analytics and AI data pipelines.

The combination of Trino and Spark engines is what makes it transformative:

1. Trino delivers blazing-fast, federated SQL queries across heterogeneous data sources — without moving data
2. Spark accelerates the heavy lifting of large-scale data transformation, ML feature engineering, and AI pipeline preparation
3. Together, they cover the full analytics spectrum: from sub-second ad-hoc queries to petabyte-scale compute-intensive workloads

From query engine to the control plane for agentic AI

The shape of data consumption has fundamentally changed. For two decades, analysts wrote SQL and people read dashboards. Now humans and AI agents work through the same enterprise workflows — agents issuing SQL programmatically inside multi-step reasoning loops. They consume governed data, decide and act.

That shift turns the query engine into the control plane for AI: the point where governance, cost control and access decisions are enforced for every agent in the enterprise. It’s also where Dell’s built-in agentic experience, with Starburst’s AI Data Assistant (AIDA), lets any user reach governed enterprise data through natural language conversation — no SQL required, no data movement needed.

Real deployments. Real results.

The best validation of any platform is the work it enables. Over the past two years, the Dell AI Data Platform has been deployed globally across financial services, manufacturing, telecommunications and the public sector. Three deployments tell the story.

How CTBC Bank cut fraud detection time to 0.03 seconds

CTBC Bank, one of Taiwan’s largest financial institutions, built a sophisticated AI-powered fraud prevention capability that demonstrates what becomes possible when structured data is treated as the ground truth for AI.

Using the Dell AI Data Platform, CTBC processes every transaction through a database of over 400 risk factors — detecting suspicious transactions with multi-modal detection, entity-relationship linking and real-time scoring. The result:

• Fraud detection response time of 0.03 seconds — protecting client assets in real time
• 40% reduction in fraud losses year over year between 2024 and 2025
• Expanded real-time AI detection beyond credit cards to all transaction types, with anti-money laundering pilots now underway

How Dell modernized petabyte-scale analytics in its own house

If we ask enterprises to trust this platform for their most critical data challenges, we commit to running it ourselves. Our goal was straightforward: eliminate ETL pipelines by querying data in place — including a multi-PB data lake in Dell ObjectScale — and deliver sub-second response times for both API-driven and ad-hoc interactive queries.

What we observed exceeded the goal:

• 3x boost in query performance while gaining 30x compute efficiency and 12x memory efficiency
• Consistent performance at scale on datasets exceeding 200TB
• Dynamic filtering, cost-based optimization, caching, materialized views and query pushdown all contribute to lower latency and reduced data scanned per query

This internal deployment has become one of our most important engineering feedback loops, directly informing the production hardening, security model and operational tooling our customers rely on.

FREQUENTLY ASKED QUESTIONS

Why do enterprise AI projects stall at the data layer?

Many AI projects stall not because the models are inadequate, but because they cannot access the right data, at the right time, with the right governance. Data sprawl across on-premises systems, multiple clouds, and the edge — combined with sovereignty, compliance, and security requirements — makes centralizing data into a single warehouse impractical for many enterprises. The Dell AI Data Platform addresses this with enabling federated access, allowing AI agents to query governed data where it lives.

How is the Dell AI Data Platform different from a traditional data warehouse?

Traditional warehouses require data to be ingested, copied, and centralized before it can be queried. The Dell AI Data Platform, powered by Starburst, enables governed, federated SQL queries across heterogeneous data sources in place — with the speed and controls required for AI applications. This can reduce costly ETL pipelines, limit data duplication and enforce governance at the query layer for both human users and AI agents.

What is AIDA, and how does it support agentic AI workflows?

AIDA (AI Data Assistant) is the platform’s built-in natural language experience for governed data access. For enterprises building agentic AI workflows, AIDA helps connect business users to the governed data layer while preserving the platform’s federated, no-dta-movement approach.

What performance improvements were announced at Dell Technologies World 2026? 

At DTW 2026, Dell and Starburst announced up to 6x faster query performance on NVIDIA Blackwell GPUs, approximately 3x faster query throughput on NVIDIA Vera CPUs (targeted for GA in Q1 2027), and up to 12x faster vector indexing available in Q2 2026.

What the next two years look like

We’re more excited about the next two years than the last two. The convergence of agentic AI, GPU-accelerated data processing and federated access is creating a new enterprise operating model — one where AI agents work directly over governed enterprise data in real time.

We’re deepening the co-engineering investment in the Data Processing Engine, expanding the agentic layer through AIDA, and bringing full Blackwell and Vera acceleration to general availability in 2027.

Two years ago, we made a bet that the future of enterprise AI would be won or lost at the data layer. The market has validated what we built — and our customers have shown us what’s possible when the data layer is no longer the bottleneck.

Ready to build your AI data foundation?

Talk to a Dell Technologies specialist about the Dell AI Data Platform powered by Starburst — and learn how enterprises like CTBC Bank, NTT DATA and Dell itself are turning their data layer into a competitive advantage for agentic AI.

→ Explore the Dell AI Data Platform

→ Read the Starburst partnership overview

→ Download the ESG Economic Validation Report

References: Info-Tech CIO Priorities 2026 | Gartner Agentic AI Prediction | Dell Data Lakehouse GA Blog | DTW 2026 Newsroom | GigaOm TCO Benchmark | ESG Economic Validation | NAND Research AIDP Report