It’s been a monumental two years of partnership between Starburst and Dell. We are immensely proud of all that we’ve achieved together. And our work is far from done. If anything, it has only increased in importance. As Dell recently wrote, the data foundation for AI has never been more essential.
What brought Starburst and Dell together?
When we announced our partnership with Dell, we set out to solve the fundamental problem of enterprise data fragmentation. At the time, the industry was focused on the raw power of large language models, but we held a different conviction. We believed that the success of enterprise AI would be determined by the accessibility and integrity of the data architecture beneath those models.
Building the forefront of data foundations
Over the last two years, this relationship has moved from a strategic alliance to a deep co-engineering effort. Today, with the arrival of agentic AI, this partnership has moved to the next level.
Let’s look at what Starburst and Dell have created, where we’re heading next, and why we are actively solving one of the largest, most important bottlenecks in AI.
What Dell and Starburst created together
The Starburst-Dell partnership has always been based on large ambitions. Together, we worked to build a production-grade environment that solves the problem of data fragmentation without requiring massive data movement. We have proven that you can modernize without replatforming, delivering a solution that is rigorously validated for the production realities of enterprise IT.
Building the data foundation for AI
The foundation of our joint work is the Dell Data Analytics Engine powered by Starburst. This architecture decouples compute from storage, enabling organizations to query data directly in Dell PowerScale or ObjectScale environments. By using open standards like Apache Iceberg, we have created a future-proof foundation that eliminates vendor lock-in and provides the stability required for long-term AI strategies. This setup delivers up to 90 percent faster time to insight and a 53 percent lower total cost of ownership compared to traditional cloud migration paths.
Building the context layer for AI
Data without context is just noise for an AI agent. Our joint solution creates a unified context layer that applies common definitions and governance across distributed sources. By integrating NVIDIA cuDF and cuVS into the Dell AI Data Platform, we enable high-performance retrieval and reasoning pipelines. This ensures that when an AI system accesses data, it understands the business logic and the security constraints associated with that information, regardless of where the data physically resides.
Solving the data architectural bottleneck in AI
The primary bottleneck for AI today is no longer the model or the GPU but the latency of the data pipeline. We have solved this by providing high-performance federated access. By eliminating the need for mandatory ETL, we allow teams to move from data discovery to AI inference in a fraction of the time. This is critical for organizations moving beyond AI pilots into production-scale systems that can operate across hybrid and multi-cloud environments.
The Agentic Control plane
As we enter the era of autonomous agents, the need for governance has never been higher. The Agentic Control plane we have built with Dell provides the guardrails for AI-driven actions. It coordinates workflows and enforces policies at the engine level, ensuring that agents operate within the bounds of corporate reality. This is supported by the Dell Data Analytics Engine Agentic Layer and AIDA, which allow business users to interact with data through conversational natural language interfaces.
Why structured data matters more than ever
There is a common misconception that the rise of AI makes structured data less important. While AI is incredible at unlocking the value of unstructured text and images, that unstructured data requires a guide. Structured data provides the indispensable framework of truth that gives unstructured insights their meaning.
Why NVIDIA believes structured data is the ground truth of business
During the GTC 2026 keynote, NVIDIA CEO Jensen Huang noted that structured data “is the ground truth of business. The ground truth of AI.” This is a recognition that the core operations of the global economy are still recorded in the rows and tables of enterprise databases. You cannot run a supply chain or a financial institution on unstructured data alone. You need the precision of structured records to ground your AI in reality.
How Starburst and Dell unlock structured data
The Dell and Starburst partnership is designed to activate this ground truth. We provide the high-performance engine required to query petabyte-scale structured datasets across on-premises and cloud environments.
How structured data works alongside unstructured data
Our architecture enables seamless integration of both data types. You can use structured data to filter and govern the retrieval of unstructured context, such as searching billions of files on PowerScale using granular metadata. This creates a more reliable and accurate AI system, where structured ground truth serves as a validator of the probabilistic outputs from unstructured models.
Starburst and Dell provide breakthrough performance with GPU acceleration
The ultimate goal of our joint engineering is to ensure that the infrastructure never stands in the way of intelligence. In our recent release, we achieved a significant performance milestone that redefines the capabilities of the Dell AI Data Platform. For the first time, GPU execution has officially surpassed CPU performance on the full TPC-H benchmark suite.
This breakthrough is the result of compounding three major improvements: Hash Join on GPU, enhanced aggregation, and full support for decimal arithmetic. By offloading these complex operators to NVIDIA GPUs via the cuDF library, we are seeing speedups that were previously considered unattainable in a federated environment.
Unpacking the benchmark results
Our latest testing shows that the Dell and Starburst solution is now delivering massive performance gains across industry-standard benchmarks.
- TPC-H SF30: We have achieved a 3.2x average per-query speedup with total wall-clock time for the full suite improving by nearly 1.75x. Some complex queries are now running up to 8x faster than traditional CPU baselines.
- ClickBench: This web analytics benchmark hit a 3.4x overall speedup. In some cases, aggregation-heavy queries have seen an 81 percent reduction in execution time.
- Query Q13: A critical indicator of join performance, this query now runs 8.3x faster on GPU, proving that the bottleneck of large-scale data joining has been effectively neutralized.
Real-world price and performance impact
These gains are not just about raw speed; they are about the economic viability of AI infrastructure. Our analysis shows that because of these performance leaps, GPU-accelerated instances are now price-competitive with standard Graviton 4 workers. In fact, the G6 instance was found to be approximately 8 percent cheaper than the standard cloud worker when measured by time to insight.
The GPU operator roadmap
We are moving quickly to expand this acceleration across every part of the execution engine. Eight key operators are already complete and shipping in our master branch, including Table Scan, Filter, and Hash Join.
We are not only building for the benchmarks of today; we are building the high-performance engine for the agentic workloads of tomorrow.
How Starburst and Dell are just getting started
The last two years have been about laying the foundation, but the next two will be about accelerating the intelligence built on it. We are deepening our co-engineering investments to make AI agents even more intuitive and powerful on Dell infrastructure. We recently surpassed 100 million dollars in ARR for this partnership, proving that the market values a data strategy built on optionality and open standards.
Together, we are ensuring that the most demanding enterprises in the world have a clear and governed path to the agentic future.
