May 2025 Summaries

Dell and Starburst are advancing enterprise AI by providing innovative data solutions that enhance data accessibility and governance across various environments, including on-premises, cloud, and hybrid systems. At the forefront is the Dell Data Lakehouse, a part of Dell's AI Data Platform, which integrates cutting-edge features like native vector search, large language model functions, and automated Iceberg table management to empower organizations in building agentic AI strategies. The platform's advancements were highlighted at Dell Tech World 2025, where Dell introduced enhancements such as native vector embedding, hybrid search, and automated maintenance, which simplify AI workflows and increase accessibility for non-technical users. Additionally, Dell's collaboration with NVIDIA enhances the platform's performance through the use of NVIDIA RAPIDS and Spark, accelerating AI and analytics workflows while reducing costs. Infrastructure improvements, like the introduction of Project Lightning and innovations in Dell ObjectScale, further support large-scale AI deployments by delivering high throughput and reducing latency. These developments, alongside the integration of NVIDIA's networking components, enable organizations to effectively manage and scale their AI initiatives, reflecting a shared vision of unlocking the full potential of data for analytics and AI applications.

In the evolving landscape of enterprise technology, AI agents are poised to disrupt traditional SaaS applications by enabling organizations to leverage their own data more effectively, thus diminishing the dominance of SaaS giants like Salesforce and SAP. While SaaS applications thrived by capturing and controlling functional data, AI agents bypass traditional interfaces and can integrate data from multiple sources, facilitating autonomous workflows without user intervention. However, the effectiveness of AI agents is hindered by limited data access, prompting SaaS vendors to expand their data capabilities through strategic acquisitions and integrations. As AI agents become more efficient, doubling their task capacity approximately every seven months, organizations are encouraged to develop data-centric architectures that emphasize unified data access and vendor-neutral solutions. This shift empowers companies to assume control over their data, positioning them strategically to harness AI's potential while reducing reliance on application vendors.

Starburst's approach to enterprise AI emphasizes the importance of integrating AI with existing data architectures through their Lakeside AI strategy, built on the Icehouse architecture using Apache Iceberg and Trino. This method allows businesses to leverage their current data infrastructure for AI applications, avoiding the costly and time-consuming process of data migration. Starburst's new AI features, including the Starburst AI Agent and AI Workflows, aim to enhance data access, collaboration, and governance, enabling enterprises to build scalable AI applications with secure, reliable performance. The AI Agent provides a natural language interface for intuitive data exploration and task automation, while AI Workflows simplify the transition from AI experimentation to production by integrating vector search and SQL-based LLM functions without data movement. Additionally, Starburst's AI Model Access Management ensures secure and governed access to proprietary and third-party AI models. These innovations, alongside advanced data cataloging and management features, position Starburst as a leading platform for building an Open Hybrid Data Lakehouse, facilitating faster and more efficient AI adoption for enterprises.

Apache Spark and Starburst are complementary processing engines that, when combined, provide versatile options for managing data workloads, solutions, and applications, particularly in machine learning (ML) and artificial intelligence (AI) workflows. Spark's Python-centric approach is well-suited for tasks involving real-time streaming and machine learning, offering sophisticated frameworks for event processing and scalable ML algorithms. It is particularly adept at handling unstructured data through transformations and integrations with libraries like Unstructured.IO, which are crucial for preparing data for generative AI (GenAI) applications. In contrast, Starburst leverages a SQL-based strategy that excels in handling structured data and interactive querying, making it ideal for traditional data lake activities and generating embeddings from text chunks. Together, Spark and Starburst offer powerful tools for handling both structured and unstructured data within a unified data lakehouse architecture, enabling enterprises to effectively implement AI data strategies tailored to their specific needs.

As AI continues to revolutionize the business landscape, the importance of data access emerges as a critical factor in maximizing its potential across enterprises. The evolution from SaaS applications to agentic AI signifies a shift where AI agents, rather than traditional software interfaces, drive business efficiency by accessing data directly. This transformation is further supported by the enduring relevance of on-premises data, which, contrary to expectations, remains a vital asset due to its critical role in sectors like financial services and healthcare. Additionally, AI's growth accentuates the need for robust data governance, as organizations navigate the complexities of data ownership and compliance. Starburst positions itself as a leader in this evolving ecosystem by offering a unified data architecture that bridges the gap between cloud and on-premises environments, supporting both analytics and AI workloads through its Icehouse architecture and promoting the concept of Lakeside AI, which integrates AI capabilities with data lakehouses for enhanced data accessibility and insight generation.

Presto, originally developed at Facebook in 2012, was designed as a highly scalable SQL query engine to efficiently manage enormous data volumes, but it eventually evolved into Trino after its creators forked the project in 2019 to continue innovating independently. Trino retains the core principles of Presto, such as using Massively Parallel Processing (MPP) to query large, distributed data sets, but has outpaced Presto in development speed and features, offering enhancements like Docker-based deployments, fault-tolerant execution, and support for multiple open-table formats. Leveraging its open-source roots, Trino has been adopted by major companies like Netflix and LinkedIn and has become integral to various data platforms, including Amazon EMR and Starburst. Starburst, in particular, facilitates the deployment and management of Trino by providing enterprise-grade features and optimizations, such as the Icehouse architecture and Warp Speed technology, which enhance performance and reduce costs. Through its robust capabilities and integration with Starburst, Trino continues to lead the SQL query engine market, offering scalable, fast, and flexible data access solutions for organizations dealing with vast, complex data ecosystems.

The rise of AI presents significant opportunities for companies, necessitating the development of an AI data strategy that aligns with business needs and leverages the right data architecture. A successful AI data strategy integrates fast and flexible data access through technologies like the data lakehouse, which combines the analytics capabilities of a data warehouse with the flexibility of a data lake, forming a strong foundation for AI initiatives. This strategy involves three main pillars: ensuring data access across various platforms, fostering collaboration through data products that simplify data sharing, and implementing robust governance to maintain data quality and compliance. The Icehouse architecture, which combines the Iceberg open table format with the Trino SQL query engine, offers a scalable solution for handling AI data needs, supporting decentralized access while allowing for centralized processing of critical workloads. Starburst's Icehouse architecture enhances AI data strategies by providing a unified framework for data access, collaboration, and governance, integrating seamlessly with existing infrastructures to support evolving AI and analytics demands.

Data virtualization is crucial for managing data during experimentation with new systems and migration processes, as it provides a unified interface to access data spread across multiple systems, including Icehouses and legacy data warehouses. The two main types of data virtualization architectures are pull-based and push-based systems. Pull-based systems retrieve data from underlying systems to process locally, while push-based systems delegate as much data processing as possible to the underlying systems, reducing data transfer. Despite the efficiency of push-based systems in minimizing data transfer, pull-based systems are more common today due to their easier implementation and compatibility with modern, high-throughput networks. In scenarios where data from multiple systems needs to be accessed, data virtualization systems are indispensable, minimizing the complexity of data management and integration. However, performance considerations remain crucial, with pull-based systems potentially facing bottlenecks in data extraction and push-based systems encountering compatibility issues. For Icehouses, which utilize Iceberg tables and distributed storage, pull-based virtualization is the norm, as Iceberg lacks built-in query processing capabilities that push-based systems could utilize. This discussion underscores the importance of understanding the trade-offs and performance implications of using different data virtualization architectures, especially when dealing with complex data environments like Icehouses.