The 3 foundations of an AI data architecture

In my last blog, I showed how Starburst has become the foundation for an emerging Artificial Intelligence (AI) data architecture. I also talked about how this is a natural evolution for us that extends many of the same approaches that we take with data analytics, including our use of hybrid data lakehouses and Icehouse architecture. Both AI and analytics rely on a strong foundation, and that foundation is built using data architecture. 

What makes an AI data stack successful? 

When we think of a data architecture serving AI models, three pillars stand out. I think of these as the three pillars of the AI data stack. Whatever AI model you’re using, whatever AI agent, and whatever source the data has originated from, you will need each of these three things to be true to be successful. 

Starburst helps serve each of these, which is one reason it’s helping companies adopt a single foundation for all their data needs. Let’s look at these three pillars in more detail before turning to some examples of how our customers are implementing AI data architecture today using Starburst. 

 

1) Your AI architecture needs a single foundation for all your data

Need: A single foundation for all of your AI data

Inaccessible data is useless. That’s universally true, whether you’re talking about analytics or AI. Just like analytics, for AI models to thrive, they require a strong data architecture to support them. You need access to all of your data, wherever it lives, regardless of data structure. You also need the freedom to be able to move data in a way that makes sense for your business. The problem is that most businesses don’t have data in one place or in one structure, they have it in many places–in the cloud, on-premises, across multiple clouds, and in many structures. Overcoming this problem is not only difficult, it’s often impossible. Data silos in AI are no different from data silos in data analytics, and they hold back the ability to derive value from your data.  

Reality: Data silos build your AI on a shaky foundation

AI models suffer from the same data architecture problems that face data analytics. Big data can’t be big without data from across your organization. Similarly, AI models can’t learn from data they can’t reach. Meanwhile, delivering datasets to AI models in near-real time is not easy. Data is held in different formats across different technologies, some in the cloud and some on-premises. 

The familiar problem from analytics–vendor lock-in–is no less of a problem for AI. It’s simply a new frontier for your data stack and one that many businesses are struggling to deliver. This is made even worse when you consider that context is particularly important for AI. Contextual finetuning, in the form of RAG strategies, is the key to making general purpose models fit proprietary use cases. Data silos inhibit access to quality context, reducing your ability to fine-tune your dataset for a particular model. 

Solution: Build your AI model on the best data foundation

Starburst is a single foundation for all of your AI data. It allows you to feed AI models using an Icehouse architecture predicated on Apache Iceberg and Trino, just like we do with your analytics. Just like analytics, this foundation includes the ability to apply real choice to your technology, which is essential for fine-tuning your AI models or using a RAG architecture. And just like analytics, strong AI models rely on a firm foundation comprising a strong data stack. 

 

2) AI needs to be easy to use and solve real business problems

Need: AI needs to provide insights for your whole business

It might sound obvious, but no one sets out to implement an AI or analytics solution that doesn’t actually work for their business. 

In practice, this means several things. First, any solution has to be easy to use. If it’s too complex to be practical, it will struggle to attain value. Second, it needs to work with your existing infrastructure. No one is looking to replace all of their analytics overnight with AI. Instead, AI is an iterative extension of what your data can do for you.

How do you get there? This is where it gets interesting. There’s a technological aspect to the problem, but there’s also a business aspect. For AI to be useful for your business, both need to work together. Your technology can’t fight your business; it has to help it achieve its desired outcome. 

What you really need is something that brings your company together around this project. You need technology that’s more adaptable, fast, and collaborative.     

Reality: Making AI work across your entire business requires a plan 

For AI to transform your business, it takes a village. Data is generated by each and every one of your departments, but often not in the same way, using the same technology, or speaking the same language. 

With analytics, this can inhibit insights by limiting the size and context of the datasets in question, and it’s no different with AI. AI models need as much context as possible to provide value, which means that data collaboration is now more important than ever. In pursuing AI, businesses face risks on a return-on-investment basis. The money they spend on AI architecture needs to deliver real value, and the projects that underpin that initiative need to foster collaboration at a company level. 

Solution: Starburst lets you easily build an AI architecture for your whole organization

Starburst is built on choice. A good foundation for data architecture has to be robust and flexible. That’s why our technology is designed to connect to over 50 data sources, creating opportunities for collaboration across your whole organization. 

It’s also easy to use. This means that as your team works to pull together data sources, identify context, and feed this into an AI model, their energy and efforts are actually able to create results. As more and more organizations move to become AI-ready, having a technology that makes this process seamless, powerful, and easy makes the difference between success and failure. 

And it just so happens that what works for your AI architecture also works for your analytics data stack too. The same foundation can serve both use cases, so as you prepare for an AI-ready future, you can also optimize your analytics data stack too. 

 

3) For data to provide value, it has to be governed securely

Need: Effective AI is built on a foundation of strong data governance 

Without well-governed data, you don’t have anything. AI is exciting, and nearly every organization is planning to implement an AI adoption project in one way or another. But one thing that gives people pause is data governance. 

Data is sensitive. In many industries, it’s highly regulated and subject to regulatory and compliance requirements. All of this means that as you build out your AI capabilities, your ability to govern your data needs to grow in parallel. 

Reality: Strong data governance requires a technology that supports it 

For too many organizations, AI governance is fractured. The same exponential growth that promises to unlock revolutionary value also comes with the potential for compliance concerns. Organizations are faced with opportunity, but also risk. 

This is particularly true in high-compliance industries, like financial services, the public sector, or healthcare, but concerns around transparency, data protection and handling, intellectual property, data sovereignty, and access controls are largely universal across industries.  

All of this creates risk. 

Solution: Secure the foundation of your AI with strong data governance

Starburst is built to scale data governance as quickly as it scales data itself. This is already a focus for data analytics, where it creates a secure foundation for your data across multiple environments–cloud, on-premises, and hybrid–ensuring that the right people are able to access it in the right way.

The same is true of data governance for AI. Starburst lets you secure your AI data as it moves from source to model. Data products are integral in this regard. They provide the perfect AI governance layer, allowing you to create curated, governed collections of data that can feed AI models directly. 

Often this can make the difference between a project going forward, and one encountering difficulties. As more and more organizations move towards a hybrid model alongside their AI, this governance layer becomes even more critical. 

 

How our customers use Starburst for their AI data architecture

Importantly, this isn’t something that’s happening tomorrow–it’s happening today. Starburst already provides the foundation for a growing number of organizations’ analytics, and increasingly, we’re also extending that foundation to include AI. Because both analytics and AI involve data architecture, it’s a natural extension. AI needs many of the same things as analytics, and this need is not only growing, it’s exploding. 

Let’s look at a few examples to see how our customers are using Starburst to power their AI architecture. 

1) Going uses Starburst to build the foundation of their AI architecture

Background: Going provides real-time airfare intelligence for predictive pricing and personalized recommendations. To do this, they analyze rapidly changing airfare prices around the globe. 

Challenges: Going needed something interesting–an analytics foundation that doubled as an AI foundation. Their existing infrastructure lacked scalability, making it difficult to support AI-driven price forecasting, deal scoring, and personalized travel recommendations.

Solution: Because Starburst provides a foundation for both analytics and AI, Going was able to pursue this dual-use case. Specifically, they use us to ingest data into an Icehouse architecture using Apache Iceberg. This prepares the groundwork for future AI expansion. 

This use case shows an important point about Starburst. Organizations that might have data analytics needs today may very well have AI needs tomorrow. By creating a robust foundation for your data architecture, Starburst serves both use cases. It provides room to grow–and that includes room to grow your AI in a way that works for you and alongside your data analytics. 

Read the Going Case Study to learn more. 

2) Asurion uses Starburst to make AI work for their business

Situation: Asurion is a global technology solutions company. They help their customers protect their devices with tech support, insurance, and device protection plans.

Challenges: Asurion’s team struggled with insufficient data quality. At its peak this led to identifying over 80 data quality incidents in just 90 days. 

Solution: Asurion uses Starburst, to apply advanced ML algorithms to detect poor data quality. This has reduced data quality incidents by over 50%. Check out this video to learn more about Asurion’s solution. 

And they didn’t stop there. As they continue to innovate and push further into AI uses, Asurion is also using Starburst to build a scalable AI-powered analytics platform by enabling seamless access to enterprise data. This approach eliminates data silos, accelerating AI-driven insights, and empowering teams to access data using natural language (NLP) queries—without requiring SQL expertise. 

Hear more from the team implementing this project on our upcoming webinar on March 26th, 2025. 

3) Vectra uses Starburst to achieve AI data governance 

Situation: Vectra AI  is a cybersecurity company. It uses AI for hybrid attack detection, investigation, and response solutions.

Challenges: Vectra’s previous data architecture limited its ability to grow, particularly with regard to historical data access and data governance. In essence, their data foundation was not broad enough or agile enough to capture their need for data velocity. This resulted in query failures, increasing costs, and a growing technical debt. 

Solution: Vectra uses Starburst Galaxy to power the foundation of its AI data application. In doing so, they are able to leverage strong data governance across multiple data sources. Because Starburst is designed to secure your data at any scale, Vectra embeds Starburst to serve an security platform application requiring data from multiple sources using near real-time ingestion. This allows its AI models to detect and investigate threats faster—without the need for costly data movement.

Check out this case study discussing Vectra’s AI solution. 

 

Starburst fuels your AI journey

Starburst’s superpower is its ability to build the foundation of your data architecture wherever it lives–either in the cloud, on-premises or in a hybrid environment–and to do it in a way that works for you and your business. This is already upending the analytics industry, but with the AI era now here, it’s set to push into overdrive. 

Today, we sit at the pivotal intersection between data as a raw resource and data as a valuable asset. Every organization is facing this challenge, looking to turn their data into insights, and we are ready to meet that challenge. 

Starburst: The foundation of your AI architecture

Starburst is the foundation upon which your analytics and AI house is built. We unify all of your data and make it all accessible to a single query or a single AI model. We foster collaboration and unity, so your entire business can move in the same direction, and we secure that data foundation with industry-leading data governance that ensures that you meet the compliance and regulatory requirements of the AI age. 

As the world embraces AI in new and exciting ways, we are ready to help organizations meet that moment and derive more value from their data than ever before. This is why I started Starburst, and it’s why we will continue to lead as the value of data continues to evolve.