What does it take to achieve real AI change in an enterprise environment? It’s a question on every organization’s mind as we enter the next phase of the AI revolution. We know that AI can and will change entire industries. We see it each day, but we also know that it can go much further, and we anticipate that change and seek ways to channel and guide it in certain specific ways.
It’s also an area where success has specific meanings. We know success, intuitively, we see it. Whether we consider the meaningful integration of AI into enterprise production use cases or the deployment of agentic intelligence, these initiatives work when they understand the betterment of the enterprises in which they operate.
AI works when it provides change within a business context
Change, then–change in and of itself–is not the problem. This is not a story of change for change sake, or AI for AI sake, but rather, change within a business context, and AI within a business context.
This is the AI leap, and understanding how to make it happen, in the right way, for our businesses is now essential for enterprise success.
To repeat the question, what is required to achieve AI change in enterprise environments in ways that work within a specific business context? In other words, how do we approach the AI leap with context in mind?
Why enterprise AI doesn’t happen automatically
The first step in navigating the AI leap is understanding where it can go and what it takes to get there. The real question is not where we’re going, but how to get there, and how to engineer that change.
And that’s an important word, engineered. The truth is that the AI leap won’t just happen. The AI leap needs to be engineered. It needs to be purposefully, intentionally constructed.
Perhaps because AI has rapidly removed so many barriers, casting open the floodgates of potential in ways that seem, at times, limitless, it can be easy to forget that AI still has dependencies.
It still needs things to be successful, accurate, trusted, and impactful. In a word, it needs to be engineered. That work of engineering the AI leap is only just getting started, but it represents the true bottleneck in AI success.
Let’s take a look at two examples.
There’s an invisible engineering AI bottleneck impacting the AI leap
The first is from venture capital firm a16z.
Recently, they looked at where AI is having the greatest success, and there is absolutely no question that the shift to AI is real. If you look at a16z’s recent analysis of where enterprise AI revenue is concentrated, billions of dollars are flowing directly into applications supporting coding, legal, support, and search activities.
What’s missing from this picture?
It’s an impressive picture, but there’s something even more impressive that’s missing. If you pause and look closely at that picture, you notice what is glaringly absent. The infrastructure underneath is not mentioned. It’s assumed, but it’s not explicitly discussed.
In this absence, the data, the context, and the governance are being left behind. Enterprises are aggressively buying and building frontend AI tools while the data foundation beneath them remains completely broken.
This absence is important and points to one of the true bottlenecks in the AI story that is only now being revealed. It points to the need for the AI leap to be engineered.
NVIDIA is pivoting to deal with the need to engineer the AI leap
This isn’t just our perspective. It’s a shift underway in other areas of the AI industry as well.
NVIDIA, the world’s largest technology company, recently addressed the issue of engineering the AI leap directly. They argue that you cannot build a scalable AI strategy on top of a fragmented data estate.
In fact, Jensen Huang speaks to this exact point in a recent conversation with Michael Dell of Dell Technologies. He explains how structured data is now its own AI bottleneck, and that bottleneck controls access to the context layer needed to make AI accurate, governed, and valuable.
In other words, structured data is needed to engineer the AI leap. Starburst is mentioned by name because it solves this problem. It engineers the AI leap.
The ground truth of AI
It’s a powerful idea, but it’s also one that’s been building for some time. A few months ago, at NVIDIA GTC, the idea was shared in an incredible slide discussing the importance of structured data as the ground truth of AI. In other words, the importance of contextual data and business context to engineering the AI leap.
There’s something else important in the photo too.
If you look closely, right there in the center of a massive, 120 billion dollar structured data ecosystem, Starburst is highlighted in purple. We are incredibly proud of our NVIDIA partnership, especially since Starburst on Vera delivers 2.7x faster query performance, though we will discuss the technical details later.
Engineering the AI leap is the key to enterprise AI success
But the real key to this slide is that NVIDIA sees the exact same bottleneck that we see. They see it because they are steeped in the needs of AI for enterprise environments, and are aware of how important context is in enterprise AI success.
Simply having the data isn’t the same as having the meaning behind it. Structured data is the operational ground truth of your business, but most enterprises are completely failing to unlock it.
They see the need to engineer the AI leap.
Why the data foundation is the true bottleneck of AI
Given this reality, it’s important to think of your AI initiative in the right way. Most importantly, it’s not really your AI model that’s failing. It’s your data foundation. Alongside the context layer and semantic layer, the data foundation is the true ground truth of AI, exactly as NVIDIA says.
When an enterprise attempts to run modern AI workloads on fragmented backend systems, the structural cracks appear instantly.
Every day, we watch organizations pour millions into sophisticated large language models and autonomous agents, only to find them stalled, hallucinating, or entirely disconnected from the business reality.
Time and again, the pattern reemerges.
The model itself is rarely the weak link. It is simply waiting on data it cannot properly access or understand.
The pain of the AI leap when engineered badly
The traditional, centralized approach to a massive enterprise data estate is getting exponentially expensive in an AI-focused world.
The story has other dimensions too.
Recent findings from Mavvrik highlight a stark financial reality. An overwhelming 84 percent of companies report that skyrocketing AI costs are actively eroding their gross margins. Even worse, 85 percent of organizations are completely missing their initial AI cost forecasts. It is costing a fortune, and most companies cannot accurately predict the operational damage. This absolute lack of predictability is forcing a massive shift, with 67 percent of enterprises now actively planning to repatriate their AI workloads back on-premises just to regain control over their bottom line.
This is what comes of engineering the AI leap using the wrong tools.
Caught between volatility and gravity
It’s an ongoing trend. This operational pain happens because enterprises are caught between two opposing forces pulling in entirely different directions.
At the top, you have extreme AI and application volatility. The models, parameters, and application tools shift constantly. What your team picks today will not be what they rely on next year.
At the bottom, you have enterprise data gravity. Your data is heavy, fragmented, highly regulated, and immensely expensive to move. Trying to force this heavy, unyielding data to migrate at the frantic speed of volatile AI models creates immense executive risk, soaring overhead, and technical instability.
The tension between the two can pull in opposing directions. Managing it, again, requires stable foundations.
A stable data foundation is key for engineering the enterprise AI leap
All of this means that enterprise AI needs a stabilizer to ensure that the data foundation is sound.
What does this mean in practice?
Well, it means that before anything else can happen, you need to think about data and the context around it. Everything that comes afterward relies on that foundation. This architectural stability must happen before anything else, because every single downstream query, real-time inference, and agentic workflow relies entirely on that foundation.
The goal of most enterprise AI, achieving enterprise intelligence, relies on this foundation. In many ways, you can see this as the objective of engineering the AI leap.
What does it take to achieve an enterprise intelligence platform?
All of this begs the question, what is needed for enterprise intelligence? The same two answers circulate:
- Data
- Context
Once you accept that you need a way to absorb application volatility above while respecting data gravity below, the path forward becomes clear. You cannot rely on the old playbook of forcing massive, multi-year migration projects.
How Starburst built an enterprise intelligence platform
At Starburst, we approached this problem head-on.
We designed our enterprise intelligence platform architecture to deliver complete AI readiness without forcing you to rebuild your entire data stack from scratch. It sits cleanly as a resilient layer between your data sources and the tools your teams use to access them.
In effect, it allows you to query your data in place, meaning you move only what you choose. This stabilization is delivered through four distinct architectural layers.
Analytics engine
This high-performance core is built on Trino to provide universal, federated data access. It allows you to query directly across your distributed data estate, reducing the need to move or duplicate your assets.
Ultimately, it creates the scalable analytics workflows that form the true foundation for AI.
Enterprise context layer
This layer is critical to making AI a real business solution rather than just an isolated experiment. It applies common definitions, metadata, and policies across all your domains.
This process turns raw data into a shared ground truth that both humans and AI can actually understand.
Agentic control plane
This is the governance boundary where you coordinate agents, tools, and complex AI workflows. It allows you to enforce strict policy and governance for AI-driven actions.
By setting these guardrails, you maintain total flexibility without risking corporate reality.
Agentic interface
This layer is built around our core belief in model optionality. While it provides the flexibility to use your own custom models and third-party agents, it also serves as the gateway to our native capabilities.
It ensures you never lock your business into a single vendor or a single bet.
How do you make it happen? That’s where Starburst’s AIDA comes in.
Starburst’s AIDA
AIDA is the ultimate expression of enterprise intelligence in action. It’s what happens when you have a data foundation with access to all of the data and context needed to make decisions with confidence and accuracy. It’s what becomes possible when you engineer the AI leap with data and context in mind.
In many ways, this is also the opposite of what many tools offer. Look at the market right now.
Every AI tool out there positions itself purely as a passive advisor. You ask a question, you get an answer, and then a human still has to manually go and do all the heavy lifting to execute.
We believe the next logical step is AI that does not just tell you what is happening, but actually helps carry out what should happen next. This has to be done in a governed, auditable, and trusted environment.
That is the critical shift we are driving, moving from analytics that merely inform to analytics that act.
Grounded reasoning over guessing
Most generic text-to-SQL tools or standard chatbots struggle in a real business environment because they are just guessing about user intent. They look at a database blindly without any appreciation for how the business actually runs. This worsens the problem of AI hallucination.
AIDA is built on a completely different premise. Its reasoning is fundamentally grounded in your governed data products and established business context. AIDA does not just run queries against a raw database. It inherently understands your data assets, respects your data policies, and explicitly explains its logical path to deliver trusted answers in seconds.
In other words, it’s engineered for the AI leap.
Unified governance at the engine level
Governance is another aspect of this that can’t be ignored. AI needs far more than basic data connectivity. It requires an active Agentic Control Plane.
AIDA operates entirely within this layer to coordinate workflows, enforce your existing data policies, and set strict guardrails for AI-driven actions. Because security, access control, and lineage are embedded directly into the data foundation rather than bolted on afterward, you can safely expose your distributed data estate to AI.
You get full compliance while maintaining total operational flexibility.
Eradicating the business intelligence backlog
This approach is revolutionary.
I wrote before how traditional BI is failing the modern enterprise because the operating model is slow, static, and old-fashioned. Think about the reality inside most organizations today. Teams routinely spend over four months building a single dashboard. Then, 51 percent of users say they cannot even interact with that data meaningfully, and 72 percent bypass the dashboards entirely by exporting raw files to Excel. You wait a third of a year for a report, only to end up with unmanaged spreadsheets and answers nobody trusts.
Dashboards are not dead, but they are absolutely being demoted from the default interface for enterprise intelligence. The world is shifting rapidly toward conversational, contextual, and actionable intelligence.
AIDA shatters this bottleneck by providing a natural language interface to explore trusted data without SQL, dashboards, or waiting in an IT ticket queue.
Engineering the AI leap is the key to enterprise AI success
All of this points to a need to approach the AI leap, especially in an enterprise environment, as an engineering problem that needs an engineering solution. The ingredients of this engineering challenge include data and context, but it is an engineering problem at its core. Understanding this is the next
The speed of AI change demands engineering the AI leap at speed
These shifts are intensified by the accelerated pace of AI itself. The operational window for enterprise AI requires immediate, decisive action. We are operating in a market where strategy windows open and close in fractions of a second, meaning true operational excellence is defined by speed and clarity.
To win, your organization must be able to instantly ask questions across your entire data footprint, understand the exact business context behind those numbers, and execute decisions immediately before that competitive window closes forever. Leaders simply cannot afford to sit on the sidelines waiting for multi-year transformation programs or massive cloud migrations to finish.
Your data and your context are the ground truth of your enterprise
In short, you cannot wait for perfect data. You need your data accessible, contextual, and AI-ready immediately. The shift required is simple. Stop waiting to fix the backend and start activating what you already have.
The winners of this race are not rebuilding. They are intentionally engineering the AI leap.
By combining a high-performance federated analytics engine with an enterprise context layer and AIDA, the right data foundation turns your AI strategy into immediate business execution.
Change has always been a part of enterprise environments. The real question is, what is required of this change at this moment?
