Data, AI & Analytics • 5 hours ago • Shruti Das
For much of the generative AI boom, enterprise strategy has revolved around a relatively straightforward question: Which AI model should we use?
Companies compared large language models. They experimented with copilots. They tested different providers and debated whether proprietary or open models would be better suited to their needs. That question is still important. But it is becoming less central.
As enterprises move beyond experimentation, a more fundamental issue is emerging: Where does the data live, and how should AI interact with it?
This shift was reflected in a series of announcements from Broadcom at VMware Explore on August 31, including its VMware Private AI Cloud initiative. The company explicitly framed the approach around bringing models to enterprise data rather than moving enterprise data to the models. The announcement is significant beyond the technology itself because it reflects a broader change in enterprise AI architecture.
The next phase of AI may be less about moving information into a model and more about building AI systems that can operate securely around the data enterprises already have.
The model is only one part of the equation
The early generative AI conversation naturally focused on models. That made sense. The sudden emergence of powerful foundation models created a visible technological breakthrough, and organizations rushed to understand what these systems could do.
But enterprises rarely operate on a single clean dataset. Their information is distributed across data warehouses, operational databases, SaaS platforms, cloud environments, internal applications and decades of legacy systems. Some data is highly sensitive. Some are regulated. Some are poorly structured. Much of it lacks the context required to make it useful to an AI system. That means the challenge of deploying AI in a real enterprise is rarely as simple as connecting a model to a database.
The more difficult questions are often:
- Which data should the AI be allowed to access?
- Where should that data remain?
- How can access be governed?
- How can the AI understand business context?
- How can organizations prevent sensitive information from being unnecessarily copied across environments?
These are fundamentally data architecture questions. And they are becoming increasingly important as AI moves closer to production systems.
Data gravity is becoming an AI strategy issue
For years, the concept of data gravity has influenced enterprise technology decisions. Large volumes of data are difficult and expensive to move. Applications and services often end up being built closer to where that data already exists. AI is beginning to reinforce the same principle.
Moving enterprise data repeatedly between systems can introduce security, compliance, latency and cost challenges. For highly regulated industries, the question can be even more complicated because data may be subject to residency, sovereignty or industry-specific requirements. That creates a growing incentive to bring AI capabilities closer to established enterprise data environments.
Broadcom’s recent announcements illustrate this direction, with the company positioning its private AI infrastructure around running and governing AI workloads where enterprise data already resides. Its related AI-ready data foundation announcement also focused on governed agents and governed data within private cloud environments.
The larger lesson is that AI architecture is increasingly becoming inseparable from data architecture. An organization can select an excellent model and still struggle to create useful enterprise AI if its data remains fragmented, inaccessible or poorly governed.
The real bottleneck is often the data foundation
This is becoming one of the most important realities of enterprise AI. Many organizations have successfully built demonstrations. A team can take a carefully selected dataset, connect it to an AI model and produce impressive results in a relatively short period. Scaling that experiment across the enterprise is much harder. The AI system eventually encounters the realities of corporate data: duplicate records, disconnected systems, inconsistent definitions, outdated information and unclear ownership. The problem is not always that the AI model lacks intelligence. The problem may be that the organization has not created a reliable environment in which that intelligence can operate.
IBM has made a similar argument around the changing requirements for AI-ready data, emphasizing that enterprises moving beyond experimentation increasingly need real-time, contextualized and trusted data rather than simply broad access to information.
That distinction is important.
Available data is not necessarily AI-ready data. For an AI system to produce useful results, it needs more than access. It needs context. It needs to understand what a customer identifier represents. It needs to distinguish between an approved policy and an outdated document. It needs to understand which data is authoritative and which information should not influence a decision. Without that context, connecting more data to AI can simply create a more sophisticated way of processing confusion.
Private AI is about more than privacy
The growing interest in private AI environments can therefore be misunderstood if it is viewed only as a security trend. Security is clearly important. Enterprises do not want sensitive information flowing into uncontrolled environments. But the strategic value of running AI closer to enterprise data also involves control, cost and operational simplicity.
An organization that moves large volumes of information between multiple platforms may face growing infrastructure costs and complexity. As AI usage expands, those costs can become increasingly significant. There is also the question of governance. When data, models and AI applications are distributed across disconnected environments, it becomes harder to understand how information is being used and where decisions are being made. Keeping more of the AI stack connected to established enterprise infrastructure does not automatically solve those problems. But it can make them easier to manage.
The emerging enterprise architecture is therefore likely to be more hybrid than the early AI conversation suggested. Some workloads will run in public clouds. Others will use private infrastructure. Organizations will continue using external foundation models while also deploying models and AI services closer to sensitive data. The future may not belong to a single AI environment. It may belong to organizations capable of managing where different AI workloads should run and why.
AI is making data architecture strategic again
For years, data architecture has sometimes been treated as a technical concern sitting behind more visible business initiatives. AI is changing that. The quality of an organization’s data environment increasingly determines the quality of its AI ambitions. A fragmented data estate can slow deployment. Poor governance can create security risks. Missing context can reduce accuracy. Excessive data movement can increase costs. These are no longer isolated data management problems. They are AI problems. That could significantly change the role of data leaders within the enterprise.
Chief data officers and data teams are increasingly being asked not simply to manage information but to create the conditions in which AI systems can operate reliably. That includes building:
- Trusted data foundations
- Clear semantic definitions
- Real-time access where required
- Strong governance controls
- Reliable identity and permission models
- Clear ownership of critical data
In other words, the infrastructure behind AI may become just as strategically important as the model itself.
The next AI race may be about architecture
The technology industry has spent much of the past several years competing on model capabilities. That competition will continue. Models will become faster, more capable and increasingly specialized. But enterprises may find that access to the best model is not their biggest competitive advantage. If multiple organizations can access similar AI capabilities, differentiation will increasingly come from what those models can do with proprietary enterprise information. That depends on data.
A company with fragmented systems and poorly governed information may struggle to operationalize even the most capable model. A company with trusted, well-organized and context-rich data may be able to create significantly more useful AI applications with the same underlying technology. This is where the enterprise AI conversation begins to shift. The question is moving from: Which model is smartest? To: Which organization has built the best environment for AI to work?
That is a much broader challenge. It involves infrastructure, governance, data engineering, security and business architecture.
Bringing AI to the data changes the operating model
The idea of bringing AI closer to enterprise data also changes how organizations think about deployment. Instead of treating AI as a separate destination where information is sent for processing, enterprises can increasingly think of AI as a capability embedded within their existing technology environment. That could mean AI working alongside databases, analytics platforms, applications and operational systems. The advantage is not simply proximity. It is integration.
Enterprise AI becomes more valuable when it can work with current information, understand organizational context and participate in existing workflows. This is particularly important as AI evolves from answering questions to performing more complex tasks. An AI system that summarizes historical information is useful.
An AI system that can analyze current information, identify a problem and support a business workflow is more significant. That progression requires AI to move closer to the systems where data and decisions already exist.
The model matters. The data environment matters more.
None of this means enterprises should stop thinking about models. Model selection still matters. Cost, performance, reliability and security will remain important considerations. But the industry may be approaching a point where model capability becomes easier to access than high-quality enterprise data. That could make the data environment a more durable competitive advantage.
Organizations cannot simply purchase years of accumulated business context. They cannot instantly recreate trusted data relationships, operational knowledge or well-established governance frameworks. Those capabilities have to be built. The next phase of enterprise AI will therefore be shaped by an increasingly simple principle: AI is only as useful as the environment in which it can safely access and understand enterprise data.
The companies that succeed with AI may not be the ones that chase every new model release. They may be the ones that recognize that the foundation of enterprise AI was never just the model. It was the data all along.
Key Takeaways
- Enterprise AI strategy is shifting from model selection toward data architecture. Where data lives and how AI interacts with it are becoming critical decisions.
- Data gravity is influencing AI deployment. Security, compliance, latency and cost are creating incentives to bring AI closer to enterprise data.
- Available data is not automatically AI-ready data. AI systems need trusted, contextualized and well-governed information.
- Private AI is about more than privacy. It also reflects growing concerns around operational control, cost and governance.
- Data architecture is becoming a competitive AI advantage. Organizations with strong data foundations may be able to generate more value from similar AI models.
- The next enterprise AI question may be less about which model to use and more about building the right environment for AI to operate.