AI’s real bottleneck isn’t compute — it’s context

By now, it’s old news that AI is everywhere in the enterprise. Copilots, agents, predictive systems — they’re running finance, supply chains, HR, customer ops. A recent survey found that by the end of 2025, half of companies had AI in at least three business functions. That’s fast adoption, but it’s also exposing a hard truth that a lot of leaders didn’t see coming.

The problem isn’t model performance. It isn’t compute power either. It’s the quality and context of the data feeding those systems. AI introduces a new requirement: it doesn’t just need access to data — it needs to understand the business context behind that data. Without that, you get answers that are technically correct but operationally useless. Or worse, actively harmful.

Irfan Khan, president and chief product officer of SAP Data & Analytics, puts it bluntly: “AI is incredibly good at producing results. It moves fast, but without context it can’t exercise good judgment, and good judgment is what creates a return on investment for the business. Speed without judgment doesn’t help. It can actually hurt us.”

I’ve seen this play out in real deployments. A company rolls out an AI for supply chain optimization. It crunches inventory levels, lead times, supplier scores. It spits out recommendations in seconds. But it doesn’t know which customers are strategic accounts, what tradeoffs are acceptable during shortages, or which contractual obligations are in play. So it optimizes for the wrong thing. Both speed and accuracy, but no judgment. That’s not a win.

This is where data fabric comes in. Not the old-school approach of just dumping everything into a data lake and hoping for the best. A proper data fabric is an abstraction layer that spans across infrastructure, applications, clouds, and operational systems. It preserves the semantics — the meaning — of how data relates to business processes, policies, and real-world decisions. For agentic AI, that fabric becomes the primary interface. Agents query business knowledge, not raw storage.

Khan breaks it down into three components: intelligent compute for speed, a knowledge pool for business understanding and context, and agents that act autonomously but are grounded in that understanding. The magic is in how these three work together. Speed alone is dangerous. Context without action is useless. Action without grounding is chaos.

Here’s what I find interesting: most companies know they’re not ready. Only one in five consider their data approach highly mature. Only 9% feel fully prepared to integrate and interoperate their data systems. That’s a huge gap. And it’s not a technology gap — it’s an architecture gap. Companies have spent two decades building data warehouses and lakes that aggregate data but strip away context. Now they need to rebuild with context as a first-class citizen.

Knowledge graphs are central to this. They let agents query enterprise data using natural language and business logic, not SQL or API calls. That’s a big shift. Instead of asking “what’s the inventory level for SKU 123?” you ask “which strategic customers might be affected if we run out of component X?” The system needs to know what “strategic” means, who those customers are, and how component X maps to finished products. That’s context.

The old approach of consolidating everything into one repository is dying. The new approach is integration — connecting data where it lives, with meaning preserved. It’s harder to build, but it’s the only way AI delivers real business value instead of just fast wrong answers.