Share with your CDO
The enterprise AI investment thesis has a structural blind spot, and Vinod Bijlani of HPE Services names it directly in this datanomics framework piece. Organizations have spent two years obsessing over inference efficiency (tokens per second, cost per million tokens) while largely ignoring the preparatory economics that determine whether those tokens mean anything. MIT research pegs 95% of enterprise generative AI pilots at zero measurable return. Gartner predicts 60% of AI projects unsupported by AI-ready data will be abandoned through 2026. The argument is that measuring AI performance without measuring data readiness is like grading a restaurant on plating speed while ignoring whether the ingredients are fresh.
What this means for your business
Where you sit on this depends less on your AI budget and more on your data estate’s current condition. Organizations that built clean, governed data platforms before the generative AI wave arrived are now watching their inference investments compound. Organizations that rushed to model deployment while deferring data quality work are accumulating a kind of hidden AI debt, where each new use case requires a fresh round of expensive remediation before it can produce anything trustworthy. Gartner’s figure that 63% of organizations lack adequate data management practices for AI isn’t a prediction; it’s a current diagnostic.
The agentic AI dimension sharpens this considerably. Copilots and assistants operate within relatively constrained boundaries, so a data quality gap produces a bad answer. Agents, which are designed to reason across workflows and take actions autonomously, amplify bad data into bad decisions executed at speed. McKinsey’s finding that fewer than 10% of enterprises have scaled agents to tangible value, with 80% citing data limitations as the blocker, suggests the bottleneck isn’t model capability. Bijlani’s proposed metrics (cost to AI-ready data per terabyte, time to AI-ready data, trusted data coverage as a percentage) are a more honest scorecard than anything currently dominating infrastructure procurement conversations.
The sovereign AI framing in this piece is where the argument gets genuinely sharper than most datanomics writing. Sovereignty discussions habitually collapse into compute location and model provenance, but real control requires auditable governance over where data resides, who accesses it, and how AI decisions get traced back to the data that produced them. Accenture’s finding that only 15% of organizations have made AI sovereignty a CEO or board-level priority means most governance conversations are still happening at a compliance tier that can’t actually enforce those controls. The CDO who has sovereignty defined only at the infrastructure layer is one regulatory audit away from discovering the gap.
Bijlani writes from inside HPE Services, which has obvious commercial interest in expanding the AI cost conversation beyond GPU procurement and toward managed data infrastructure, so the datanomics framing conveniently expands the addressable problem to HPE’s portfolio. That incentive doesn’t make the argument wrong, but it does mean the proposed metrics (cost per TB to AI-ready, trusted data coverage) should be stress-tested against your own environment before they become the basis for a vendor conversation. The falsification condition here is straightforward: if your organization can point to scaled agentic deployments running on data you haven’t formally governed, the datanomics premise needs revision. Most organizations won’t be able to make that case.
Concept deep-dive: Datanomics
Datanomics, as Bijlani defines it, is the economic discipline of measuring what it costs to make enterprise data usable by AI systems, covering discovery, quality assessment, deduplication, access governance, and pipeline reliability. Tokenomics measures the cost of generating AI output. Datanomics measures the cost of making that output trustworthy. The analogy is manufacturing quality control: you can optimize a production line for throughput, but if the raw materials are contaminated, speed makes the waste problem worse, not better.
Based on reporting from Beyond Tokenomics: Why Datanomics Is the Missing Half of the AI Economics Equation, originally published 2026-07-27 03:06:00.
