Vinod Bijlani is an AI practice leader at Hewlett Packard Enterprise.getty​Over the past two years leading enterprise AI programs across Asia Pacific, one pattern has become impossible for me to ignore: The organizations I see spending the most on AI are not the ones getting the most value from it.That's not for lack of investment. Gartner forecasts worldwide AI spending will reach $2.52 trillion this year, a 44% jump from 2025. Almost all of that conversation happens in tokenomics, which includes inference cost, tokens per second and time to first token. But it's the wrong ledger. Stanford's 2026 AI Index found that 88% of organizations now use AI in at least one business function, but fewer than one in 10 have scaled it into production. Ask those same organizations what worries them most, and the answer isn't the model. It's what's underneath it: According to McKinsey, 74% of respondents now cite data inaccuracy as their top AI risk, ahead of cybersecurity and regulatory compliance. Enterprises have been measuring how cheaply and quickly they can generate an answer when the bigger economic question is how much it costs to earn the right to trust one.I call the missing discipline "datanomics." It includes the cost, risk and effort required to turn fragmented enterprise data into something an AI system can safely act on. Datanomics rarely appears as a line item on a budget today. Instead, it reveals itself downstream as stalled pilots, unexpected retraining cycles and the quiet retirement of projects no one wants to justify to the board.The Three Eras Of Enterprise AI EconomicsEvery enterprise I've worked with over the past two years has moved through the same three phases, whether they used these names for them or not:1. Token Economics​Token economics came first, and it's still where most budgets live. This is the era of cost-per-million-tokens, latency benchmarks and model bake-offs. It's a real discipline, and it matters for margin. But it optimizes a stage of the pipeline that comes after the hard part is already solved—or ignored.2. Agent Economics​Agent economics is where the failure mode changes. A chatbot with bad data gives a wrong answer. An agent with bad data takes a wrong action—approves the invoice, updates the record or escalates the case without a human in the loop to catch it. McKinsey found that while nearly two-thirds of enterprises have experimented with agents, fewer than 10% have scaled them to deliver tangible value, with eight in 10 citing data limitations as the roadblock. In the AI factory and private cloud AI deployments I've worked on, this is the exact moment the conversation stops being about model choice and starts being about governance.3. Data Economics​Data economics is where security and sovereignty converge, and it's the era most enterprises haven't admitted they're already in. IBM's "2026 Cost of a Data Breach" report puts the global average breach cost at a record $4.99 million, with AI-driven attacks adding roughly $1 million more per incident. The organizations getting hit are disproportionately the ones that never extended basic access controls to their AI systems and data. At the same time, IDC projects that by 2030, half of all new economic value created by digital businesses in the Asia Pacific region will come from organizations that are investing in and governing their AI capabilities today, not just deploying them. In every sovereign AI conversation I've had with regulated enterprises across the region, the real question was never where the GPUs sit. It's whether the organization can show, on demand, where a piece of training or retrieval data came from and who approved its use.Each era doesn't replace the one before it; it sits on top. By the time an enterprise is running agents under a sovereignty mandate, it's paying the token economics bill, the agent economics bill and the data economics bill simultaneously. Most boards are only tracking the first one.What Datanomics Actually MeasuresDefined holistically, datanomics isn't a single number; it's four parameters, and an enterprise's real AI readiness is only as strong as the weakest one.1. Cost To AI-Ready Data (Per TB): This is the true entry price of AI, and almost no enterprise budgets for it because it's invisible until a project fails. It covers discovery across every system data lives in, deduplication, cleaning, privacy masking and the pipeline engineering required before a model can safely touch the result.2. Time-To-AI-Ready: This is the operational delay spent cleansing, governing and pipelining data before an AI initiative can launch. While teams wait on data readiness, expensive engineering talent and compute resources sit idle, directly inflating project TCO.3. Trusted Data Coverage: This is the share of an enterprise's total data estate that actually meets its own governance, security and sovereignty bar. In most organizations I've worked with, this is a small, well-tended island surrounded by a much larger ocean of ungoverned data. Every AI use case built outside that island isn't running on evidence. It's running on faith.4. Cost To Secure Data (Per TB): This is the operational budget needed to safeguard active AI workloads—a discipline fundamentally different from traditional cybersecurity. Cyber defenses protect network perimeters and databases; AI security must neutralize runtime threats like prompt injection and LLMjacking. Treating AI security like standard IT security leaves the semantic layer wide open to exploit.None of these show up on a typical AI dashboard today. Together, they predict, far more reliably than inference speed, which era of AI economics an organization is actually paying for—and whether it can afford the bill that's coming.The Bottom LineThe first generation of enterprise AI focused on reducing the cost of generating intelligence. The next generation will be defined by optimizing the cost of securing and trusting it. That is the fundamental difference between tokenomics and datanomics. As autonomous agents and private cloud architectures take center stage, the organizations that win won't necessarily be those with the largest LLMs or the highest token throughput. They will be the ones that master the economics of their data foundation.​Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?