In search of AI value: The first rule of tokenomics is ‘Don’t count tokens’

Reading Time: 7 minutes

If AI were a TV series, it has so far aired for three seasons. In season one—the Adoption Era—the storyline was, “How does our organization get people to use AI?”

Organizations encouraged tokenmaxxing (an approach that persuades workers to use their maximum token allocation) with scant regard for the associated costs. When the bill finally arrived in this first AI season, the Adoption Era came to an abrupt, cliff-hanger ending—with lots of spending but few beneficial outcomes.

Then, in season two—Consumption Shock—the reality shifted from the subscription-based, all-you-can-eat token buffet to consumption-based bills. The question became “What does AI use cost our organization?” And of course, the models, and the methods to get the most value from those models, were—and are—evolving.

FinOps teams (cross-functional groups that optimize cloud financial spending and bridge the gap between engineering, finance, and business departments) tried to determine the return on investment of organizational AI spending. FinOps focused on token spending: prompts, outputs, model calls, retrieval steps, agent actions, and cost per token.

But as the plot thickened in season two, it became clear that tokens were only one character in a much bigger cast. Compute, storage, memory, data, model training, infrastructure, governance, and people all drive the total cost of AI.

Fast-forward to season three: The Value Era.

Today, the question is no longer “How much does AI cost?” but rather, “What value does our organization get from the full AI stack?” Instead of waiting until the cloud bill arrives, organizations need to move cost and value decisions earlier in the AI lifecycle: into design, architecture, infrastructure planning, governance, and business measurement. And these teams must look beyond costs to how AI affects workflows, risk, mission impact, compliance, user productivity, operational readiness, governance, adoption, strategic alignment, and business outcomes.

The AI lens has expanded from ROI alone to a broader view of total cost of ownership and cost-benefit analysis.

“The first rule in tokenomics is ‘Don’t count tokens,’” noted J.R. Storment, executive director at the Tokenomics Foundation, a Linux Foundation project, during the session “Dynamic ROI, FinOps, and tokenomics in the AI era.”

The session was one of many informative presentations on AI value and costs at VMware Explore, the virtualization conference taking place Aug. 31 to Sept. 3, 2026, at the Venetian Convention and Expo Center in Las Vegas.

The first rule in tokenomics is “Don’t count tokens.”

From token spend to total AI economics

Understanding how to calculate AI value took center stage at the VMware Explore conference.

In the tokenomics session, for example, speakers addressed the AI value lifecycle and how to understand where the bulk of costs rack up.

Indeed, token costs represent “really just the tip of the iceberg,” said Shreenivasan “Raj” Rajagopal, head of ValueOps, AI, and Infrastructure at Broadcom, during the session.

“Only about 25 percent of the cost [comes from] the models at the token layer,” Rajagopal said.

Finance professionals, IT infrastructure leaders, and C-suite executives should take note if they care about quantifying AI value. The data center infrastructure layer generates the bulk of the costs Rajagopal continued.

“Seventy-five percent sits lower—in all the other stuff. Yes, the people, the compute, the storage, the memory, the [model] training, and the attached models, revenues, the inputs, the outputs are all part of the bigger AI economics picture. This iceberg of costs represents a new way of thinking about the infrastructure layer.

C-suite executives should take note. The data center infrastructure layer generates the bulk of the AI costs.

Tokenomics ‘shifts left’

Accordingly, the tokenomics conversation is “shifting left” and moving AI cost management and cost-benefit analysis earlier in the AI lifecycle. Instead of assessing value only during the cloud billing phase (which is effectively the end of the AI lifecycle), experts urge operators to factor in AI costs per outcome which rack up during the design, IT architecture, and coding phases. As tokenomics matures, making cost-efficient decisions at these points in the cycle becomes increasingly important.

How the IT infrastructure layer enables and constrains AI was a key theme at the VMware Explore conference. The newly released VMware Private AI Cloud strives to give organizations a production-ready path to securely build, run, and govern inference workloads, agentic applications, and traditional enterprise workloads together on a single private cloud platform. One of the key goals of this unified platform approach is to scale AI cost-effectively.

The tokenomics conversation is ‘shifting left’ and moving AI cost-benefit analysis earlier in the AI lifecycle.

Creating a map of tokenomics

Rajagopal urged more organizations to think about AI value in the context of three buckets in this AI lifecycle and seek to answer this question: “What value are we getting from a set of inputs?”

AI leaders can approach tokenomics through three interconnected domains: production, consumption, and value.

As AI adoption, and costs accelerate, organizations should as, “What value are we getting from a set of inputs?”

But organizations that focus only on those three areas often discover a gap. They can measure infrastructure costs and token usage but still struggle to determine who owns those costs, whether spending aligns with business priorities, and how AI investments translate into outcomes. This is where governance becomes critical. Governance serves as the connective layer across all three domains, establishing accountability, measurement standards, and decision-making processes that help organizations link AI spending to business value.

1.Production. This bucket involves the key inputs that drive AI costs. It includes the cost generated by data centers for the inputs to generate AI. Electricity, computer chips, and data are key inputs that go into servers, while finished tokens emerge as text, code, predictions, or decisions. Measuring inputs and their associated costs is a foundational step in understanding AI value. From a governance perspective, organizations must determine who owns infrastructure investments, how costs are allocated among business units, and which efficiency metrics define success.

2. Consumption. Most enterprise users occupy this domain by consuming AI models. There are several dimensions to the consumption bucket.

• Inputs and outputs. Users consume tokens when they submit prompts (inputs) and when AI systems generate responses (outputs). This has traditionally been the primary method of AI cost measurement.
• Agent workflows. Agentic applications execute additional behind-the-scenes actions, including model calls, tool invocations, retrieval operations, and retries that can rapidly multiply token consumption. Without visibility into these activities, organizations may see costs rise without understanding why. Governance frameworks provide the transparency to monitor usage patterns, establish guardrails, assign ownership, and ensure that consumption aligns with business objectives rather than simply tracking token volumes.

3. Value. In the final bucket, organizations evaluate the total cost of ownership for AI and determine whether AI spending is producing meaningful business outcomes. In this context, “good” tokenomics is not about minimizing token use. It is about ensuring that every token contributes to a valuable result.

Rajagopal emphasized that AI creates a unique dynamic. While AI generates new costs, it also enables agentic workers capable of generating significant business value. The challenge for organizations is determining whether AI is improving productivity, accelerating decision making, enhancing customer experiences, or driving revenue growth. Governance closes this loop by connecting spending and usage metrics to business outcomes, enabling leaders to measure cost per outcome, assign accountability, and make informed investment decisions.

For SHI customers, this is where the conversation extends beyond infrastructure optimization. Organizations need governance frameworks, operating models, and measurement strategies that connect production costs and consumption patterns to business results. The goal is not simply to spend fewer tokens. It is to achieve better outcomes for every dollar invested in AI.

How SHI can drive better AI value

SHI helps organizations take a holistic approach to AI value realization by evaluating the full AI stack— from infrastructure, data, and governance to models, applications, and business outcomes. By exploring how AI affects the entire stack within a data center, IT leaders can better identify true costs but also new value generation (such as agentic workers). Rather than focusing solely on cost per token, SHI helps customers measure cost per outcome, identifying where AI investments create productivity gains, accelerate decision-making, reduce operational costs, or drive new business value.

SHI’s expertise spans AI infrastructure, FinOps, IT asset management, cloud economics, and advanced AI strategy. As a founding member of the Tokenomics Foundation, SHI launched and is an active participant in shaping emerging standards for AI value measurement. It helps organizations establish governance models, visibility frameworks, and optimization strategies that connect AI spending to business results.

Rather than focusing solely on cost per token, SHI helps customers measure cost per outcome.

This includes helping organizations with the following:

• Gain visibility into AI costs across infrastructure, models, agents, applications, and cloud services.
• Optimize workload placement, model selection, and model-routing strategies to balance cost, performance, governance, and security requirements.
• Measure AI value through meaningful business metrics rather than token consumption alone.
• Align IT, finance, and business stakeholders on a shared framework for AI investment decisions.
• Develop long-term AI operating models that support governance, accountability, and continuous optimization as AI usage scales.

As Rajagopal indicates, consistent monitoring of inputs and outputs starts to show real benefits. “When you do this dynamically, consistently—daily, weekly, monthly—you realize ROI dynamically.” According to some studies, Rajagopal said, dynamic and consistent measurement could yield 5,00% ROI.

The organizations that succeed in the next season of AI will develop frameworks to optimize the relationship between AI costs and business outcomes. SHI helps customers build that capability, turning AI economics into a strategic advantage and ensuring that every AI investment delivers measurable AI value.

NEXT STEPS:
Why AI tokenomics needs more than cost per token

How model routing improves AI economics

Stop chasing tokens; start measuring outcomes

Speak to an SHI expert about AI spending