Tokenomics: What Enterprises Can Learn from ITAM and FinOps

Every new technology wave introduces new revenue models for vendors, which create new complexities for enterprise end users. At one level of abstraction, tokenomics is emerging as the word of choice for characterizing the new complexities of financial engineering and innovation in new AI opportunities and revenue models. Although many of the specifics are novel and unique to the characteristics of AI, and in particular large language models, the overall patterns are similar to those the industry experienced with novel licensing models for software and hardware and later cloud services. Those waves produced disciplines of their own: IT asset management (ITAM) for software licensing, and financial operations (FinOps) for cloud. Shane Cronin, Head of FinOps & ITAM Services at SHI, has had a front-row seat to previous waves of financial innovations over several decades. Now, as uncertainty emerges around AI tokenomics, he is leading SHI’s engagement with the open source community to sort out best practices for AI financial models and practices as part of the recently-launched Tokenomics Foundation, under the broader umbrella of the Linux Foundation. Is AI tokenomics a genuinely new discipline, or the third act of a story that started with software license audits and ran through cloud egress charges? Cronin says it’s a bit of both: > The underlying problem is the third act of the same story: organizations adopt a new form of technology consumption faster than they build the financial and governance model around it. The same was true for software licensing and cloud, and now we’re seeing it with AI. Where tokenomics differs is the relationship between consumption and value. With a software license, I can identify what I bought. With cloud, I can trace a workload to infrastructure and cost. With AI, a token is only one component of an economic system where model choice, infrastructure, prompts, context, agents, and the quality of the output all affect what I ultimately spent and whether I got anything valuable for it. > > > I believe we shouldn’t keep creating entirely separate organizational disciplines every time the consumption model changes. As a result of ITAM not evolving quickly enough when cloud arrived, FinOps emerged; so now tokens arrive, there’s a new discipline. These capabilities need to converge around technology management rather than being three separate, disconnected silos.” Fascination to discipline The wild success of ChatGPT in 2022 had many enterprise boards and executives wondering how their companies might take advantage of the AI enthusiasm to do something different and gain a competitive advantage. Unfortunately, it’s taken nearly four years to realize that doing something different with AI does not always or necessarily mean more profitable, particularly when disconnected from AI costs. Vendors have taken a few years to figure out how to price the new AI capabilities of foundation models, and more recently agents, often framed by leading AI labs as token consumption. Early enterprise AI experimentation is only now catching up with the new financial models. Cronin explains: > There was a period of time when enterprises were being encouraged to use AI almost indiscriminately, and adoption itself was treated as success. When consumption models changed, usage scaled, and companies started to discover costs they weren’t aware of before or hadn’t anticipated, this push for disciplined use emerged. > > > We’ve seen AI capabilities being enabled without users realizing they were moving token-based consumption to organizations deliberately incentivizing people to consume more tokens. This is where it becomes a CFO conversation. They care whether the investment increased revenue or reduced cost. Once AI spending became material enough that the questions had to be answered, it stopped being just an engineering line item, and that’s when it became a board conversation.” These new financial models are butting up against new bottlenecks in the overall hardware supply chain and rising operational costs, so the landscape is still evolving. For example, Cronin sees many companies telling financial teams to “find $10 million in the next six months” that can be directed toward any number of broader strategic priorities. In the early days of motivating developers to kick the tires of AI innovation, many enterprises encouraged a sort of friendly competition by creating token leaderboards. But as with many new metrics, the enthusiasm quickly unraveled, with sticker shock from consumption that created no lasting value. As Cronin explains: > A token leaderboard tells you who consumed the most, but it doesn’t tell you whether that person produced anything useful. Usage as a point of performance measurement naturally led to tokenmaxxing, but that doesn’t discern if someone is using it to code or to create a cookbook, something with no real value driven by a fundamentally wrong incentive. Thus, the next wave requires smartly combining AI usage with better visibility through metrics and leaderboards that provide showback and attribution, meaning each team sees what it consumed even when nobody bills it for it. This can help to identify which business unit is spending, which use cases are consuming tokens, and the resulting outcomes associated with that spend. However, it also raises a more difficult and nuanced question: how a particular measurement of efficiency translates into overall business performance. Cronin argues that navigating this transition to business performance will require more than just discouraging token waste, because some of it can lead to new opportunities: > The reality is that hard quotas can prevent waste, but they can also prevent experimentation that might create disproportionate value. The goal should be guardrails, not paralysis: make consumption visible, establish ownership, constrain obviously wasteful behavior, and then evaluate whether the outcome justified the cost. One emerging challenge for CFOs is that AI breaks many traditional assumptions used to drive financial forecasting compared to a traditional software contract. Visible token pricing is only part of the economic equation because the supporting infrastructure can change as model behavior evolves. For example, prompt design and data architecture can change how much work the AI component has to do. Also, the amount of compute effectively supporting a model can affect how much consumption is required to achieve the same result. Cronin argues a defensible AI forecast today needs ranges and scenarios rather than a single false-precision number: > Start with workload and business demand, questioning what the use cases are, how often they will run, and what value they are expected to deliver. Then model the technical variables: model choice, token volume, infrastructure, architecture, and expected growth. These create thresholds that trigger review as actual consumption diverges from assumptions. Budgeting for an AI workload and the business outcome attached to it is much more realistic than putting a number on ‘x’ amount of money that will be spent on tokens. Think culture, not tools Each new wave of technology has also driven the development of new financial management technology, which can help enterprises find the right balance between technology innovation and financial discipline. But improving visibility into wasted spend is the easy part. The hard part is transforming the culture. Cronin argues that 25 to 30% of technology spend has been wasted for a decade. For example, countless companies have software and SaaS debt of licenses and subscriptions shelved unused, or tools that can complete the same actions across different teams with little reconciliation or overlap. He says: > Organizations are strong at identifying savings, but much worse at actually making the organizational decision required to realize them. The emerging AI equivalent of a hundred idle Oracl