Jeff Clarke’s keynote at Dell Technologies World 2026 put the operating model and tokenomics in context. His big prediction is that token consumption will become a line item on the income statement and will be a fundamental part of a new AI operating discipline within organizations. Early in Clarke’s keynote he hit the audience with two key stats:
- Cost per token was down roughly 80% this past year;
- Token consumption was up 320x.
Clarke’s point is best understood in classical economic terms, first put forth by William Stanley Jevons in his book “The Coal Question.” While Jevons didn’t coin the phrase Jevon’s paradox – it was a key chapter in this book which led to the idea that as unit costs collapse, usage explodes, and total spend still rises. Clarke tied the demand curve directly to enterprise value suggesting that by 2030 “token consumption will explode,” and the job for operators is how you can most cost effectively generate the tokens that you’ll need for the long term. The point is that when AI factories manufacture intelligence in the form of tokens that drive business value, CFOs will manage them the way they manage cloud COGS – except this time the bill can move faster because the workload moves faster; and the token spend can be more directly tied to revenue.
Two other pieces from Clarke’s keynote are relevant for investors and enterprise buyers:
- Clarke explicitly introduced token routing as an infrastructure decision – i.e. where to put the tokens becomes one of the most important choices enterprises will make (perhaps a bit self-serving but an important proclamation nonetheless);
- He also suggested that every action needs a “receipt” so you can prove what an agent did, why it did it, and how to undo it if it got it wrong. This speaks to governance, auditability, and trust, all packaged as part of operating costs.
The takeaway is that Dell is trying to get decision makers to think about AI investments beyond innovation budgets and into the same operating model enterprises use for production systems. This aligns with Jensen’s Pareto curve thesis and is why Clarke leaned into time to first token and fast deployment as a competitive variable for Dell and its customers.
We believe Clarke’s two most important concepts were: 1) Turning token consumption into financial value. Once tokens are budgeted like power and cloud, the conversation changes from AI strategy to AI as driving your business model; and 2) This leads to a new operating model, where scaling with substantially less labor becomes possible as agents begin to act on a single version of organizational truth.
The operating model becomes the product
The most profound part of Clarke’s keynote – and follow on roundtable conversations with Clarke and other executives including Doug Schmitt, Dell’s CIO – is that the operating model can’t be delegated to an outsourced team. It has to be owned, driven, and validated by the leadership team. This isn’t “command and control” but it is leadership driven with hands-on involvement that is explicitly business-led, not IT-led. The goal is outcomes, which enables cross-functional work and forces the organization to stop treating AI as use case by use case.
We believe this requires a shift in workflow thinking. The point is not to automate existing workflows A-to-Z. We call that “paving the cowpath.” Rather it calls for reimagining what the workflow should be in an AI environment and making it nonlinear across domains – services into product engineering into sales, tied to customer value. That’s a very different posture than the old plan-design-implement-manage (PDIM) lifecycle and increasingly moves organizations toward what we call service-as-software (SaSo).
Leading AI practitioners, of which Dell is proving to be one, say senior execs should plan to initially spend a day a week deep diving with teams, on whiteboards, stitching agents together, looping in security, legal requirements, running reinforcement learning and ensuring business resilience. Our advice is you can’t outsource this, it’s your business . The promise of AI is operating leverage that outcome requires doing the hard integration work.
In our opinion, one of Dell’s differentiators goes beyond its end-to-end infrastructure and ability to ship systems. It’s the “customer zero” aspect and the company’s experience with AI as a practitioner, moving from prioritizing thousands of use cases, choosing the ones that matter, experimenting with RAG-based chatbots, realizing the limits of such early approaches and iterating into agentic AI. Dell, led from the top of the company, and is explicitly forcing a new operating mode into its business by investing the time and energy into “AI-ifying” its business.
Client zero as proof
The reason Dell has credibility on operating model talk is they’re living it. We’d observe that Dell’s internal modernization is focused on simplifying, standardizing, and automating, then applying AI with discipline and intent. What we uncovered in private conversations is that initially Dell was driving toward standardization around its software products, rationalizing duplication and limiting internal software development. Over the past two years that has changed. Dell is now actively and aggressively writing its own software as part of its transformation. This in our view is a fundamental principle of service-as-software.
This is also where the keynote idea of a receipt gets traction. In the Jeff Clarke analyst roundtable, the conversation turned to securing, logging, and ultimately understanding actions so you can back it out afterward – the “digital receipt” of what agents did and why. This is more than feature – it’s an operational trust layer, similar to what Veeam’s Anand Esweran put forth in his VeeamON Keynote.
The point is a lot of enterprise processes are still manual workflows with heavy human intervention. Often crudely documented and not well understood by all. Now the push is to digitize, possibly refactor and map processes so agents can operate against them. That’s the muscle memory Dell is trying to bring to customers – specifically the messy middle.
Considerations and keys to watch
We believe client zero is Dell’s credibility pass to push beyond product launches into operating model advice. Most vendors talk about agentic outcomes. Dell is showing how the hard parts play out in practice – i.e. governance, logging, security, business resilience, and the executive time required to make it real. That’s why we think customers should listen.
Clarke is building a straight line from token routing to agentic receipts to operating model to the P&L. The keynote and analyst roundtables are aligned on the single idea that enterprises will not scale agents without trust, and they will not pay for trust without a measurable operating model that reduces friction and risk.
In our view, Dell’s opportunity is not just selling more racks– that’s how they make money. But Dell is more strategic than ever as a supplier. It’s playing a key role in the how of token-era operations – the routing decisions, the receipts, and the executive commitment needed to rewire workflows. In addition, it is becoming an ecosystem leader well beyond silicon suppliers and distributors of its hardware. If Dell can keep proving it internally and packaging it credibly for customers in easier to use solutions, the company’s opportunity will not only be tied to product cycles, but increasingly a a platform narrative with an operating model attached.
Action item
We believe CIOs should treat Jeff Clarke’s “tokens move from PoC to the P&L” premise as an operating-model shift, not a cost-optimization exercise. That means instrumenting token production and token consumption the same way the enterprise tracks cloud spend today – but tying it directly to business processes and top line revenue. Agents and model-driven workflows need a measurable token bill (cost per outcome, not just cost per token), with observability that makes them auditable – i.e. who invoked what, what data was touched, what action was taken, and what the approved recovery path is if it something goes wrong.
Importantly, tokens will hit both sides of the income statement. The cost side is obvious (compute, api calls, orchestration, storage, security). The revenue side is the real AI upside. Tokens are the throughput that powers new products and services, faster cycle times, higher conversion, fewer abandoned transactions, better upsell, and entirely new AI-mediated services. The discipline CIOs need is a value ledger where token consumption is mapped to revenue impact, margin impact, and productivity impact by use business outome. If an enterprise can’t explain where the tokens went and what they returned, it’s not ready to run tokenomics as a line item – it’s just running experiments with a bigger bill.
If you get this right, your technology spend will double or triple as you scale with less labor. Your business will begin to see software-like marginal economics, which should increase the value of your enterprise as you scale.
