AI token budgeting is moving from an engineering footnote to a finance function. According to Dell Technologies CFO David Kennedy, most chief financial officers now give token spend its own line item and track it the way they track any other operating cost. The goal isn’t to starve AI projects. It’s to make the spending visible so teams can move faster with it.
Kennedy made the case in a conversation with theCUBE Research’s Dave Vellante at the Dell Technologies New York Symposium, as first reported by SiliconANGLE. His views on AI token budgeting land as enterprise AI shifts from experiments into production, and as finance leaders try to work out what that shift should cost.
Key Takeaways
- AI token budgeting has shifted from engineering to finance, allowing CFOs to track spending as a dedicated line item.
- Tokens represent the cost of using large language models, making it essential to manage AI spending as usage increases.
- Financial services lead the way in adopting AI token budgeting, with expectations that other sectors will follow.
- Effective AI token budgeting involves sorting workloads by predictability and reviewing outcomes versus token consumption.
- The real value of AI token budgeting lies in improving speed and productivity, not just controlling costs.
Table of contents
Why AI token budgeting now sits with finance
Tokens are the unit of work for large language models, and every prompt and response consumes them. When a handful of pilots run, the bill is easy to ignore. When AI runs inside daily workflows, consumption climbs and the cost becomes hard to wave away.
Kennedy’s answer is plain accounting discipline. Give token spend a dedicated line, watch it over time, and assign ownership. He said IT and finance can work together to set guardrails around it. In his view the spending needs managing, but it isn’t the main story.
That framing matters for anyone drafting a 2027 plan. A tracked line item turns a vague worry about “AI costs” into a number that someone can explain to a board. AI token budgeting also gives teams something to improve against, which is harder when usage is buried inside general cloud or software bills.
Financial services got there first
Not every industry is moving at the same pace. Kennedy pointed to financial services, and high-frequency trading in particular, as an early group to push AI out of pilot mode and into production. He expects pharmaceutical companies to be next. For other sectors, these early movers offer a preview of what AI token budgeting looks like once AI is a working part of the business.
Dell offers a few numbers to show the broader trend. The company counts 6,500 AI factories among its customers. In May, it expanded its AI Factory offering with Nvidia, aiming to help more customers close the gap between an experiment and a result. Kennedy also noted that Dell has just come off another very strong quarter driven by AI servers, and that demand is spreading beyond GPU-heavy systems into traditional servers and storage.
That last point is worth a second look for infrastructure buyers. If AI workloads are pulling along conventional compute and storage, budgets for those categories may need revisiting alongside the AI line. Kennedy didn’t give a forecast, so treat it as a direction rather than a prediction.
Where the model runs shapes AI token budgeting
One of the more practical ideas from the interview concerns model placement. Kennedy said repetitive finance workflows can run on open-weight models hosted on-premises. Doing so converts a variable, usage-based token cost into a fixed one.
For a finance team, that’s a familiar trade. Variable costs scale with demand and can surprise you. Fixed costs are easier to forecast, though they commit you to capacity whether or not you use it. Routine, high-volume tasks tend to suit the fixed route because their demand is steadier, though Kennedy didn’t spell out the break-even math.
A sensible AI token budgeting process would therefore sort workloads before it sets limits. A few questions help:
- Is the task repetitive and predictable, or exploratory and spiky?
- Does the work need a frontier-scale model, or will an open-weight model do?
- Who owns the spend, and how often is it reviewed?
- What outcome, in time saved or work completed, justifies the tokens consumed?
Efficiency beats the price of a token
Kennedy was careful to say that the price per token isn’t the point. What counts is which tokens a company spends and whether it spends them well. An organization that burns tokens on low-value requests will see poor returns at any price, while a disciplined one can get strong outcomes from the same budget.
This is where AI token budgeting differs from ordinary cost cutting. Trimming a number across the board is easy and often counterproductive. Asking what each category of usage produces is slower work, but it points spending toward the places where AI helps most.
It also changes how finance and technology teams talk to each other. Instead of finance asking IT to spend less, both sides can ask which uses deserve more and which deserve less. That’s a healthier conversation, and it’s the one Kennedy seems to want CFOs to have.
Speed is the real return
Kennedy’s central argument is that cost control is necessary but secondary. The real value, he said, comes from the speed and productivity a company gains. In that sense, AI token budgeting is a means to an end. He also said there are now enough proof points that AI works and delivers efficiency, and that the open question is whether companies can do it at scale.
It’s worth remembering who is speaking. Dell sells the servers, storage and AI Factory systems that enterprises buy to run these workloads, so it has a commercial interest in rising AI adoption. SiliconANGLE’s disclosure notes that theCUBE is a paid media partner for the symposium, though it says sponsors have no editorial control. Readers should weigh the advice with that context in mind. Even so, the budgeting steps Kennedy describes are ordinary financial practice, and they don’t depend on buying any particular product.
Read Next
A few related pieces worth your time:
- Paid Antivirus Is Hard to Justify on Windows PCS in 2026
- RIP Samsung Messages: Gone for Good in July 2026
- Claude AI Surges Past ChatGPT: Paid Users Double in 2026
What to watch next
Several things are worth tracking over the coming quarters. One is whether AI token budgeting and its dedicated line item become standard in budget templates and board reporting. Another is how many finance teams move repetitive work onto on-premises open-weight models, and whether the shift to fixed costs delivers the predictability it promises. A third is whether the pharmaceutical sector follows financial services into production, as Kennedy expects.
For finance leaders, the starting point is modest. Create the line, assign an owner, sort workloads by how predictable they are, and tie usage to outcomes. AI token budgeting won’t produce results on its own, but it gives a company the visibility to spend on AI with confidence and the room to focus on speed.
Source: SiliconANGLE
