Dell Technologies vice chairman of products and operations Jeff Clarke sure does love a list. At his keynote address at Dell Technologies World 2026 in Las Vegas he peppered his speech with a number of them. In fact, so popular are his lists, a commemorative T-shirt with one of his lists on it is available as part of the Dell merchandise range!
One topic that came up multiple times this year was the use of tokens with Clarke saying that many organisations are dramatically underestimating the impact that AI token consumption will have on their operations, budgets, and technology infrastructure.
Clarke described a rapidly emerging reality in which tokens (the units used to process and generate AI interactions) are becoming a critical business resource, potentially reshaping enterprise cost structures in much the same way as compute, storage, and bandwidth did in previous technology eras.
Clarke said the use of tokens had exploded in the past year with costs falling by around 80% year-on-year. What that has led to is token consumption up 10x to 100 trillion tokens used in 2025.
“Token use is exploding. Token use for reasoning is up 320 times. If this pattern sounds familiar, it should. We’ve seen this before in bandwidth, storage, compute for decades,” Clarke explained. “Cheaper units unlock so much new usage that the total spend goes up.”
What makes the current AI cycle different, he said, is the unprecedented speed at which the change is occurring.
Dell is already seeing examples of engineering teams consuming monthly token allocations within a matter of hours rather than weeks.
“And that’s not because anything was broken. It is because it’s working. This is what success looks like with agents, and you better be ready for it, because it’s coming,” he told the large audience.
The rise of AI agents is accelerating this trend. Unlike traditional AI assistants that simply answer questions, agents perform tasks, execute workflows, analyse information, and repeatedly interact with models to complete objectives. Every interaction consumes tokens. The result is that token consumption is rapidly becoming a strategic business issue rather than merely a technical metric.
Perhaps Clarke’s most significant observation was that tokens are beginning to replace traditional labour costs for certain work.
“If AI already has it, tokens are about to be aligned in your P&L,” he said. “Agents are doing in minutes and hours what took multiple sprint teams days and weeks through writing code, testing code, deploying it and planning the next generation iterations.
“That’s not a 30% productivity gain, that’s a 10x gain, that’s 100x gain, and often times even greater than that. That’s structurally a different company.
“As the agents take on more of this cognitive work, cost migrates from head count to tokens. Tokens are going to be much greater in the rounding error in the future, and the companies who are planning for this shift now are going to have a big advantage over everyone that waits.”
For many organisations, this will require a complete rethink of how AI spending is measured and managed, he said.
“As token consumption rises, organisations will face increasing pressure to ensure workloads are being processed on the most appropriate infrastructure and AI models,” he said.
This has given rise to what Clarke described as one of the most important emerging disciplines in enterprise AI: token routing.
Under this approach, organisations must determine which AI workloads should run on edge devices, which belong in private datacentres, and which should be processed in public cloud environments.
“The question isn’t whether consumption grows,” Clarke said. “The question is, are you running the right tokens on the right infrastructure?”
Clarke warns that not every workload requires access to the largest and most expensive frontier models. Routine summaries may be handled by smaller models running closer to users, while complex reasoning tasks may justify more powerful infrastructure.
“You need to think the right workload, right model on the right tier,” Clarke said.
He said Dell believes organisations that fail to develop this capability risk significant cost overruns as AI usage scales.
“Run everything in one place because it’s easy, be ready for a surprise,” Clarke warned. “And that surprise is a large bill that’s only going to get larger.”
For Clarke, the emergence of token economics represents more than a technical challenge. It is becoming a defining characteristic of the AI-native enterprise.
As AI adoption accelerates and autonomous agents become more deeply embedded into business processes, organisations will increasingly need to manage tokens with the same discipline they apply to traditional infrastructure resources.
“The notion of token routing, where to put that token, is going to be one of the most important infrastructure decisions you will make,” Clarke said.
Leave a comment