[Las Vegas, United States=DigitalToday reporter Jin-ho Lee] Dell Technologies presented “tokenomics” as a key variable in corporate infrastructure strategy in the artificial intelligence (AI) era. It judged that corporate cost structures could be reshaped as token use surges with the spread of agentic AI.

Dell Technologies unveiled an infrastructure strategy for AI-native companies in a keynote speech on the second day of Dell Technologies World 2026 (DTW 2026) in Las Vegas on May 19, local time. The message was to move beyond simply adopting AI and to design in detail which tasks to run on which models and infrastructure.

◆Surging token costs... detailed infrastructure build is key

Jeff Clarke (제프 클라크), vice chairman and chief operating officer of Dell Technologies, reviewed changes in the AI market over the past year. He stressed that token costs have emerged as companies’ biggest concern as use of AI agents that execute tasks directly increases. Prices for AI models have fallen, but total token consumption is growing exponentially as the scope of AI use widens.

Clarke advised companies to focus on tokenomics in this situation. Tokenomics is a newly coined term meaning that token use generated in the process of using AI determines a company’s cost structure. For companies that provide AI models, cost per token and model efficiency are key. For companies using AI, managing token use across inputs and outputs becomes a core task.

Clarke compared it to the past storage and computing markets. He said that as the unit cost of using such solutions falls, usage can surge and total spending can instead rise. “When the price (of a service or device) goes down, new usage goes up,” he said. “We’ve seen this pattern in storage and computing for decades.”

The keynote introduced a case in which one developer used 10 agents and consumed 1 billion tokens in a day. Another case drew attention in which some engineering groups inside a company used up a monthly AI usage allocation in a short time. Clarke said this was not due to system errors but a result of AI agents being used effectively for actual work.

Dell recommended against funneling all AI tasks into a single high-performance model to solve the problem. It said companies should divide use across local workstations, on-premises data centres, the edge and the cloud depending on the nature and sensitivity of work, performance requirements and cost structure. “The real issue is not whether token use increases,” Clarke said. “The issue is whether you are running the right workloads on the right infrastructure.”

Clarke presented “token routing” as a solution for future AI use. Token routing is a method of deciding which model to run where depending on the nature of the task. Routine summarisation work is handled by smaller AI models, while sensitive financial analysis is run on in-house infrastructure. It is linked not only to cost savings but also to personal data protection and regulatory compliance. “Token routing and the operational burden of tokens will become one of the most important infrastructure decisions,” Clarke said.

◆“Data infrastructure determines token costs”

Dell emphasised data infrastructure as a key factor in lowering token costs. If corporate data is not organised, it is difficult for AI to get the desired answers even while using more tokens. If data is well refined, the same results can be achieved with smaller AI models and fewer tokens.

The “Dell AI Data Platform” could be an answer in this situation. Arthur Lewis (아서 루이스), president of Dell Technologies’ Infrastructure Solutions Group, said the Dell AI Data Platform would be infrastructure that increases token efficiency. The platform refines data scattered across a company into a form AI can use and links it with inference workloads. It processes not only structured data but also unstructured data such as images, video and audio, converting it into data sets AI models can use.

Lewis also proposed designing full-stack AI infrastructure through Dell’s AI factory strategy and AI workstation solutions. Dell unveiled various solutions at DTW spanning storage and servers, security and automation.

“Good data infrastructure is a starting point for improving token efficiency as well as AI performance,” Lewis said. “Companies must prepare data in a form AI can use and be able to run it on the most suitable infrastructure.”