Amazon.com Inc.'s cloud business is observing a clear shift in how corporate customers use artificial intelligence, with more companies moving beyond model training to embed those models into everyday operations via inference, according to the unit's chief executive.
"We still see large training clusters being used by a number of companies, but as these models get really popular and really powerful, more and more companies are integrating that inference into their workloads," Amazon Web Services Chief Executive Officer Matt Garman said on the Bloomberg Tech television show.
Training denotes the compute-heavy phase of creating an AI model. Inference is the subsequent stage in which that trained model is applied to respond to prompts or perform tasks as part of business workflows.
In response to growing infrastructure needs, Amazon last week raised its planned capital expenditures for 2026 to $220 billion, up from an earlier forecast of $200 billion. The company said the higher projection reflects rising costs for memory and other components required to run data centers.
Garman did not lay out a spending projection for the following year, but called the potential for Amazon's artificial intelligence business "just massive." To help plan for new facilities, Amazon is securing extended customer commitments. "As we're investing, we're getting five-year commitments from customers," Garman told Bloomberg Television. "Today, demand still significantly outstrips supply and we're trying to build and invest to keep up with what customers are asking for."
Investor response has been positive in recent sessions, with Amazon's shares lifting the company's market capitalization above $3 trillion. The company also reported cloud-computing revenue growth for a fifth straight quarter.
Context and implications
Amazon's comments underscore two linked developments that affect its cloud business: a change in customer workload composition toward inference, and an escalation in the infrastructure investment required to support that shift. The larger capex plan and long-term customer contracts are presented as measures to align supply with accelerating demand.
Data limitations
The company did not provide a forecast for next year's spending level, and details on the specific mix of components driving cost increases were not disclosed beyond higher prices for memory and other parts for data centers.