Oracle and AWS are among major cloud and AI vendors pushing tools to control rising enterprise AI costs

Enterprises are increasingly looking for ways to control the cost of deploying generative AI, where the price of using advanced models can rise quickly with how many “tokens” are processed. Tokens are fragments of text that large language models read and produce, and higher token usage generally means more compute, which translates into higher bills for businesses and their cloud providers.

Against that backdrop, multiple major vendors, including Oracle and Amazon Web Services, are working on product and program changes aimed at limiting runaway spending. The immediate concern is not just model access, but also the way organizations are adopting newer “agentic” systems that can run multi-step workflows. Those tools can expand usage and, in turn, increase total token consumption beyond what many buyers initially budgeted.

The cost pressure is also changing how business leaders evaluate AI initiatives. Instead of treating generative AI as a one-time experiment, many companies are moving toward broader deployment across customer support, software engineering, and internal knowledge work. That broader rollout makes cost predictability and cost controls more central, since even small overruns can compound across teams and environments.

In enterprise cloud and AI offerings, vendors typically have to balance two competing goals. They want to encourage higher-quality outputs and faster adoption, but those improvements can increase resource usage. The current push described in industry coverage focuses on giving customers more direct levers to manage spend, including options to reduce waste in workflows and to better align usage with planned budgets.

Oracle, which offers cloud infrastructure and a range of AI services, is one of the vendors identified as part of the larger market response. Amazon Web Services is also named in the reporting. Both are prominent players in enterprise cloud computing, where AI-related charges can include usage-based fees for model inference, storage, and related services that support building and running AI applications.

Still, the coverage provides limited detail on what specific mechanisms Oracle and AWS are rolling out, such as whether they include new discount structures, tighter rate limits, revised billing models, or governance features designed to cap usage per application. The article also does not quantify how much customers’ costs have changed, or specify which industries are feeling the pressure first.

What to watch next is whether these cost-management efforts show up as concrete changes in product packaging and pricing. Enterprises will likely look for clearer controls around agent workflows, more transparency into token and compute consumption, and service-level options that help administrators enforce guardrails without fully restricting innovation.