Business · Companies · 5 hrs ago
Companies cap AI spending as autonomous tools drive up costs
Uber reached its annual budget limit for artificial intelligence within four months of 2026, according to its technology chief.
Companies including Uber, Meta, Cisco and Walmart have set spending caps as they adopt AI agents, which can carry out complex tasks with little human input.
These agents can use large amounts of computing power, and their costs can be hard to predict.
One Amazon project reportedly went 860% over budget after unfinished tasks kept running in the background for five months, costing more than $1 million.
Some AI services have moved from fixed subscriptions to charges based on use, leaving customers exposed to higher bills when tasks run longer.
Most charges are based on computing use rather than the value of the work completed, prompting calls for pricing tied more closely to results.
How companies and AI providers balance result-based prices with sustainable revenue matters as the industry grows and investors weigh potential stock market listings by Anthropic and OpenAI.
Uber had reached its annual artificial intelligence spending limit just four months into 2026.
Companies including Uber, Meta, Cisco and Walmart have responded to rising AI costs by setting spending caps.
Autonomous AI agents can consume large numbers of tokens with little human oversight, making costs harder to predict.
An Amazon project exceeded its budget by 860% after unfinished tasks continued running in the background for five months.
AI companies’ shift from fixed subscriptions to usage-based pricing has transferred more cost risk to customers.
- Who
- Companies adopting AI agents, including Uber, Meta, Cisco and Walmart, and the AI labs that provide their services.
- What
- Companies are capping AI spending as autonomous tools make costs harder to predict and control.
- When
- By early 2026, some enterprise services had moved to usage-based pricing; Uber had reached its annual limit four months into the year.
- Where
- Not stated
- Why
- AI agents can run complex, prolonged tasks and consume large amounts of tokens with little human oversight, while usage-based pricing exposes customers to higher costs.
This story does not have two clearly opposing sides.
We prepare budgets based on what we know, but things change rapidly in this space.
Charging by tokens makes no sense.
The GPT-3.5 model that helped trigger the ChatGPT revolution had an initial maximum context length of 4,096 tokens.
Some enterprise AI services moved from fixed subscriptions to usage-based pricing without usage limits.
Uber had reached its annual AI spending limit.
Unfinished tasks in an Amazon project continued running in the background, causing the project to exceed its budget by 860%.
- Uber AI spending
- Annual limit reached four months into 2026
- Amazon project
- Budget exceeded by 860%
- Amazon project duration
- Unfinished tasks ran in the background for five months
- Affected companies
- Uber, Meta, Cisco and Walmart set spending caps
- GPT-3.5 context length
- 4,096 tokens initially
Sources
Related news
Philip Morris aims for smoke-free products to make up over 66% of revenue by 2030
Dollar holds near 18-month high as markets await US inflation data
EU banks cut financing for coal companies by 46% in four years
US-China supertanker freight cost falls from record
Conflict around Hormuz drives up shipping costs beyond oil
Asian Markets Open Mixed as Investors Watch Middle East Tensions and US Earnings
Capitalism and Socialism Debate Who Should Guide Growth and Distribution
Oil prices fall on hopes of Russia-Ukraine energy ceasefire
Asian shares cautious as earnings season approaches and oil prices slip
Investors turn to rental housing in Japan and Australia as costs rise
Gold Edges Higher as Traders Assess Trump’s Energy Ceasefire Claim
