Firms Cut Jobs to Cover Rising AI Token Bills

Companies are financing rising AI token bills through workforce reductions while engineering teams and vendors report ways to lower those costs. Nvidia is targeting a $2 billion annual token bill for its engineers, and security firm ProjectDiscovery raised its prompt cache hit rate to 84% and reported a 59–70% cut in large-language-model spending.

Nvidia chief executive Jensen Huang made the comments on a podcast at the close of the GPU Technology Conference in 2026, warning he would be “deeply alarmed” if a $500,000 engineer used less than half their salary in annual AI tokens. He confirmed Nvidia is working toward a $2 billion yearly token bill for its engineering force.

A survey of 350 executives at companies with more than $1 billion in revenue found about 80% of firms deploying AI agents or automation reduced headcount. Analyst Helen Poitevin commented, “Workforce reductions may create budget room, but they do not create return.”

Uber provided AI coding tools to 5,000 engineers in December and exhausted its 2026 AI budget by April. The company’s chief operating officer Andrew Macdonald acknowledged the changes have not yet produced clear customer-facing results. Uber later set a $1,500 monthly token cap per engineer to control spending.

Prompt caching stores static inputs such as system instructions and reference documents so they are not reprocessed on every API call. Published pricing from major providers indicates prompt caching can cut repeated-input costs by up to 90%. ProjectDiscovery said it raised cache hits from 7% to 84%, served 9.8 billion tokens from cache and reduced total LLM spending by 59-70% as a result.

Other technical measures include routing routine tasks to smaller models when high-end models are unnecessary; providers’ flagship models can cost about five times more per token than lighter alternatives. Batch processing can roughly halve per-token cost for non-real-time workloads. Retrieval-augmented generation narrows the context sent to the model to only the relevant slice of a knowledge base, and prompt compression removes redundant examples that increase each call’s token count. Teams that run open-weight models on their own infrastructure can lower costs further for routine tasks.

A fintech that replaced roughly 700 customer service roles with an AI assistant reported a decline in service quality and moved to a blended model with AI handling routine volume and humans managing complex cases. The fintech’s chief executive Sebastian Siemiatkowski remarked, “The result was lower quality, and that’s not sustainable.”

A study from Stanford’s Institute for Human-Centered AI found employment for software developers aged 22 to 25 fell nearly 20% from 2024 levels while older cohorts increased, a change that reduces entry-level hiring and training opportunities.

Some companies have applied engineering controls such as caching, model routing and batching to reduce token spending; others have used layoffs or set spending caps. Reported measures, savings and workforce effects vary by company and team.