Cut AI Token Costs, Not Your Team: Smarter Spending Can Protect Jobs

 

As companies pour billions into artificial intelligence, many executives are facing a difficult question: should they reduce their workforce to pay for growing AI costs?

Nvidia Chief Executive Jensen Huang believes engineers should be using AI aggressively. Speaking on the All-In Podcast after GTC 2026, Huang said he would be concerned if an engineer earning $500,000 annually consumed less than half that amount in AI tokens.

Nvidia is reportedly preparing to spend as much as $2 billion per year on AI token usage for its engineering workforce.

His statement reflects a broader shift in corporate spending. Money that once went mainly to salaries is increasingly being redirected toward AI infrastructure, cloud computing, and token-based services.

The world’s four largest hyperscale technology companies are expected to spend about $700 billion in combined capital expenditure in 2026. At the same time, AI has become one of the most frequently cited reasons for job cuts in the United States.

However, reducing headcount to fund AI investments does not automatically improve business performance.

Cutting Employees Does Not Guarantee AI Returns

Research firm Gartner surveyed 350 executives from companies generating more than $1 billion in annual revenue. All of the companies were deploying AI agents or automation systems.

Around 80 percent had already reduced their workforce.

However, Gartner found no clear connection between those job cuts and improved financial returns.

Analyst Helen Poitevin explained that workforce reductions may create temporary budget space, but they do not automatically generate meaningful returns from AI.

The findings suggest that many companies may be solving the wrong problem.

Instead of treating AI spending as fixed and employees as expendable, businesses should examine how their token consumption can be reduced through better engineering.

Uber’s AI Budget Problem

Uber reportedly provided AI coding tools to around 5,000 engineers in December. By April, the company had already exhausted its entire AI budget for 2026.

Around 70 percent of newly committed code was reportedly AI-generated. Despite that high level of adoption, Uber Chief Operating Officer Andrew Macdonald admitted that the company had not yet established a clear link between AI-generated code and improvements customers could actually experience.

Following the budget overrun, Uber introduced a monthly AI spending cap of about $1,500 per engineer.

The experience demonstrates that high AI usage does not automatically equal higher productivity, better products, or stronger business results.

Companies still need clear objectives, spending controls, and performance measurements.

Where Companies Can Reduce Token Spending

AI token budgets are more flexible than many businesses assume.

One of the simplest cost-saving measures is prompt caching.

Prompt caching prevents companies from repeatedly paying to process the same instructions, reference documents, and background information. Static content can be processed once and reused at a lower cost.

Published pricing from OpenAI and Anthropic indicates that prompt caching can significantly reduce the cost of repeated input processing.

Security company ProjectDiscovery reportedly increased its cache hit rate from seven percent to 84 percent by restructuring its prompts. The improvement reduced its total large language model spending by between 59 and 70 percent.

The company served about 9.8 billion tokens through its cache.

That level of savings shows that prompt design and system architecture can recover more money than some workforce reduction programs.

Use the Right Model for the Right Task

Another major source of unnecessary spending is the use of powerful flagship models for simple tasks.

Routine classification, summarization, data extraction, and formatting do not always require the most expensive model available.

Smaller models can often perform these tasks at a fraction of the cost.

Companies can also use intelligent model routing, where simple requests are sent to smaller models while complex tasks are reserved for more advanced systems.

Batch processing can further reduce costs for workloads that do not require immediate responses.

For example, businesses processing reports, documents, transcripts, or large data sets overnight may not need real-time AI responses. Batch processing can therefore provide substantial savings.

Retrieval and Prompt Compression Can Lower Costs

Retrieval-augmented generation, commonly known as RAG, can also reduce token consumption.

Instead of sending an entire database or long document to an AI model, a RAG system retrieves only the most relevant information for each request.

This reduces unnecessary input while improving the accuracy and relevance of AI-generated responses.

Prompt compression offers another opportunity.

Many organizations use prompts filled with repeated instructions, excessive examples, and unnecessary context. Removing redundant text can lower costs across thousands or millions of AI calls.

Open-weight models may also help companies handle routine workloads more affordably, although organizations must be prepared to manage their own infrastructure, security, maintenance, and model performance.

These solutions are not revolutionary. They are simply examples of disciplined technology management.

Companies routinely optimize electricity, cloud storage, bandwidth, and computing resources. AI tokens should be managed with the same level of care.

Human Workers Remain Essential

Reducing token costs only matters if the savings are used productively.

Evidence increasingly suggests that companies achieve stronger results when AI is used to support employees instead of replacing them entirely.

Klarna became one of the most visible examples.

The financial technology company replaced around 700 customer service positions with an AI-powered assistant. The system handled large volumes of customer inquiries, but customer satisfaction reportedly declined.

Klarna Chief Executive Sebastian Siemiatkowski later acknowledged that the result produced lower service quality and was not sustainable.

The company eventually moved toward a blended workforce model.

AI now handles repetitive and routine requests, while human employees manage complex problems that require judgment, empathy, negotiation, and contextual understanding.

Gartner expects more companies to follow this approach. The research firm predicts that by 2027, half of the companies that reduced customer service staff because of AI may begin hiring people again.

Companies Risk Losing Their Future Talent

Another major concern involves entry-level workers.

Research from Stanford University’s Institute for Human-Centered Artificial Intelligence reportedly found that employment among software developers aged 22 to 25 fell by almost 20 percent compared with 2024 levels.

Employment among older software developers continued to grow.

This trend creates a long-term workforce problem.

Junior developers do more than complete basic programming tasks. They learn company systems, gain practical experience, develop judgment, and eventually become senior engineers, technical leads, and AI system managers.

If companies eliminate entry-level positions today, they may struggle to find experienced professionals capable of directing advanced AI systems five years from now.

AI can generate code, but it cannot independently create a sustainable talent pipeline.

Leadership, Not Layoffs

A company that reduces its AI token spending by 50 or 60 percent may suddenly have enough budget to retain employees, hire junior talent, improve training, and strengthen customer service.

Whether that money is reinvested in people is ultimately a leadership decision.

The biggest winners in the AI economy may not be the companies that buy the most tokens or eliminate the most jobs.

They may be the organizations that manage AI spending intelligently, assign the right model to the right task, reduce waste, and preserve the human expertise needed to turn artificial intelligence into real value.

The smarter strategy is not to shrink the team just to expand the token budget.

It is to shrink the token budget through better engineering, then invest the savings in the people who make AI useful.