Technology companies globally are becoming more tight-fisted about spending on artificial intelligence (AI) tools. Tesla recently capped employee spending on AI tools at $200 a week. Uber imposed a monthly limit of $1,500 per AI tool after exhausting its annual budget for Anthropic's Claude Code in just four months. Microsoft cancelled Claude Code licences for one of its engineering teams.
These moves signal a new phase in enterprise AI adoption. Businesses are questioning if productivity gains justify the rapidly rising bills. The debate is spilling into pricing models, infrastructure investments, hiring decisions and even geopolitics, as companies and governments search for ways to make AI productive and sustainable.
Token trap
Enterprises are pouring money into AI. Global IT spending is expected to reach $6.31 trillion in 2026, fuelled by investments in AI infrastructure, software and cloud services, according to projections by Gartner in April. The share of US businesses with paid subscriptions to AI models, platforms and tools has increased from 7.5% in 2023 to over 54% now, according to Ramp AI index data. It also says the spending on AI among its customers increased 13-fold over the past year, with the biggest spenders averaging $7,450 per employee each month.
Many companies initially treated token consumption as a proxy for productivity. Meta even maintained an internal leaderboard that rewarded heavy users. However, higher cost and other issues are prompting a rethink. Research by Jellyfish found that developers with the largest token budgets produced twice as many code submissions, but at ten times the cost. Similarly, CodeRabbit found AI-generated code created more problems than human-written code. Executives at Uber, Meta and Palantir have since argued that AI should be judged by measurable business outcomes, not token counts. The restrictions on token use are expected to move businesses in that direction.
Agent overload
As enterprises move beyond chatbots to AI agents, costs are rising even faster. Around 70-90% of enterprises are already experimenting with AI agents, according to a report by Goldman Sachs, citing McKinsey and PwC surveys. Tata Sons chairman N Chandrasekaran has said Tata Consultancy Services will have more AI agents than employees in three years. Goldman Sachs estimates AI agents will consume over 100 quadrillion tokens a month by 2030.
Unlike conventional chatbots that respond to individual prompts, AI agents can plan, execute and monitor multi-step tasks with minimal human intervention. As a result, they consume more tokens. Researchers at the University of Michigan, Stanford University and other institutions found that an AI agent consumed roughly 1,200 times more tokens than a coding chat on average.
Companies are now tweaking their Agentic AI strategies. A KPMG survey of 2,145 executives showed nearly half of them pulled back the use of AI agents where costs exceeded expected benefits. Some are adopting model-routing software that directs workloads to the most cost-effective AI models instead of treating all tasks equally.
Pricing shifts
In recent months, AI providers have overhauled how they charge customers. In April, OpenAI moved its Codex tool from per-message pricing to API token usage. Around the same time, Anthropic began billing business customers for actual token usage once they exceed the credit limits in their subscription tiers. In May, Google replaced Gemini's daily prompt limits with a compute-used model. In June, Microsoft shifted GitHub Copilot to usage-based billing.
Investor expectations are driving the change. OpenAI and Anthropic are preparing for initial public offerings (IPOs). Other tech companies face close scrutiny of their returns. All have poured money into AI infrastructure: OpenAI has committed more than $1.4 trillion to infrastructure over coming years. Anthropic plans to invest $50 billion in US data centres and pay SpaceX $1.25 billion a month for computing capacity.
Alphabet, Meta, Microsoft and Amazon together expect to spend over $700 billion on capital expenditure in 2026. Recovering these costs is reshaping pricing. For enterprise customers, it has made AI spending harder to predict.
Headcount lever
Rising spending on AI is coinciding with another wave of layoffs across the technology sector. A total of 219 companies have laid off over 119,000 employees so far this year, compared with 125,000 in the whole of 2025, according to layoff.fyi. In May alone, technology companies announced 38,242 job cuts as they redirected spending towards artificial intelligence. Executives increasingly acknowledge that these cuts are frequently aimed at funding expensive AI deployments rather than reflecting genuine automation gains.
CloudBees chief executive Anuj Kapur said reducing headcount is often “the only lever they can pull” to offset growing AI bills. OpenAI chief executive Sam Altman has also warned against “AI-washing”—using AI as a convenient explanation for layoffs that companies had already planned.
Whether the trend persists remains unclear. Some companies have already reversed course after discovering that automation alone could not match human performance. Klarna rehired customer service staff after AI-driven support led to a decline in service quality. Ford Motor similarly brought back and promoted 350 experienced engineers.
Sovereignty push
The economics of AI are also reshaping geopolitical competition, as governments seek to reduce dependence on a handful of US model providers. China has emerged as the cost leader. Its AI models are significantly cheaper than those offered by OpenAI and Anthropic. In May, DeepSeek cut prices for its flagship V4-Pro model by 75%, widening the cost gap with leading US rivals.
Lower costs stem from abundant domestic electricity, efficient model architectures and optimized use of computing resources. Companies are noticing.
AI startup Lindy said switching to DeepSeek reduced inference costs by millions of dollars. China is investing $295 billion over five years to build a national computing backbone that pools computing capacity across the country, according to a Bloomberg report. The growing adoption of Chinese open-source models abroad also advances Beijing's ambition to shape global AI standards.
European companies are increasingly spreading workloads across US, European and Chinese models to reduce cost and strategic risks. The European Union is pursuing policies to lessen dependence on American AI providers. In India too, the calls to pursue AI sovereignty is growing more intense.
www.howindialives.com is a database and search engine for public data.