AI companies and enterprise users rethink token costs as infrastructure spending climbs

Traders Union
All News
Business & Wealth
AI token costs & infrastructure

Highlights

Enterprise AI expenditure rose 60 per cent in the past nine months, with median per-employee spending increasing from $2.86 in January 2024 to $11.95 in July 2026.
Tightly managed AI token prices hit a nine-month low at 95 U.S. cents in early October, but agentic workflows are driving much higher overall enterprise compute usage and costs.
Adoption of open-weight AI models grew to 6 per cent of businesses by July 2026 as firms seek savings versus closed-weight models like OpenAI's GPT and Anthropic's Claude.

Token pricing shifts as AI usage deepens

As reported by Financial Times, AI providers are moving away from the subsidized adoption phase and charging more explicitly for usage as training, infrastructure and compute costs mount.

Basic consumer chatbot access often remains free, while paid subscriptions for tools such as Claude and ChatGPT start at about $8 a month for smaller users and rise sharply for heavier use. For more intensive workloads, billing is increasingly based on tokens, the units AI systems use to process and generate text, with costs varying by model sophistication and the amount of computation required.

Input tokens are generally cheaper because reading text requires less compute, while output tokens cost more because generating responses demands repeated processing. Advanced models also carry higher charges even for relatively short tasks, because deeper reasoning needs more computing power.

At the same time, overall token prices have come under pressure as switching costs stay low and customers move between platforms. The article cites Silicon Data's LLM Token Expenditure Index, which fell to 95 U.S. cents in early October, its lowest point in nine months, even as providers such as Google try to create stickier customer relationships by bundling AI services with other offerings.

Enterprise bills rise with agents and heavier workloads

Even with lower aggregate token prices, enterprise AI spending is rising because more companies are using complex workflows and autonomous agents that consume far more compute.

Sara Hooker, co-founder and chief executive of Adaption Labs, says agentic workflows have changed pricing dynamics by making usage longer, more tool-intensive and less predictable. Instead of a single prompt producing a single answer, agents may call multiple tools and continue searching for data until they judge an answer complete, pushing token volumes and bills much higher.

Ramp says spending on AI in the 18 months to June 2026 grew nearly sixfold as businesses consumed more tokens for more complex tasks. The vendor management platform also says enterprises' aggregate spending on AI computing has risen 60 per cent in the past nine months, while median expenditure per employee per month increased from $2.86 in January 2024 to $11.95 in July 2026.

For the top 1 per cent of spenders, median monthly expenditure per employee rose from $755 to $7,400 over the same period. JP Morgan, citing data published in June by Ramp AI, says that group, largely made up of users performing complex coding tasks, consumes 600 times as many tokens as a median user.

That cost pressure is changing procurement strategies. Companies are comparing model providers more closely and increasingly considering open-weight systems such as DeepSeek and Meta's Llama, which can be hosted internally and tuned with proprietary data, as a way to reduce spending compared with closed-weight models such as OpenAI's GPT series and Anthropic's Claude.

Ramp AI Index data shows that more than 6 per cent of AI-adopting businesses used open-weight AI in July 2026, up from 4.5 per cent in January. Even so, enterprises still face the challenge of measuring whether productivity gains from AI, especially from complex agent-based deployments, outweigh rapidly rising compute costs.

In our earlier article on AI agents in large U.S. companies, we covered survey findings showing many CEOs already see efficiency and growth gains, even if only a small share consider agents transformational so far. That piece also noted that adoption is being paired with continued investment in AI infrastructure, along with hiring and retraining plans as firms adapt processes and workflows to make agentic systems work at scale.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.