AI spending stalls as price wars and efficiency gains curb revenue growth

Corporate AI usage is hitting record volumes, yet total spending has stalled, according to new data from financial operations platform Ramp. This divergence suggests that intense price competition and improved customer efficiency are capping revenue growth for major model providers like OpenAI and Anthropic just as they prepare for public markets.

What happened

Ramp’s analysis shows that token spend grew by only 1.5% across August and September 2026, a sharp deceleration from the 42% month-over-month growth seen in July. While October saw a slight re-acceleration of 8.9%, the overall trend indicates that the era of hypergrowth may be pausing. Ara Kharazian, Ramp’s lead economist, notes that the top 1% of AI-intensive firms, often called the most “AI-pilled,” have seen their per-employee spending dip slightly below the $10,000 monthly peak reached in summer. Since this small cohort drives a disproportionate share of total revenue, even a minor stutter in their consumption is enough to flatten overall market growth.

The slowdown is not due to a lack of demand. Token volumes continue to rise, indicating that companies are using AI more frequently. However, the cost per unit of work has dropped significantly. Median firms now spend between $12 and $15 per employee per month, while the top 10% spend $600 to $1,000. Kharazian argues that it is unclear if broader industries, such as healthcare or automotive retail, will ever reach the high spending levels seen in software and finance. Without new, highly expensive use cases emerging to replace the current plateau, revenue trajectories for model providers face significant headwinds.

How it works

The primary driver of this spending stall is a fierce price war among leading AI providers. In September, OpenAI cut API prices by half for its GPT-6 Sol and GPT-6 Luna models. Anthropic responded with significant reductions for Fable 5.1 and Opus 5.5, a move the company admitted slashed its projected revenue growth by approximately $600 million. These cuts, combined with enterprise discounts designed to lock in customers, have transformed the market into a buyer’s landscape. Unlike previous tech monopolies where low introductory prices eventually led to higher costs, the AI sector remains highly competitive with entrants like Meta, Nvidia, and various open-source providers keeping pressure on pricing.

Beyond external price cuts, enterprises are becoming more sophisticated in managing their AI budgets. Companies are implementing stricter defaults to prevent the use of expensive models for long-running tasks and imposing hard budget limits on employees. They are also employing routers to offload simple queries from standard models to cheaper, lightweight alternatives. For instance, tools like Jev allow firms to handle simple yes/no outputs at a fraction of the cost of general-purpose models. As Notion co-founder Akshay Kothari summarized, "tokens go brrr. costs go ahhh," highlighting how increased usage no longer translates linearly to increased spend.

Key details

• Token spend growth slowed to 1.5% in August and September 2026, down from 42% in July.
• OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna in September, prompting similar cuts from Anthropic.
• Anthropic’s pricing reductions reduced its projected revenue growth by about $600 million.
• The top 1% of AI-spending firms are seeing per-employee costs dip below the $10,000 monthly peak.
• Median firms spend $12-$15 per employee monthly, while the top 10% spend $600-$1,000.
• Anthropic’s potential IPO valuation has dropped from $2 trillion to an estimated $1.2–$1.3 trillion.

Why it matters

For software leaders and technical founders, this shift signals a transition from experimental adoption to operational efficiency. The assumption that AI budgets would grow indefinitely alongside usage is breaking down. Instead, organizations are focusing on unit economics, seeking ways to maintain or increase output while reducing cost per token. This environment favors teams that can implement robust routing strategies, monitor usage closely, and select the most cost-effective model for each specific task rather than defaulting to the most capable or expensive option.

Investors and executives should also recalibrate expectations for AI vendor revenue. The path to profitability for model providers is becoming steeper as they compete on price in a crowded market. While new products like personal agents are touted as the next growth engine, experts remain skeptical that they will match the economic productivity of coding agents. Consequently, businesses may find themselves in a stronger negotiating position, able to leverage competition between OpenAI, Anthropic, and open-source alternatives to secure better terms and lower costs.

What you can do

• Audit your current AI spend to identify tasks currently using expensive models that could be handled by lighter, cheaper alternatives.
• Implement model routers to automatically direct simple queries, such as classification or yes/no questions, to low-cost endpoints.
• Set strict budget caps and default model restrictions for employee access to prevent unchecked usage of premium tiers.
• Negotiate enterprise discounts with multiple providers, leveraging the current price war to secure better long-term rates.
• Monitor token volume versus cost trends internally to ensure your efficiency gains match or exceed market averages.
• Evaluate open-source or specialized tools like Jev for specific, narrow use cases where general-purpose models are overkill.