Corporate spending on artificial intelligence is shifting from unchecked experimentation to strict fiscal management as companies move away from flat-rate subscriptions toward usage-based billing. Firms like Coinbase, Salesforce, and Walmart are implementing price caps and internal audits to curb “tokenmaxxing”—the practice of maximizing AI usage regardless of cost—after realizing that unlimited access to powerful models like Anthropic’s Claude and OpenAI’s GPT series can lead to exponential, unsustainable budget growth, according to corporate executives and industry analysts.
Why Are Companies Capping AI Spending?
The primary driver for new AI budget constraints is the transition by major model providers, including OpenAI, Anthropic, and GitHub, to usage-based billing models. According to GitHub’s chief product officer Mario Rodriguez, the previous flat-rate structures were “no longer sustainable” as the gap between simple chat queries and massive autonomous coding sessions widened.
This shift has led to significant sticker shock. A senior software engineer at Deloitte noted that GitHub’s new billing, which took effect in June, has caused developers to burn through monthly quotas rapidly. One highly detailed prompt that previously carried no marginal cost can now exceed $100 under current usage-based pricing, according to the same engineer. Consequently, companies are now prioritizing “hard-nosed utility” over the novelty of AI, as noted by Niranjan Krishnan, head of AI solutions at FPT Americas.
How Are Businesses Managing Their AI Budgets?
Major firms are deploying diverse strategies to control costs while maintaining productivity. Coinbase has introduced a tiered system of weekly price caps, ranging from $500 to $5,000, depending on an employee’s specific role and seniority. Rob Witoff, a Coinbase executive, stated that while the company wants to encourage innovation, it must ensure that usage is intentional rather than wasteful.
Other organizations are taking different approaches:
- Salesforce: CTO Parker Harris reported that while the company has allowed high spending, it is now implementing an “Effective Output” score to measure the tangible return on investment for engineering tasks.
- Walmart: The retail giant has instituted hard usage limits on its internal programming tools.
- IT Consultancies: Companies including IBM, Oracle, and JPMorgan Chase have joined the “Tokenomics Foundation” to standardize how AI usage is measured and budgeted across the industry.
Will Cheaper Models Replace Industry Leaders?
The rising cost of premium models is creating a market opportunity for lower-cost alternatives. As executives look to balance their books, many are offloading basic, repetitive tasks to smaller or open-source models. Ahmad Awais, founder of Command Code, reported that his startup gained 10,000 customers in a single 30-day period, driven largely by demand for more cost-effective AI solutions.
This trend mimics the “Ferrari to the grocery store” analogy used by Harness senior vice president Trevor Stuart; companies are realizing that using state-of-the-art models for simple text summarization is a misuse of capital. While OpenAI and Anthropic are attempting to mitigate these costs through “prompt caching” and more token-efficient model releases, the competitive landscape is widening as firms seek to avoid diverting significant portions of their annual upside into AI infrastructure costs.
Frequently Asked Questions
What is “tokenmaxxing”?
Tokenmaxxing refers to the practice of using high-end AI models for every possible task without regard for the cost of the tokens (the units of data the AI processes). It became a focal point for budget cuts in 2026 as companies realized the behavior was fiscally irresponsible.
Why did AI prices increase in 2026?
Prices rose because AI providers transitioned from flat-rate, subsidized billing to usage-based models. According to GitHub, the previous flat-fee structure was not sustainable as the computational load of autonomous agents grew significantly larger than standard chat queries.
Are companies cutting AI budgets entirely?
No. Most companies are moving toward a “value-based” spend. According to Salesforce CTO Parker Harris, the goal is to forecast spending based on the expected return, rather than simply limiting the use of tools that provide measurable profit or productivity gains.
How is your team handling the shift in AI pricing? Share your experiences in the comments below or subscribe to our newsletter for more industry insights on the future of enterprise software.
