Enterprise AI costs explode as token pricing sends bills from $200 to $20,000 per month

Why AI tokens will send your enterprise cloud bill sky-high again

Follow ZDNET: Add us as a preferred source on Google. ZDNET's key takeaways * AI usage is moving to token-based pricing. * Token pricing is far more expensive than the previous flat-fee model. * Measuring the value derived from AI remains an unsolved problem. SAN DIEGO -- A few months ago, most people paid a flat fee for their AI access. That was then. This is now. The days of AI pricing as a loss-leader are over. As everyone has discussed here at FinOps X 2026, AI's token-based pricing model is becoming the foundation of the entire generative AI economy, and it's far more expensive than older models. Just ask CoPilot users who are having fits over the new token-based pricing. For many enterprise customers, this reminds them of the early days of cloud pricing when they had to deal with volatile invoices and business models shifting under their feet. Underneath the confusion, tokens are quietly standardizing how labs translate scarce GPU capacity into billable units, how enterprises measure AI usage, and how software vendors reprice their products. Also: Rolling out AI agents? 4 ways to move fast and furious - but with extreme caution Tokens: The atomic units of AI In this new world, the token is the basic unit of AI work. J.R. Storment, executive director of the FinOps Foundation, calls it "the atomic unit of AI." In his FinOps keynote, Storment said that "tokens serve more roles in the modern economy than almost any other commodity has in modern history, maybe, maybe oil in the 20th century." Tokens, he told the FinOps X audience, are simultaneously "the unit of output from all of the hardware and compute and data centers," "how the labs price their outputs and inputs," and "the value unit that enterprises are looking to monetize." That abstraction is precisely why labs and hyperscalers like it. Instead of charging for GPU types, memory, and power directly, they can expose a single unit -- tokens per million -- over a bewildering mix of architectures and deployment topologies. OpenAI, Anthropic, Google, and others now publish per‑model rate cards with separate prices for input tokens (everything you send the model) and output tokens (everything it generates back), usually quoted in dollars per million tokens. Also: Building an agentic AI strategy that pays off - without risking business failure So what are tokens anyway? An AI token, said Storment, "is the smallest unit a word or phrase can be broken down into when being processed by a large language model (LLM)." Before a model can work with text, it breaks it into fragments, a process called tokenization. For English, a common rule of thumb is that "one token is roughly four characters, or about three-quarters of a word," so "100 tokens ≈ 75 words." The token hides enormous complexity. As SAP's FinOps team put it in their session, "You pay per token, and this little token hides an enormous complexity underneath predictability," from model choice and quantization to how aggressively you use caching or agents. That complexity is exactly what FinOps teams are now being asked to decode. The all‑you‑can‑eat token era is over. If 2023 through early 2025 was the era of cheap experiments, the last 18 months have been a rude awakening. Storment describes three distinct phases: The "old days of AI" before ChatGPT, the "good old days of AI" when chatbots "could write some decent code," and then the post‑November‑2025 world when major model releases "took AI from pretty good to really good." In the good old days, the era of all-you-could-eat tokens and subscriptions, we went through a brief period of token maxing. Then everybody was excited about their token leaderboard, which showed who had the most token usage. Today, token leaderboards are painfully obsolete because no one can afford to waste tokens. As Amazon senior vice president Dave Treadwell begged, "Please don't use AI just for the sake of using AI." Objectively, between June and November last year, Storment said global token usage grew in a "nice linear path." Then those new models and agentic patterns landed. Context windows "went from a few thousand or tens of thousands or hundreds of thousands up to millions of tokens in a single conversation," and "agentic hit the scene and exploded," adding "loops and retries and corrections and all this insanity." Also: The autonomous business is coming. Here's why that shift is good news for professionals Companies had happily subsidized that behavior... until they saw the bills. Storment recounted how some "$200-a-month" power users actually cost "upwards of tens of thousands of dollars a month when you were running everything on the latest model." For example, SemiAnalysis, an AI analytics company, recently estimated that a $200 Anthropic plan used to give $8,000 worth of Claude tokens, while a similar OpenAI offering gave $14,000 worth of Codex tokens. Those days and prices are done. Moving forward, companies will have to pay the real cost of AI tokens. "So now what matters more than anything is AI value," Storment told the room. "We've got to bring value back to what we're doing... We're in an era where tokens are the main measurement. We're in an era where tokens are in everything in software, and they're driving a lot of the global token economy." Scarcity keeps token prices from collapsing If Moore's law and hyperscale competition were the only forces at work, you'd expect token prices to keep falling. To some extent, they have. "Since 2023, token prices have fallen dramatically," Storment acknowledged. SAP's internal telemetry tells a similar story. "This is our cost per token over the same time period," said SAP data scientist Maida Nazifi, showing their internal chart. "It's clearly trending down, even with a bit of flattening at the end. And honestly, it matches the narrative that everyone wants to believe, right? Token prices keep on falling." But both stress the caveat: The floor may be in sight. Storment notes that if "you look at the top labs and their pricing, you go back to the Wayback Machine. Token prices have been pretty flat since November 2025," which he links directly to hardware and power constraints: "We can't get enough hardware, we can't get enough power... we're seeing backlogs, we're seeing long commitment periods, and we're seeing shortages." Also: AI agents are getting their own search engine He cited Intel's CEO saying he doesn't expect real relief in GPU and related component supply "until 2028." Nazifi and SAP VP Frederik Pohl are seeing the same patterns at their company: Pohl warned, "We have supply chain constraints, we have hardware prices that are rising, and the prices of new frontier models are growing ever more expensive." The net result is a classic Jevons paradox: Falling unit cost, exploding total spend. "Even with falling token prices, we see that our spend is still rising, and that's the famous paradox," Pohl said. "At our scale, we had unit costs falling, but we saw in some months that spend was doubling." Storment thinks the paradox is just beginning. Goldman Sachs, he said, estimates global usage rising from "6 quadrillion tokens" today to "120 quadrillion forecasted tokens" within about 3.5 years. Even if token prices drop further once supply loosens, they are unlikely to fall 24x as fast as volume grows." FinOps discovers token economics For the FinOps community, which cut its teeth on cloud right‑sizing and reserved instances, token pricing is both familiar and completely alien. The familiar part is that its usage‑based, the invoices are big, and forecasting is hard. The alien part? The unit is tied to language, not infrastructure, and it changes as fast as model releases, not as slowly as server depreciation schedules. Pohl asserted that "AI does not just stretch the cloud playbook, it breaks it; AI is more different from the cloud than cloud was to the data center." Unlike CPUs, "AI models are nothing like that... they have their unique strengths and weaknesses... They have different cost profiles, and swapping out an LLM is not just a pricing decision. It's also a quality-of-output decision." SAP's experience is a case study in how enterprises are retooling. Its Business AI platform, Pohl explained, runs across "multiple different LLMs," including "ChatGPT, Anthropic, Gemini... other open source models," layered on "different hyperscalers." Also: Work IQ is Microsoft's big bet on agent-first enterprise IT, and I have questions When SAP first went looking for AI cost data, "we immediately hit a wall," Nazifi recalled. "The existing [cloud] tools were very blind to the nuance of LLMs, so they could tell us we spent this amount on [a provider], but not really which model, or how much the model. It really was like trying to optimize your gold mining operation by looking at the total weight of ore." So they did it the hard way: "We pulled data manually, we merged data across tables, and then we had this first picture by hand." That picture, once it reached their global infrastructure lead and then the CTO, transformed the conversation. "Within days, it went from like, OK, this is interesting, keep me posted,' to... 'I need this regularly, I need more,'" Nazifi said. Pohl added the FinOps lesson: "If you have a CTO asking for a number, that's not a question, it's a mandate. That demand forced SAP to formalize an internal AI FinOps framework built around three pillars: * Spend visibility: "What we consume, how we consume it, and where we consume it," across models, platforms, business units, and regions. * Economics: "How efficiently are you leveraging AI," measured with token‑level metrics like input/output ratios, cached token ratios, and "token to spend drift" to see whether costs are rising because of volume or mix shifts to pricier models. * Value: Connecting AI spend to business outcomes with "cost per use case" and "inference cost by revenue," so they can tell "which AI features are economically viable" and whether "your AI product margins actually work." "Every token needs to earn its cost," Pohl said, echoing Nvidia CEO Jensen Huang's phrase "token factory effectiveness." That factory spans everything from silicon and data center leases to model routing and prompt design. Tokenomics: beyond just counting tokens If FinOps is about cost control and accountability, tokenomics, at least as the Linux Foundation is positioning it, is about the full lifecycle of tokens as an economic good. Storment defines it as "the emerging discipline of converting energy and capital into AI tokens and resources, consuming those tokens and all the related technology to drive efficient intelligence, and then ultimately drive value on the backend." In his view, that breaks into three buckets: * Production: "Take energy and capital and create tokens," whether in cloud data centers, colos, edge devices, or, as Elon Musk likes to imagine, "data centers in space." * Consumption: All the allocation, forecasting, and optimization, which kind of sounds a lot like FinOps for AI," spanning model routing, quantization choices, agent limits, and cache strategies. * Value: "How do we monetize those tokens? How do we adjust our pricing based on the cost of those tokens? What are the labor implications in our entire company based on the cost of that AI?" That last piece is where token pricing directly collides with software-as-a-service (SaaS) business models. As Storment told me in an interview, "Tokenomics is getting over to the price of the tokens and how effectively we manage this production and consumption of them is changing pricing models for Fortune 100 companies." He points to Microsoft's GitHub moves, shifting Copilot toward more explicit usage‑based charging, as an early example. Developers "who love the unlimited tokens" are now "really just angry at Microsoft," because their implicit subsidy vanished. Also: Why Anthropic suddenly pulled Fable 5 and Mythos 5 for everyone The labs themselves are also tightening the screws in ways that are invisible at the token level. He raised as a fresh example Anthropic's Fable model card: "If you're going to use Claude at Fable to try to build an LLM, they will silently drop you to a different model, and you aren't going to know." Since then, Anthropic has walked back this policy, but other companies may not. Such silent policies make a mockery of any naive "cost per token" metric, because "not all tokens are created equal by any stretch of the imagination." Storment agrees. "A token can cost two cents per million, or it can cost 35 per million, just from a cost perspective," he said, and even at the same rate, "one might drive a lot of value, and one doesn't, based on how you're using it." For him, the point of embracing "tokenomics" as a term is to harness the fact that the C‑suite has already latched onto tokens as a mental model. It also doesn't help that today's advanced LLMs, such as Anthropic Fable 5, can chase after an answer and burn tokens without users having a clue what's actually happening. For instance, Simon Willison, co-creator of the Django Web framework, reported that "Based on a screenshot and a one-line prompt, Claude Fable 5 + Claude Code," launched a web server, used numerous and different web browsers, built and launched its own web server, and performed many other tricks, all to track down a simple CSS display bug. Had he used token pricing, it would have cost him only $12. It's easy to envision a frontier model taking on a more complex problem and burning hundreds or thousands of dollars. Business models: from credits and seats to blended token bundles These pricing experiments show a pricey future. Most customers will never see a line item labeled "120 quadrillion tokens." Instead, vendors are building layers of abstraction on top: * Credits and opaque consumption: Storment described signing up for an unnamed service where "every time I ran a video, it was like, 'Put more quarters in the machine, put your credit card down. These credits go fast.'" Under the hood, those quarters are tokens. * Hybrid subscription + usage: Others use "a basic monthly, and then some level of consumption," giving customers a predictable base and then exposing them to token‑denominated overages at the margin. * Direct pass‑through models: A smaller set, especially in infrastructure‑adjacent products, are "starting to direct allocation, direct pass through," essentially showing customers the token meter more honestly but wrapped in their own dashboards and guardrails. These are all vulnerable to upstream shocks. Storment warned, "Anything changes in this, your token factory changes, you route to the wrong model and blow your cache up, you inefficiently forecast or estimate. Anything changes, this affects consumer pricing at the end, and you may have to change your prior pricing model for how you go to market, and this isn't just software companies, it's cascading into banks and everyone else today." Also: How this travel company's AI rollout drove a 73% satisfaction boost: A 5-step playbook for your business That cascading effect is why the Linux Foundation is spinning up a Tokenomics Foundation alongside the FinOps Foundation: to give big consumers and suppliers a vendor‑neutral place to hash out specifications and best practices for measuring and allocating token‑based costs. The FinOps Focus specification, originally designed to normalize cloud billing data, is already being extended for token‑level telemetry. A new "FinOps certified Focus generator" program aims to validate that providers' billing pipelines conform. The human side: AI haves versus have‑nots Beyond the spreadsheets, token pricing is already shaping who gets to use powerful AI -- and who doesn't. Storment sees a "societal divide between those who can afford the AI and those who can't" if high token costs persist. At the enterprise level, you can already see the outlines: "Certain teams are being deemed worthy of getting the latest model, and others are not," with some users routed automatically "to cheaper model[s]" and others granted exceptions. Yet there is also a strong argument against crude caps. One Fortune 100 executive told Storment to "look across your usage... and you're going to find some outliers of people... Don't cap them, don't shut them down. Go talk to them, find out what they're doing, because they might actually be doing something really interesting." In a world where YC‑backed startups receive "millions of dollars of tokens" from frontier labs to disrupt incumbents, shutting down internal experimentation could be an existential threat. Also: 5 ways to grow your business with AI - without leaving employees behind For individuals, and especially new workers, trying to use AI, token pricing feeds into broader anxieties about AI and jobs. You raised the backlash to AI‑heavy commencement speeches and the sense among graduates that AI is "coming directly for their jobs in an already tough job market." Storment's view is more nuanced but still stark: "I don't think AI is immediately coming for everybody's job, but I think the person who's better at AI is coming for the job of the person who's not using AI." If token prices and quotas restrict who can learn and experiment, that divide will only deepen. For both companies and individuals, we're moving quickly into an AI-token-based economy. This, in turn, will lead to a far more expensive AI world. What all that will mean is a question we don't yet have an answer to. The one thing we know for certain is that it will be orders of magnitude more expensive than it has been.

AI coding agents could soon cost more than the developers using them

Consumption-based pricing and scant cost controls are sending monthly bills into five figures, Gartner warns Gartner has slammed AI vendors' lack of transparency, saying developers are facing sharply increased costs from coding agents. Since the main AI coding agent vendors have shifted from seat-based licensing to consumption-based pricing, developer teams now face highly variable cost structures. Developer teams face an emerging problem of escalating costs, said Nitish Tyagi, senior principal analyst at Gartner. AI coding bills were leaping from $20 or $100 to $2,000 to $5,000 per developer per month, while in extreme cases, the bill might hit $20,000 in token charges. However, software engineering departments get little insight into how token consumption is calculated and billed, making it difficult for them to forecast and control costs accurately. AI coding vendors have not yet delivered built-in features to allow developers to optimize costs in AI coding agents, resulting in cost escalation, Gartner said. "None of the vendors have incredible features when it comes to cost optimization," said Tyagi. Instead, he said vendors were focusing on the concept of "tokenmaxxing" to "boost the high" of token consumption, suggesting that if developers increase the number of tokens, they will increase productivity gains. "There is no direct relation between the increase in token consumption and an increase in productivity gains," he said. Gartner recommends developer teams optimize token consumption and adopt strategies such as context engineering practices, where software engineers improve the input context provided to AI systems. Another recommended strategy is model routing, where engineering and platform teams direct simpler, high-frequency tasks to smaller models, using frontier models only for complex, high-value work "All of these things will improve the output quality, and, therefore, will increase the productivity gains as well, so while there is no direct relation between tokenmaxxing and productivity gains, there is a relation between optimization of token consumption with the output quality," Tyagi said. As a result of the lack of cost optimization tools among vendors and consumption-reduction strategies among users, AI development is in a situation where a developer's coding agents may cost more than they earn, at least in some parts of the world. Gartner predicts that by 2028, AI coding costs will overtake the average developer's salary due to rising LLM token consumption and the shift to consumption-based licensing models. "We're not saying AI token cost will be higher than every developer's salary on the planet, because US salaries tend to be higher than in India, for example. But current token costs are already more than most of the salaries in India," he said. However, the cost of coding agents does not vary according to where they are consumed, meaning AI coding costs for a developer in India may be equivalent to the salary of an engineer with four to six years' experience, Tyagi said. "We're not saying AI token costs will be higher than every developer's salary on the planet, because US salaries tend to be higher than in India, for example," Tyagi said, adding that token costs do not vary by location, and in India they may already be equivalent to the salary of an engineer with four to six years' experience. ®

Tech Workers Maxed Out Their A.I. Use. Now They're Trying to Minimize It.

Earlier this year, the message from tech companies to employees was clear: Use as much artificial intelligence in your work as possible. Employees called it "tokenmaxxing," with a token referring to a unit of A.I. use roughly equal to a word fragment. Companies like Meta and Amazon even encouraged workers to compete on leaderboards that tracked token use. Then came the bills from companies, like Anthropic and OpenAI, that provide A.I. tools -- and they were not cheap. Now the tokenmaxxing era appears to be over. Meta told employees last week that it would soon limit A.I. use after seeing an "exponential increase" in costs. In May, Uber said it had blown through its projected A.I. spending for the year in just four months, and it has placed some monthly limits on A.I. coding tools. Walmart also set limits for different A.I. tools. And Amazon and Meta have taken down their tokenmaxxing leaderboards. In other words, "tokenminning," short for "token minimizing," is now in. The reversal, within just a few months, underlines how A.I. use remains in flux as people try to figure out how to best use the tools. "The biggest problem is this is all changing so fast, people and companies don't know what to do," said Rob May, the chief executive of Neurometric, a start-up that helps companies better use A.I., and the author of "The Tokenminning Manifesto." "C.E.O.s who did not know how to measure the A.I. savviness of their employees thought, 'Well, who's using the most tokens?'" he said, adding that the philosophy ended up promoting volume over efficiency. OpenAI and Anthropic offer subscriptions that cost $10 to $200 a month for use of their A.I. models; when subscribers hit their usage limit, they are cut off. But the bulk of the revenue comes from offering tools to companies like Meta, Shopify and Amazon, which pay not only subscription fees but also for the tokens used by their tens of thousands of workers. So the more tokens that are used, the more money the A.I. costs. A simple task, like asking A.I. to summarize the transcript from a company meeting, may use a few hundred tokens. More complex requests, like writing code to build a new product or feature, can use tens of thousands. The costs of using A.I. models have soared as they have become more powerful and consume more tokens. Anthropic's newest A.I. model, Fable, is twice as expensive as its previous model, Opus. While there are cheaper models, many employees have fallen into the habit of using the most powerful models for everything, Mr. May said. The ways that people use A.I. have also changed. Instead of just conversing with A.I. chatbots, engineers deploy A.I. "agents," which can work on complex tasks for hours at a time. As a result, engineers can use tens of thousands of dollars' worth of tokens each month. Many companies said they were trying to be more strategic about A.I. spending after not seeing clear returns on their investment. "If you're not actually able to draw a direct line to how many useful features and functionality you're shipping, that trade becomes harder to justify," Andrew Macdonald, Uber's chief operating officer, said in a recent podcast interview. "That link is not there yet." That's not to say companies won't keep spending big on A.I. Meta told employees that it was on track to spend billions on A.I. use this year, but wanted to "find places we can spend less while getting similar or better business results." Marc Benioff, the chief executive of Salesforce, the enterprise software company, said his company planned to spend hundreds of millions on A.I. this year but now tracked "agentic work units" instead of tokens. The new metric is supposed to measure output, not just use. Meta's and Walmart's limits on employee A.I. use were reported earlier by The Information and Bloomberg. It's unclear how "tokenminning" might affect the bottom lines of Anthropic and OpenAI. At the height of the tokenmaxxing era this year, the A.I. companies reported record revenues driven by the use of coding tools. Last week, Meta told its engineers to use its internal coding assistant, MetaCode, instead of third-party tools if possible. Meta declined to comment, Anthropic did not provide a comment, and OpenAI did not respond to a request for comment. (The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. They have denied the suit's claims.) The clear path forward for companies, Mr. May said, is to use cutting-edge A.I. only on complex tasks that require it and substitute cheaper models for other instances. Companies can save as much as 90 percent by opting for less advanced A.I. models, said Andy Markus, AT&T's chief A.I. officer. He said his engineers were using the most powerful A.I. models for some tasks and the less powerful ones for most other actions. "There's an ebb and flow," he said. "What we do find is that, for most use cases, the latest greatest frontier model isn't needed."

AI coding costs could top developer salaries by 2028

By 2028, the AI tools a developer uses could cost more than the developer's salary, Gartner warns. AI coding costs are climbing fast, and most companies cannot even see what they are spending. By 2028, the AI tools a developer uses could cost more than the developer's salary, Gartner warns. AI coding costs are climbing fast, and most companies cannot even see what they are spending. The AI coding boom has a bill, and it is growing fast. By 2028, AI coding costs will overtake the average developer's salary, Gartner predicted on 24 June. The cause is simple: every move an AI agent makes burns tokens, and the meter is always running. Tokens are the units of data an AI model processes. Under the new pricing models, more tokens mean a bigger bill. "Organizations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact," said Nitish Tyagi, a senior principal analyst at Gartner. The warning lands at a strange moment. AI coding tools are wildly popular. Engineers love them, and managers credit them with real speed gains. Yet the same tools now threaten to cost more than the people they help. Gartner's point is blunt: popularity and price are rising together. From $20 to $5,000 a month The jump is already visible. AI coding bills are leaping from $20 or $100 a month per developer to $2,000 or even $5,000, Tyagi told The Register. The driver is a quiet change in how the tools are sold. Vendors used to charge a flat seat fee. Now most charge by consumption. A developer pays for what an agent uses, and an agent can use a lot. That shift turns a predictable line item into a wild one. It is the same dynamic that has already seen enterprise AI bills triple even as token prices fell. Consumption pricing rewards the vendor when usage grows. The more an agent writes, tests and retries, the more it bills. A single autonomous task can chew through tokens a developer never sees. Multiply that across a whole team, and the monthly invoice swells. None of that means the tools fail. Used well, they ship features faster and free engineers from grunt work. The worry is the gap between the promise and the invoice. Right now, far too few teams measure it, and fewer still act on what they find. Nobody can see the bill The deeper problem is visibility. Many vendors do not show how they calculate or bill token use, Gartner said. Companies cannot forecast the cost, so budgets run dry early. "Most organizations still lack the maturity and frameworks to effectively measure cost versus business impact," Tyagi said. Gartner is blunt about the result. Engineering leaders find token-driven spend harder and harder to justify. Budgets vanish sooner than planned. And without a way to tie spend to business value, the next budget meeting gets awkward. Developers are not the ones policing it. They optimise for speed, not cost. "Token discipline will not emerge through developer choice alone," Tyagi said. Without rules, he warned, costs can climb faster than the productivity the tools promise. Why the costs keep climbing Several forces push the meter up. Agents left to run on their own burn tokens freely. Bloated context windows, where the tool receives more text than it needs, add more. And teams rarely build a feedback loop to trim the waste. The tools themselves offer little help. AI coding vendors have not built in mature cost controls, Gartner said. So the job of restraint falls on the buyer. Most buyers are not ready for it. The user base is changing too. As people grow comfortable with the tools, light users become heavy users. Gartner expects model prices to rise as well, as AI firms chase profit. More usage, at a higher price, points one way. This is already reshaping behaviour. Some firms have started to cap how much AI their staff can use. The most AI-hungry companies now spend around $7,500 per employee each month. The industry scrambles for a fix The pain has created a market. Database vendors are now pitching themselves as the cure for runaway AI costs, arguing they can cut the number of calls a coding agent makes. Others want an industry standards body to explain the bills. Even big players are pulling back. Microsoft quietly retreated from heavy Claude Code use over cost, and GitHub paused some Copilot sign-ups as agentic demand strained the economics. The tools work. Paying for them at scale is the hard part. Gartner sees the wider market entering a fresh phase of expansion and competition. That should, in time, bring better cost tools and clearer pricing. For now, the buyers sit ahead of the products. They are scaling fast on tools that were never built to be cheap. What Gartner says to do Gartner's advice is about discipline, not retreat. It tells engineering leaders to sort tasks into three buckets: developer-led, developer-with-agent, and fully agent-led. Each one gets a set level of autonomy. Take model routing first. A frontier model is overkill for a simple function. Gartner wants teams to send easy, high-frequency tasks to smaller, cheaper models, and reserve the expensive ones for complex work. Done well, that alone can trim the bill sharply. Context engineering is the other lever. Every extra line fed to the model costs tokens. Trim the input to what matters, summarise the rest, and the meter slows. Then set token limits, monitor usage automatically, and review the heaviest workflows in every sprint, rather than panicking once the budget runs out. The bottom line None of this kills the case for AI coding. The tools genuinely speed work up, and few teams will hand them back. But Gartner's forecast is a warning that the savings are not automatic. A tool that writes code faster is no bargain if it costs more than the person using it. The open question is whether discipline arrives before the bill does. Pricing keeps rising, usage keeps growing, and the visibility stays poor. 2028 is not far away. The companies that win will be the ones that learn to count tokens before the tokens count them.

Tokenminimizing: firms cap staff AI use as bills bite

A year ago firms ranked staff on leaderboards by how much AI they burned. Now AT&T, Meta, Uber and Walmart are capping it, and Amazon has scrapped the leaderboard entirely. A year ago, the smart move inside a big company was to use as much AI as humanly possible. Some firms even ranked employees on leaderboards by how many tokens they burned, a status game that earned its own name: tokenmaxxing. That era is ending. The same companies are now capping AI use, and the new buzzword is its mirror image: tokenminimizing. The latest is AT&T, which has started limiting some employees' access to GitHub Copilot, according to The Information. Meta is reportedly reining in staff spending on Anthropic and other AI tools, a sharp reversal from the months when workers raced each other to consume the most. The bill came due The trigger is simple: the spending got frightening. The most AI-obsessed firms now spend $7,500 per employee per month, and agentic tools that call a model over and over have tripled enterprise AI bills even as per-token prices collapsed. Uber blew through its entire 2026 AI coding budget by April and now caps employees at $1,500 a month per tool. Walmart has capped use of its in-house AI agent. Amazon scrapped the internal leaderboard that ranked staff by AI usage after people gamed it, sending compute costs up. Even individual engineers were a problem: Microsoft found some spending $500 to $2,000 a month on Claude Code tokens alone. Cue the 'I told you so' Some companies are enjoying the moment. "We never celebrated tokenmaxing," Box chief executive Aaron Levie said. "We never had leaderboards, so we didn't get ahead of our skis on... incentivizing the wrong thing." Not everyone is pulling back. At Databricks, an engineering leader said the AI budget for engineers is still unlimited, "so tokenmaxxing still exists", a sign that firms confident their staff use AI efficiently see less reason to ration it. That is the tension under the trend. Caps control costs, but they can also throttle the productivity gains that justified the spending in the first place. The real winners are the cost-cutters' tools The more lasting shift is what tokenminimizing pushes companies toward. To cut bills without cutting use, firms are swapping expensive frontier models for cheaper or open-source ones on simpler tasks. That hands an opening to the plumbing. Microsoft and Databricks have launched 'gateway' tools to monitor and cap staff AI spending, and Nvidia-backed Factory, valued at $1.5bn, just launched a model router that shunts cheaper tasks to cheaper models. Satya Nadella captured the mood in a weekend essay, arguing AI models should be swappable rather than dominant. "The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see," he wrote. Coming from the boss of a company whose software is under pressure from the very labs it depends on, it is also a tell about where this is heading.

Tokens are getting cheaper, but companies are spending even more on AI as a result, top economist warns | Fortune

The ghost of a 19th century English economist may be haunting yet another part of the AI boom. In 1865, William Stanley Jevons observed that when the Watt steam engine made coal use more efficient -- decreasing the amount required to a task -- coal consumption actually skyrocketed. More than 150 years later, one economist is citing this phenomenon, dubbed Jevons paradox, to explain why the cost of AI will continue to creep up. Despite the price of a single token dropping more than 90% since 2023, spending on large language models has doubled since late last year, according to the Silicon Data Token Expenditure Index. Essentially, token price -- or the cost to process the most basic unit of AI -- has gone down, but companies are spending more than ever on AI. Apollo chief economist Torsten Slok said it's yet another example of "Jevons paradox in action." "As tokens get cheaper, companies don't spend less but instead run more AI agents, automate more workflows and generate more code, pushing aggregate expenditure higher even as the unit cost of intelligence collapses," Slok wrote in a recent blog post. The cost of tokens has become a major concern for companies racing to leverage AI. The trend of "tokenmaxxing," in which employees blitz to increase their AI use, has emerged as companies like Meta and Amazon incentivize the technology's use. However, the deployment of AI just for the sake of it is proving unsustainable. Uber president and chief operating officer Andrew Macdonald recently said the rideshare company burned through its entire AI budget in the first four months of the year amid the company's increasing use of Claude Clode. Bloomberg reported the company has now capped monthly AI spending to $1,500 per employee. Others are reckoning with AI -- which tech leaders promised would boost productivity -- still costing more than human labor: "For my team, the cost of compute is far beyond the costs of the employees," Bryan Catanzaro, vice president of applied deep learning at Nvidia, recently said in an interview with Axios. Jevons paradox and the AI boom With token prices dropping as new AI models become more efficient, the era of tokenmaxxing may be over, but that won't necessarily solve companies' AI budget crises. A group of Bain and Co. analysts confirmed Slok's point in a brief last week, finding that while token costs were halved from December 2024 to December 2025, the tokens consumed grew by 450% in the same period. The analysts attribute this paradox to companies feeling compelled to upgrade their AI models to take advantage of the upgraded technology, rather than stick with their current models and pocket the savings. Moreover, tokens per query have increased as agents become more capable of complex tasks. And to Slok's point, once a team believes AI can complete these more significant tasks, they will ask more of the technology and subsequently use more tokens. "The models get cheaper. The usage gets heavier. The bill stays stubbornly high," the brief said. Token costs are just one area of the AI boom where economists have seen paradoxical economic data emerge. Slok similarly found that despite AI being able to automate 86% of tasks for customer service workers, employment for call center workers in the Philippines has actually nearly doubled over the last decade. A similar trend can be seen in radiologists, another profession deemed endangered because of AI's ability to automate it. The number of radiologists in the U.S. has actually increased by 10% in the last 10 years. "Lower cost per interaction does not mean fewer interactions," Slok wrote last month. "It means more customers served, more channels opened and more markets worth reaching. The technology that was supposed to shrink the industry is fueling its expansion." Bain and Co. sees an AI future where a company's operating expenses come 70% from human headcount, and 30% from tokens. In order to make this shift sustainable, analysts warned companies will have to navigate uncertainty regarding costs of AI by not just creating a budget for AI spend, but determining the true financial returns from employing certain AI tools to assess if the tokens are worth it. "The opex shift from headcount to tokens isn't a budget problem," analysts said. "It's a structural transformation."

Why Companies Are Already Blowing Through Their 2026 AI Budgets in Just Two Months

"The future is going to be good for the AIs regardless; it would be nice if it would be good for humans as well." That's a quote from Ilya Sutskever, an AI researcher who cofounded OpenAI and became its chief scientist. It reminds us that though AI is powerful, it's very controversial -- for a long list of reasons. One of the biggest concerns is that as companies race to deploy AI, they may be displacing real people from real jobs as AI takes on some workplace duties. A new report at CNBC points out that displacing people may actually be happening for a different reason: the increasing cost of using AI systems means company leadership is having to question exactly where it spends the money needed to keep the business running. The news outlet spoke to two "enterprise AI CEOs" who are busy rolling out AI to their workers, and they both highlighted that AI costs are steeply rising, and that AI prices are soaring to levels above what anyone expected. At the center of the issue is the price of AI tokens -- the basic unit of input processed by an artificial intelligence system like a large language model, the core tech at the heart of systems like ChatGPT.

As AI adoption grows, token consumption comes under close scrutiny

Businesses are facing hefty AI bills, prompting a shift from simply counting AI tokens to scrutinising their actual value. Companies are now tracking cost per outcome and implementing stricter controls like usage dashboards and approval mechanisms. Firms are exploring cheaper models and consolidating tools to manage expenses and ensure measurable business results. The "tokenmaxxing" trend has drawn scrutiny from auditors and risk officers, who are now conducting deeper reviews of how artificial intelligence tokens are consumed, not just how many are used. As enterprises grapple with mounting AI bill shocks, they are increasingly tracking metrics such as cost per workflow, cost per transaction, cost per document reviewed, cost per customer query resolved, cost per software change delivered and productivity gains per process, said Ashvin Vellody, chief strategy and innovation officer, Consulting, at Deloitte South Asia. "The stronger organisations are also putting in place usage dashboards, model-selection guidelines, consumption thresholds, approval mechanisms for high-cost models, and governance forums to assess whether AI usage is delivering measurable value," he added. Companies are drowning in token bills due to heavy spending on AI tools without clear returns, success metrics, or guardrails. Goldman Sachs has forecast global token usage could surge 24 times to 120 quadrillion tokens per month between 2026 and 2030. Cheaper tokens may not necessarily translate into lower AI bills, Gartner says. A recent report by brokerage firm Jefferies noted that "AI for now is costing more money than it is saving". At the same time, large IT services firms are under growing pressure to demonstrate measurable business outcomes rather than simply reporting productivity improvements. To save token costs, HCLTech said it is shifting workloads to sovereign models. "We have AI offerings that leverage sovereign models and GPUs, as well as non-GPU inferencing, to reduce dependency on high-cost external compute," said chief technology officer Vijay Guntur. Tech Mahindra, meanwhile, is deploying smaller open-source models that can deliver comparable business outcomes for clients at a lower operating cost. Some clients are consolidating tools and platforms to improve governance and cost control, said Nikhil Malhotra, the company's chief innovation officer. It has also invested in building its own AI models suited to enterprise needs. "Auditability is emerging as a critical component of enterprise AI adoption," said Chandrashekar Mantha, partner-Assurance at Deloitte India. "Large AI programmes are typically reviewed weekly or fortnightly, while smaller projects may be tracked daily to prevent budget overruns."

After cloud spending, IT companies see 'token maxxing' as AI's next cost challenge

IT firms are flagging a new AI challenge: 'token maxxing,' where companies prioritize AI usage metrics over actual business results. This trend, akin to past cloud spending issues, sees token consumption treated as value. Experts emphasise tracking AI's impact on workflows and outcomes, not just its input. As enterprises rush to deploy generative AI across their organisations, IT services firms are beginning to warn of a new challenge: token maxxing. The term refers to the growing tendency among companies and employees to measure AI adoption through token consumption -- the units used whenever large language models process information -- rather than through actual business outcomes, according to a report by the Times of India. Tokens have emerged as the basic currency of the AI economy. But after an initial wave of experimentation, technology companies are increasingly cautioning that unchecked token usage could create a fresh cost problem, much like cloud spending did for enterprises over the past decade. "Tokens are an input to delivery, not a measure of value," Arumugam Kumaradassan, vice-president and head of AI industrialisation and enterprise IT automation at Cognizant, told TOI. "When token consumption is treated as the primary metric, costs scale linearly with demand without a corresponding return in business outcomes," he said. The phrase "token maxxing" has gained popularity in Silicon Valley, where the "maxxing" suffix -- borrowed from internet and gaming culture -- refers to aggressively optimising a particular metric. In the AI world, it increasingly describes a mindset where higher AI usage is automatically equated with higher productivity. According to the report, IT firms are now trying to ensure that AI spending remains tied to measurable outcomes rather than raw consumption. Cognizant has introduced systems that track token usage alongside business workflows and results, allowing customers to better understand whether AI spending is generating value. The issue is becoming more relevant as enterprises move from AI pilots to production deployments involving thousands of employees and autonomous AI agents. At Happiest Minds, co-chairman Joseph Anantharaju told TOI that the company is building token metering and optimisation capabilities as customers scale agentic AI projects. "I think that's going to be very important -- the ability to meter it," he said. The company is also exploring outcome-based commercial models that combine software, AI agents, platforms and token consumption into a single framework tied to business goals. The focus on efficiency is not limited to IT services firms. Salesforce chief digital evangelist Vala Afshar said organisations are increasingly tracking not only how many tokens are being consumed, but also how many tasks are actually being automated as a result. "It's wasteful to just spend tokens unless you're creating value at the speed of need," Afshar said, according to the report. Technology providers are also becoming cautious about linking pricing directly to token consumption. Mphasis chief executive Nitin Rakesh told TOI that customers are increasingly bearing variable AI costs as usage scales, making outcome-based pricing models more attractive. "What we are pricing is the economic outcome," Rakesh said. "There is a base price that you will pay me, and the rest will be linked to the outcome I can drive." The debate highlights a broader shift underway in enterprise AI. As companies move beyond experimentation, the conversation is increasingly changing from how much AI is being used to whether that usage is generating measurable returns.

Soaring AI Costs Push Enterprise Buyers to Cheaper Chinese Models | PYMNTS.com

Companies are looking to better manage their use of AI after seeing the costs of the technology rise, according to the report. The rising costs have been driven by the shift from chatbots to agents, which consume more computing power, as well as the AI labs' move from flat subscriptions to token-based billing, per the report. This has created an opening for Chinese AI labs that are able to charge less than the U.S. companies due to their more efficient models and China's lower energy costs, the report said. Chinese AI models now have greater token consumption than U.S. ones, which marks a change since the beginning of the year, the report said, citing data from OpenRouter. Companies that encouraged their employees to use AI tools when the costs were lower are now using a variety of methods to cut back, according to the report. Executives quoted in the article said their companies have introduced usage caps, tasked employees to use the right tool for each task, shared cost-saving ideas such as switching to models that are older and cheaper, and adopted open-source models. PYMNTS reported in February that as enterprises accelerated their AI usage from pilot programs to production-scale deployments, they found that the commercial infrastructure underpinning traditional software-as-a-service (SaaS) doesn't translate cleanly to AI. Rather than charging per employee, AI charges per token, per application programming interface (API) call, per generated image, per inference cycle, per autonomous workflow executed in the background while no human is watching, or, in some cases, all of them simultaneously. It was reported in May that Uber exhausted its full-year 2026 AI budget by April, leading executives to say that the company was "back to the drawing board" and that the productivity case had not closed. On June 1, it was reported that Walmart limited employees' AI usage amid rising demand. The company began offering a set number of tokens for each employee to use with its in-house AI agent Code Puppy. Before this, workers had an unlimited number of tokens. For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.

Gartner: AI Bots Will Soon Cost More to Run Than Hiring a Human Developer

Rising Token-Driven AI Spend Is Straining Budgets and Challenging Cost Justification By 2028, AI coding costs will overtake the average developer's salary due to rising large language model (LLM) token consumption and the shift to consumption-based licensing models, according to Gartner, Inc., a business and technology insights company. AI tokens are the units of data processed by generative AI models. Token consumption directly impacts the cost of AI coding tools, particularly under consumption-based pricing structures. "Organizations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact of rising token consumption," said Nitish Tyagi, Sr. Principal Analyst at Gartner. "Token discipline will not emerge through developer choice alone, as developers tend to optimize for speed and convenience over cost efficiency. Without a governed engineering operating model, costs can escalate faster than the productivity gains these tools are designed to deliver." Consumption-Based Pricing Introduces Cost Predictability Challenges The shift from seat-based licensing to consumption-based pricing among AI coding agent vendors is introducing highly variable cost structures for software engineering workloads. Many vendors lack transparency into how token consumption is calculated and billed, limiting enterprises' ability to accurately forecast and control costs. Without clear visibility into token usage across development tasks, organizations risk budget overruns and reduced ability to track cost-to-value outcomes. "Most organizations still lack the maturity and frameworks to effectively measure cost versus business impact," said Tyagi. "Software engineering leaders are increasingly concerned as token-driven AI spend becomes harder to justify, with budgets often being depleted earlier than expected." Usage Patterns and Governance Gaps Are Driving Cost Pressure Beyond pricing and visibility challenges, how AI coding agents are used within organizations is further driving cost pressures. Token overspending is often linked to how software engineering leaders govern usage, with common failure modes including ungoverned autonomy in agent-driven workflows, bloated context windows and the absence of structured feedback mechanisms to optimize usage. In addition, AI coding vendors are yet to deliver mature, built-in cost optimization capabilities in AI coding agents, further contributing to cost escalation. "AI coding costs will continue to rise as infrastructure investment and profitability challenges push model pricing higher," said Tyagi. "At the same time, as more developers adopt AI tools, light users are expected to rapidly become mainstream users as familiarity and reliance increase, driving further growth in token consumption and overall spend." To manage rising costs and avoid budget overruns, Gartner recommends that software engineering leaders implement a disciplined operating model for AI usage: * Establish a use-case-driven decision framework: Organizations should clearly define when AI coding agents should be used and determine appropriate levels of autonomy for each task. This includes classifying development tasks into three execution models: developer‑led, developer‑with‑agent, and fully agent‑led. * Align model selection with task complexity: AI coding agents are most cost-effective when work is broken into smaller tasks that can be handled by smaller models, with escalation only when complexity demands it. Engineering and platform teams should implement intelligent model routing strategies that direct simpler, high-frequency tasks to smaller models while reserving frontier models for complex and high-value development work. * Mandate context engineering practices: Developers must be trained to optimize the input context provided to AI systems by including only relevant information, summarizing content where possible, and eliminating unnecessary data to reduce token consumption without compromising output quality. * Implement governance and cost controls: Organizations should introduce mechanisms such as token thresholds, escalation policies, and automated monitoring to manage usage. Embedding these controls into engineering workflows ensures consistency and prevents uncontrolled cost growth.

Accenture Had Linked Promotions to AI Use; Now CEOs Are Reversing This Position

in barely four months, enterprise CEOs have shifted from being an AI evangelist to cutting corners on AI spends Back in February, Accenture CEO Julie Sweet made news asking staff to get friendly with AI if they wanted future promotions. Now, four months later, big businesses are reconciling AI costs with their AI evangelism mandate and quietly asking their employees to go slow with AI coding tools as their costs have suddenly spiralled. In an article published by The Economist, the discussion has veered around to the growing problem faced by enterprise around the use of AI via agentic solutions. These use massive processing power and have caused AI bills that have made their use uneconomical. The article quotes an unnamed executive to suggest that the problem is only going to grow. "Big companies, the executive points out, typically use hundreds of software programs. If each of those offer agents (as they probably will), AI costs could easily spiral out of control," the official told the magazine. Does this mean that the era of AI maximalism is over or getting there inexorably? Maybe so because now CEOs are forcing staffers to reduce AI usage in certain tasks. They are moving away from evangelising AI to cutting corners as several corporate honchos forgot to check their AI billing tabs, like we heard from the Uber CTO recently. The irony is not lost when we consider what Accenture's Sweet had said while appearing in a podcast back in February. She began by explaining that Accenture had integrated AI into its workflow over the past few years and the result was that employees had to now use these tools as part of their regular work routines. Learning to work with AI is increasingly tied to career progression within the company, she had noted then claiming that the company had spent billions of dollars training their staff in AI use. Wonder if she still thinks her idea of linking promotions to AI use is workable. We won't know for sure till Accenture shares their compute bill from the expenditure column. Of course, the current reticence to spend on AI might be a temporary phase as it plays out in parallel to another trend whereby companies are claiming that improved AI efficiencies has resulted in them handing out pink slips to redundant workers and replace them with coding agents. In fact, Amazon even set up a leaderboard ranking staffers by the number of AI tokens they used. Of course, the board disappeared as suddenly as it had appeared. Somewhere during the course of these four to five months, the meme-worthy phenomenon called token-maxing also surfaced. Stories of token utilisation without a clearly defined outcome or one could say an overkill of AI usage started emerging. We heard of an employee who spent $150,000 a month on AI tokens. Someone at Nvidia claimed he spent more on AI costs for research than on his entire staff salaries. Another company blew away $500 million on Claude usage fee within a month. In fact, there was one research conducted by Ramp Economics Lab that created a few ripples among those occupying the corner office and wanting to be part of the AI evangelism club. It said the top 1% of companies spend $7500 per employee per month on AI expenses. Compared to the $16,000 a month an average software engineer makes, it is still decent. Of course, those are the actual power users. The top 10% of companies are spending about $611 per month per employee on an average while the median are only dishing out a measly $11.38, which is equivalent to the cost of a seat on an enterprise plan. However, the study also noted that AI spends grew 14.1% per employee per month in recent times. These numbers are concerning as we might think. And the immediate response is that most CEOs now appear to have ditched their exuberance for a modicum of caution. They all know that the problem isn't AI itself, but where and when one uses it and how enterprises can find ways to maximise both its potential and their profits by using the technology. In recent times, we've seen AI consultants go around asking companies to impose token limits, ensure a more selective approach to AI deployment and usage of cheaper models. As we said earlier, the first step was to discard the euphoria as Amazon showed by removing the AI leaderboards. We aren't sure Accenture has followed suit on reducing the euphoria just yet. The fact remains that customers are reducing AI usage in consonance with the phenomenon of services companies like OpenAI and Anthropic pushing their unbridled capex on to costs for the users. Some experts warn that token costs could be the cheapest that they will ever be, while others point to Microsoft who are looking for cheaper alternatives, even if they come from the forbidden land of China! Looks like the AI ecosystem is facing its first big challenge. While getting customers to use AI indiscriminately for all sorts of tasks, they forgot to ascribe a cost to such usage. On their part, customers went all guns blazing into the "valley of death" without reminding themselves that there are "no free lunches" in corporate America.

The Wake-Up Call: A $400K AI Bill Becomes $1.4M Overnight

A founder posted this week that his company's Anthropic bill is about to jump from $400K to $1.4M per year, not because usage exploded, but because they crossed 150 seats. Past that threshold, Claude Enterprise pricing kicks in: seats no longer include usage, and every token bills at standard API rates. At their current run rate, that's a 3.5x increase overnight. This is not an isolated story. It's a signal that the invisible has become visible. For the past two years, AI spend inside companies was treated like electricity; something you use freely until someone sends you a very large bill. What's particularly interesting here is the per-role dynamic that's emerging. For engineering, the spend is worth it as the best model saves more than it costs. But for many other roles? Apps nobody uses, skills someone already built. No ROI. Companies are now being forced to do what they should have done from the start: measure it. Reading the Data: The LLM Expenditure Index is Falling The Silicon Data LLM Expenditure Index, which measures the effective price per million tokens across the actively traded LLM market has recently rolled over after a period of strong growth. The instinct is to read this as bearish for AI. We have a different opinion. The index falls for two reasons: (1) model prices decline, or (2) the market shifts toward cheaper models. We would argue we are seeing the latter. This is not a demand story, it's a model mix story. Companies aren't using less AI. They're discovering that GPT-5 nano at $0.05/M input tokens does the job that GPT-5 Pro at $15/M was doing last quarter and they're making the switch. The Bifurcation Has Arrived: Frontier vs. Everyday AI We've been flagging this for months. The AI market is splitting in two, and the token economics story is the clearest evidence yet that it's happening faster than most investors expect. Frontier AI [the $15-$270/M token range] will continue to grow in absolute dollar terms, but it will increasingly concentrate among a narrower set of firms: those with the balance sheets to absorb the compute cost and the genuine hard problems that justify it. Think pharma companies running drug discovery, financial institutions doing complex reasoning at scale, and defense contractors. Everyday AI [the $0.05-$2.50 range] is where the volume explosion happens. As spend visibility improves and CFOs start asking questions, the gravitational pull toward cheaper, good-enough models becomes enormous. This is Jevon's Paradox in reverse: for every enterprise that wakes up to token costs and downgrades, two more startups find that the cheaper model makes their product viable for the first time. Value Chain Implications: Who Wins When AI Spend Gets Rational? Most of the focus remains on "which frontier model wins", but the right question is: who captures value when enterprises start treating AI tokens like any other IT resource to be optimized? The Bottom Line The $400K → $1.4M overnight shock is not a bug in AI adoption, it's the market maturing. For two years, enterprises consumed AI tokens with the same discipline they once applied to printing costs. That era is over. What comes next is not a slowdown in AI spend; it's a rationalization. Total token volume will continue to grow, driven by Jevon's Paradox and by the explosion of use cases that only become viable at $0.05/M. But the composition of that spend is shifting: away from uncritical frontier model usage, toward intelligent routing, cost-aware architecture, and multi-model strategies. The companies that win are not the ones selling the most expensive tokens. They're the ones who help enterprises figure out which tokens they actually need.