Why Are Big Tech Engineers Maxing Out AI Tokens, And Is It Burning Billions Of Dollars?

Last Updated:March 26, 2026, 09:00 IST

Strap: Every time an employee uses AI, whether to write code, summarise documents or generate content, they are consuming tokens. These tokens are billed, often in real time

In the world’s biggest tech companies, engineers are no longer competing only on code quality, product impact, or speed of delivery. Instead, a new metric is emerging as a marker of ambition and influence — how much artificial intelligence (AI) they use.

This trend, informally dubbed “tokenmaxxing,” is turning AI consumption into a kind of workplace status game. Internal dashboards and informal leader boards are beginning to track usage, while anecdotes suggest that some engineers are running multiple AI agents simultaneously, not just to solve problems, but to demonstrate activity.

Ironically, AI was meant to make work more efficient and reduce costs. Instead, it is creating a new kind of arms race — one where the goal is not just output, but scale of usage.

What Exactly Is Tokenmaxxing?

To understand the trend, it helps to understand how modern AI systems are priced. Most large language models operate on “tokens” — small chunks of text that include both input and output. Roughly speaking, 1,000 tokens correspond to about 700-750 words.

Every time an employee uses AI, whether to write code, summarise documents or generate content, they are consuming tokens. These tokens are billed, often in real time.

For example, at OpenAI, an engineer processed an extraordinary 210 billion tokens in just a week — roughly equivalent to generating text on the scale of Wikipedia dozens of times over.

At Anthropic, a single user of an AI coding tool ran up a monthly bill exceeding $150,000, highlighting how quickly usage costs can spiral.

Across companies such as Meta and Shopify, AI usage is also beginning to influence performance evaluations. Managers are increasingly rewarding employees who rely heavily on AI tools, while those who use them less risk falling behind.

Tokenmaxxing refers to the practice of maximising this usage. In some cases, engineers deliberately push large volumes of data through AI systems or run continuous automated processes to increase their token count.

The behaviour reflects a shift in workplace signalling. Earlier, productivity might have been measured through lines of code or project milestones. Today, in some environments, heavy AI usage itself is becoming a proxy for productivity.

“Token counts are not a true measurement of productivity. While a higher token count may indicate more AI usage, productivity is truly measured by outcomes such as speed, accuracy, and business impact. A skilled prompt engineer might use 500 tokens to get something right, while someone less experienced burns through 5,000 tokens with poor prompts and iteration. Higher usage could signal inefficiency, not productivity. Relying solely on token counts can create a distorted sense of priority that values volume over value. Instead of allowing a team’s token count to dictate how productive a team is, we must focus on how efficiently tasks are completed,” said Anurag Jain, Founder & CEO, Oriserve.

How Expensive Is Tokenmaxxing

This shift has serious financial implications. Unlike traditional software tools with fixed licensing costs, AI operates on a usage-based model. The more you use it, the more you pay. This makes costs inherently unpredictable.

There are already signs of how quickly expenses can escalate. Reports suggest that individual users have generated AI bills running into hundreds of thousands of dollars per month. In some cases, single AI-driven tasks have cost thousands, even tens of thousands, of dollars.

“Many companies are moving to hybrid budgeting models. These models include baseline provisioning and variable budget buffers based on historical token use, workload classification, and predicted models of future token use. The majority of companies will incorporate AI budgeting for teams or functionalities… While FinOps techniques implemented with real-time tracking and attribution will play a critical role in ensuring adequate monitoring of AI budget expenditures, the emphasis will change from merely controlling expenditures to ensuring that AI expenditures are associated with revenue-generating or efficiency-increasing activities,” said Kanishk Agarwal, Chief Technology Officer at Judge Group, India.

The core issue is not just the cost per token, which has been gradually declining. It is the explosion in usage. As engineers rely more heavily on AI, and as autonomous agents run continuously in the background, total consumption rises dramatically.

This creates a paradox: AI is becoming cheaper per unit, but far more expensive overall.

Can You Cap AI Usage Per Employee?

As AI becomes central to work, access to it is starting to resemble a workplace benefit. Some companies are already allocating “token budgets” to employees, defining how much AI compute they can use. In high-performance environments, larger budgets can signal trust, seniority or importance within the organisation.

There is even growing discussion in industry circles about treating AI compute as a form of compensation — alongside salary, bonuses and stock options. In such a model, access to powerful AI systems becomes a resource that employees compete for. This marks a fundamental shift. AI is no longer just a tool; it is becoming a currency.

“Companies do not typically impose hard caps on the number of tokens used by a business, which would limit productivity or innovation. Instead, they use a structured governance model consisting of access by role, a clearly defined use case, and an approval process for models with high per-token costs. By using soft caps, indicating a limit on usage along with warnings, and providing visibility into usage, there is enough accountability for controlling the usage of tokens while allowing for proper use. The goal is to support the responsible adoption of AI while ensuring that AI-generated value is aligned with the priorities of the business and that AI is not used excessively or in a manner inconsistent with the objectives of the organisation,” stressed Kumar Rajagopalan, Vice President, Strategic Initiatives & Country Head India, Dexian — an IT consulting firm.

Can It Redefine Productivity Or Distort It?

The rise of tokenmaxxing is also forcing companies to rethink how they measure productivity.

On the one hand, AI can genuinely enhance output. Engineers can write code faster, analyse data more efficiently and automate repetitive tasks. In theory, higher AI usage could reflect higher productivity.

But there is a risk that the metric itself becomes distorted. If employees are rewarded, explicitly or implicitly, for higher usage, they may prioritise consumption over efficiency. Running multiple AI processes or generating excessive outputs may inflate token counts without necessarily improving outcomes.

This creates what some observers describe as “productivity theatre” — activity that looks impressive on dashboards but does not always translate into real value.

In such an environment, the definition of a “good employee” could shift in unintended ways.

How AI Can Transform Productivity

Industry leaders increasingly describe AI as a utility — something akin to electricity or water. You use it as needed, and you pay for what you consume.

For companies, this means AI spending is no longer just a technical issue. It is moving into the domain of finance, with CFOs and senior executives closely tracking costs.

“AI can dramatically increase productivity per worker, provided that it is used appropriately. If AI is implemented without governance, it can increase costs without providing an increase in output… For a growing organisation that has a ton of unprocessed data, the best use of AI is to sift through that data and surface insights that can slingshot growth,” explained Jain.

How To Prevent Runaway Token Bills In Enterprise Settings?

On the one hand, AI could boost productivity and help Indian firms deliver more value to global clients. On the other hand, large-scale adoption could lead to a surge in token consumption and therefore costs.

A key question is whether these costs will be absorbed by companies or passed on to clients. Another is whether productivity gains will be sufficient to offset the increased spending.

“The market is rapidly shifting towards companies implementing shared/value-based pricing. Customers are more likely now than ever before to pay for AI computers based on how far the solution improves speed, accuracy, and efficiency. In some instances, firms may incur initial costs to showcase their value proposition to clients and remain competitive. Long term, as AI is embedded into delivery, it is expected that these costs will be included seamlessly in pricing, by way of aligning AI usage with larger outcome-driven engagement models,” explained Rajagopalan.

There is also a workforce dimension. As AI becomes more central, junior engineers may not be replaced outright, but their roles could shift towards managing and supervising AI systems rather than performing tasks manually.

“To avoid runaway costs, organisations need to have visibility into their usage of resources, control over spending, and optimization of operational processes. We have real-time monitoring tools, usage thresholds, and anomaly detection capabilities to identify unexpected spikes in costs… Periodic audits and governance frameworks establish disciplined use of AI; this layered approach will enable organisations to be cost-efficient as they expand their use of AI responsibly,” pointed out Jain.

A New Kind Of Arms Race

In many ways, tokenmaxxing is less about technology and more about human behaviour.

Whenever a new metric becomes visible, it tends to shape incentives. In this case, the visibility of AI usage is creating a new kind of competition — one that rewards scale, speed and constant activity.

The risk is that this competition spirals into an arms race, where more usage is equated with more value, regardless of actual outcomes.

“There is growing pressure in some environments to showcase AI usage as a sign of efficiency. This can lead to AI being used in a performative way rather than as a meaningful addition to workflows. Forward-thinking companies will need to redefine performance metrics around outcomes, rather than the tools being used. Training teams and clear communication from leadership will be critical to ensure AI is seen as an enabler of business goals, not a metric in itself. Ultimately, businesses should focus on genuine productivity gains from effective AI use, rather than simply inflating usage to create the perception of success,” points out Agarwal.