Developers are Burning Tokens to Look More Productive
By Siddharth Jindal ·
NVIDIA CEO Jensen Huang is urging the tech industry to rethink how it approaches AI usage, arguing that in the age of AI agents, engineers who are not actively consuming AI compute risk falling behind.
In a recent appearance on the All-In Podcast during NVIDIA’s GTC, Huang said, “If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed.”
He likened it to a chip designer choosing paper and pencil over CAD tools, suggesting that skipping AI today is equally outdated.
Huang is pointing to a new way of thinking. Rather than cutting AI costs, companies are focusing on using it to boost employee output.
During his keynote speech at GTC, Huang shared a new model for the workforce, in which engineers are allocated dedicated AI budgets. “I could totally imagine in the future every single engineer in our company will need an annual token budget,” he said, adding that these budgets could run into hundreds of thousands of dollars to amplify productivity.
In fact, he suggested that token access could become a recruiting lever. “One of the recruiting tools in Silicon Valley is how many tokens come along with my job,” he said, arguing that engineers with greater access to AI compute will ultimately be more productive.
This framing naturally favours NVIDIA’s business. As the dominant supplier of GPUs powering models from companies like OpenAI and Anthropic, NVIDIA benefits directly from higher token consumption. More tokens translate into greater demand for compute.
Enter Tokenmaxxing
According to a recent New York Times report, a new trend is emerging where developers are increasing their use of AI to signal productivity. This has given rise to a status game dubbed ‘tokenmaxxing’, where productivity is increasingly associated with how much AI is used.
At some tech companies, including Meta and OpenAI, employees track AI usage through internal leaderboards that rank them by token consumption.
Token budgets are also emerging as a new workplace perk, with some developers spending thousands of dollars each month to automate their work.
NVIDIA’s 2026 State of AI report reflects strong momentum in AI spending. Around 86% of organisations plan to increase their budgets this year, while 12% expect no change. Nearly 40% anticipate hikes of 10% or more.
More Tokens, More Output? Not Quite
Some developers, however, see Huang’s benchmark as far removed from reality.
A common criticism is that token consumption does not directly correlate with productivity. Engineers can generate large volumes of AI output without necessarily improving outcomes. Critics argue that such metrics risk encouraging wasteful usage rather than thoughtful integration.
Vijayakumar Nadar, Chief AI Officer at NeevCloud, tells AIM that the $250,000 token benchmark is directionally insightful, but needs context. “It’s not a reflection of actual pricing; it’s a statement about the economic value AI can unlock.”
In high-value enterprise scenarios such as code generation, financial modelling, or autonomous workflows, he says, tokens are not just text but outcomes.
He further shares that NeevCloud doesn’t currently allocate tokens per employee. “We enable access and monitor usage, but focus more on outcomes than on limiting consumption,” he adds.
The company tracks token usage for visibility and capacity planning, but not as a success metric. “What matters more is the value created, faster development, better decisions, and real business outcomes.”
In a LinkedIn post, Pradeep Sanyal, an enterprise AI leader, questioned the numbers behind Huang’s claim. He noted that at current pricing, the benchmark would imply tens of billions of tokens per engineer annually, far beyond practical usage.
“No engineer is reading, validating, or directing output at that scale,” Sanyal said, adding that such figures would only make sense if “autonomous pipelines are running continuously in the background.”
Meanwhile, Rakesh Ravuri, Chief Technology Officer and Senior Vice President of Engineering at Publicis Sapient, says, “For us, AI is not about how much is consumed, but about whether it is genuinely shaping how work happens. Much like any core capability, its value shows up in the quality, speed, and consistency of what teams deliver.”
“We track AI usage patterns to understand adoption trends and where value is being created, but token consumption is not treated as a performance metric, nor do we allocate fixed token budgets per employee," he adds.
Ravuri notes that the company is still in a phase where the value created through AI adoption significantly outweighs the underlying consumption costs. As usage grows, governance will naturally evolve.
“Improvements in velocity, quality, defect reduction, and overall execution are the clearest signals that AI is being used well.”
However, in real-world use, the benefits are less predictable.
An engineer at Sarvam AI tells AIM that while AI does improve productivity, the gains are not linear. “Beyond a certain level of complexity, it may not help as much if you don’t deeply understand the system,” they say. The company currently uses tools such as Cursor and Claude Code.
Ankush Sabharwal, CEO of Corover.ai, offers a more measured perspective. He says that while leaders may set bold directives to drive adoption, the real goal is to solve problems and create value. “It should be seen as an immediate push rather than a long-term benchmark,” he adds.
Whether taken literally or rhetorically, Huang’s statement highlights a turning point in enterprise AI adoption. The question is no longer whether to use AI, but how much is enough and how to ensure that spending translates into real productivity gains.