This metric could soon become the new test for engineers, says Nvidia CEO Jensen Huang

From code output to compute usage, Huang argues token consumption may become the new productivity benchmark in AI-driven workplaces

• Nvidia CEO suggests AI token usage may become key for engineers
• Token budgets could influence hiring and pay in software roles
• Engineers may be evaluated on efficient use of AI compute

Did our AI summary help?

If any profession is being reshaped most rapidly by artificial intelligence, it is software engineering. Now, Nvidia CEO Jensen Huang says the shift could go deeper, into how engineers are evaluated, hired, and paid.

Speaking on the All-In Podcast, Huang suggested that in an AI-driven world, traditional measures of productivity such as lines of code or output may give way to a new metric: how effectively engineers use AI compute, measured in 'tokens'.

From code to compute: a new productivity benchmark

Huang argued that as AI becomes embedded in everyday development workflows, token usage, the unit of compute consumed by AI systems, could emerge as a key benchmark for performance.

In practical terms, this means companies may begin to assess engineers not just on what they produce, but on how extensively and efficiently they leverage AI tools to amplify output.

“Let’s say you have a software engineer or AI researcher and you pay them $500,000 a year, if that person did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed,” Huang said.

The argument reflects a broader shift: expensive talent, in Huang’s view, should be paired with heavy use of powerful AI systems, or companies risk underutilising both.

What are AI tokens and why they matter

AI tokens are the basic units used by artificial intelligence systems to process tasks such as generating text, writing code, or analysing data.

Every input prompt and output response consumes tokens, effectively representing the computing power used to complete a task. Since companies pay for this compute, token usage becomes a measurable proxy for how much AI assistance an engineer is deploying.

For instance, pricing for leading models, such as Anthropic’s Claude or systems from Google and OpenAI, is typically structured per token, running into a few dollars per million tokens depending on usage.

As AI adoption scales, this cost-linked metric is increasingly being seen as a way to quantify productivity in AI-heavy roles.

Token budgets could shape hiring and pay

Huang suggested that companies may soon allocate 'token budgets' to engineers, much like salaries or project resources today.

Access to compute, he argued, could become as important as compensation, with firms offering larger AI budgets gaining an edge in attracting top talent.

The implication is significant: engineers may be judged not only on output, but on how aggressively they use AI tools to accelerate work. In such a model, under-utilising available compute could be seen as inefficiency.