Coding alone could be a half-trillion-dollar AI market
Early helpers cost $20 a month, but newer tools that do the work themselves run about $100 a day in heavy use.
Invest Like the Best with Patrick O'Shaughnessy · Why the AI Boom Is Just Getting StartedWriting computer code alone could be a roughly half-trillion-dollar market for artificial intelligence.
Investor Alex of Whale Rock Capital, an investment firm, puts the math this way. The first tools, like Microsoft's Copilot at about $20 a month, just improved what he calls the grammar of coding, catching a bug or drafting a short block.
Then Anthropic, an AI company best known for its Claude assistant, released a tool around the middle of 2025 that could work agentically, carrying out multi-step coding tasks on its own. People inside Anthropic who could use it without limits were spending about $100 a day on tokens, the units of computing the models consume, or about $20,000 to $30,000 a year per user.
Multiply that by about 20 million coders worldwide, and coding alone becomes enormous. And that calculation was based on technology that was already seven to nine months old.
Questions people ask
What does it mean for a coding tool to run agentically?
It means the AI does more than suggest the next few lines. Given a goal in plain English, it can plan the steps, write files, run and test the code, find errors, and revise the work with little supervision. That is why heavy users consume far more computing than with autocomplete style helpers.
What is a token, and why does it cost money?
A token is a small chunk of text or code that the model reads or generates, roughly a few characters. AI companies charge by tokens processed because each token requires computing work in a data center. An agent that writes, tests, and rewrites thousands of lines can burn through millions of tokens quickly.
Would employers really pay $20,000 to $30,000 per coder per year?
Many likely would if the output justifies it, because a professional programmer often costs well over $100,000 a year in salary and benefits. If AI lets one programmer do the work of more than one, or finish projects much faster, the token bill can still be a bargain. The open question is how the price falls as chips and models get more efficient.