July 6, 2026
Payroll vs. the robot appetite
When AI Costs More Than the Engineer
Readers roast the math as AI bills balloon past human paychecks
TLDR: The article claims some AI companies now spend far more on computing power than on employees, with Anthropic used as the extreme example. Commenters mostly fought over the math, saying it unfairly mixes the cost of building AI with the cost of simply using it.
The big, eye-popping claim in this debate is simple: at top artificial intelligence companies, the money spent on computer power can be way bigger than payroll. The article points to Anthropic spending about $2 million in compute per employee per year, versus compensation of roughly $500,000 or more, and warns that other software companies could see their own AI costs climb fast. But the comment section did what comment sections do best: it grabbed the calculator, flipped the table, and yelled "absolutely not."
The loudest backlash was over what many readers saw as a sneaky comparison. Several argued that counting the enormous cost of building and training an AI model is not the same as counting what ordinary companies pay to use one. One commenter dismissed it as "apples and potatoes," while another delivered the thread’s funniest dunk: comparing model training to everyday AI use is like counting the effort to grow a tree and mine iron before deciding whether to hammer a nail with a hammer or your fist. Ouch.
Still, not everyone threw the whole thing out. Some said that even after removing the weirdest accounting, the broader warning remains: a small elite of companies is spending serious money on AI while the median business barely spends anything at all. Others were stunned that even the so-called bear case still assumes AI costs at 40% of salary, calling that number wild on its own. In other words, the community verdict is messy, dramatic, and very online: the trend might be real, but readers are furious about the way it was framed.
Key Points
- •The article estimates Anthropic spends about 2.3 times its payroll on compute, or roughly $2 million per employee annually based on 2026 spending and headcount figures.
- •It says the top 1% of software companies spend about $89,000 per engineer per year on AI, while the median company spends $137 annually.
- •The article models bear, base, and bull cases for AI spend per engineer through 2029, ranging from 41% to 230% of salary by 2029 depending on the scenario.
- •It argues that in the bull case, AI cost per engineer could equal an entire median SaaS employee's annual revenue contribution.
- •The article cites rising token demand from agentic workflows and falling token prices plus lower-cost open-weight models as the main opposing forces shaping future AI spend.