Tomasz Tunguz blog · 2026-02-17 · 107d

AI Inference Costs as a Fourth Component of Engineering Compensation

As AI inference costs become a significant portion of total employee compensation (21% for a $375k engineer), companies are beginning to evaluate and optimize these expenses. The author demonstrates how migrating from commercial AI services ($100k annually) to open-source models can reduce costs by 88% while maintaining performance, raising the question of whether future compensation will be directly tied to inference token consumption.

9 metrics· Cited 0× in the knowledge base ·Open source ↗

Metrics in this report

Annual Inference Cost Per 31 Tasks Daily

12000$

post-optimization

author's cost efficiency baseline

Daily Inference Invoice

92$

peak

task automation peak spending

Daily Task Completion Rate

31tasks/day

with AI assistance

author's productivity metric

Inference Costs as Compensation Percentage

21%

calculated ratio

for $375k engineer with $100k inference spend

Monthly AI Service Subscription Cost

200$

starting point

individual Claude API usage

Monthly AI Service Subscription Cost

600$

after adding multiple agents

Codex, Gemini, Claude Code subscriptions

Open-Source Model Cost Reduction

12%

percentage of original cost

post-migration inference expenses

Quarterly Inference Cost Progression

7200, 43000, 100000$

annualized run rate over two quarters

total inference spending escalation

Software Engineer Salary 75th Percentile

375000$

p75

per Levels.fyi data