Also called: CUDA Engineer, AI Performance Engineer
Writes very low-level code that makes AI models run as fast as possible on graphics chips.
You rewrite a single math routine that runs trillions of times while training an AI model, squeezing out 15% more speed. That small gain can save a company millions of dollars in electricity and chip time.
Among the highest-paid engineering niches because so few people can do it.
Approximate: anchored on Bureau of Labor Statistics data for a broader job group.
Master's degree, about 6 years after high school. Bachelor's or master's in computer science or engineering with deep knowledge of computer architecture.
Ways in: 4-year college, Graduate school. Training cost: $15k – $60k.
Majors to look at: Computer Science, Computer Engineering, Electrical Engineering.
Growing much faster than average. Demand exploded with the AI chip boom.
Low risk in 5 years, medium in 10. AI is starting to generate optimized kernels, but squeezing peak speed from new chips stays rare expert work.
Classes: AP Computer Science A, AP Physics C, AP Calculus BC.
Official data: O*NET occupation 15-1252.
Career Universe is an early version. Pay, training and AI-risk figures are estimates to start your own research, not verified facts. The research was written with AI research assistants, checked by validation scripts and spot-checked by hand. Check anything important with schools, employers and official sources. How the numbers work.