CUDA: AI Reshuffles Roles, Nvidia Salaries at $431,250

Low-level programming of Nvidia GPUs is being reconfigured under the influence of automation. Job offers related to CUDA are on the rise, salaries peak at Nvidia, while the daily tasks of specialists shift towards supervising agents and reviewing AI-generated code.

In the United States, CUDA job offers already exceed 2025, and Nvidia offers $431,250

In the United States, during the first eight months of 2026, job offers requiring CUDA skills have already surpassed the total for 2025, according to Lightcast. Nvidia remains the primary employer, with over 300 positions available in September for candidates with CUDA expertise. Nvidia's public job postings for engineering positions related to CUDA mention base salaries reaching up to $431,250, not including stock options and other benefits. These jobs are offered not only at Nvidia but also in specialized AI labs and among major cloud players.

From code to instructions: AI generates and tests hundreds of kernels

In the past, CUDA engineers wrote the kernels themselves and conducted tests to determine which was the most efficient. Now, a large portion of these tasks is handled by AI, which generates several hundred kernels, evaluates them, and retains the most effective version. Engineers are therefore dedicating an increasing amount of their work to supervising code generation agents. According to Jeremy Nixon, founder of Infinity, this evolution is part of a broader trend where developers set objectives for AI, verify its results, and intervene when difficulties arise. CUDA remains Nvidia's programming tool for executing AI optimally, but operational execution is shifting towards automation.

More demanding code reviews and rising specific demand

Automation introduces bugs described as "strange" by Anne Ouyang, co-founder of Standard Kernel, making the review of AI-generated code more demanding. Despite this evolution, the demand for CUDA engineers remains strong, while the overall demand for software engineers is lower than in 2023, according to Elena Magrini of Lightcast. High computing costs and engineers' ability to optimize chip usage, saving millions, explain the ongoing interest in these specialized profiles.

Ongoing shortage and career redistribution

The scarcity of CUDA engineers remains a challenge, even though automation has increased productivity, according to Bing Xu, founder of INT21. He notes that the most advanced expertise in CUDA has been built over more than 20 years, well before the rise of AI. Xu believes that AI partially fills the gap left by the lack of experienced engineers. Jeremy Nixon, founder of Infinity, observes that automation lowers the entry barrier for less experienced profiles and directs veterans towards supervisory tasks. Anne Ouyang, co-founder of Standard Kernel, believes that junior engineers may be the most affected, while those capable of verifying and surpassing AI will gain in value. The demand for this rare expertise continues to grow as engineers increasingly manage the AI that performs their optimization work.

Early signs of "superhuman" AI in kernel writing

In the future, artificial intelligence could play an even more prominent role in kernel creation. According to Bing Xu, the work conducted by his company indicates that AI is already capable of surpassing human engineers in certain aspects of kernel programming. Jeremy Nixon, founder of Infinity, states that in some situations, AI generates CUDA code that engineers cannot fully understand, even though they can verify its validity. He considers this ability to surpass human engineers as an early glimpse of an AI described as "superhuman" in the real world. CUDA engineers have remained among the most sought-after specialists in the tech industry for years.