Latest AI JobsBrowse Jobs
Anthropic logo

TPU Kernel Engineer

AnthropicSan Francisco, CA | New York City, NY | Seattle, WAAI Research & Engineering
Apply on company site
Salary
Not listed
Job type
Full-time
Posted
22 May 2025
Status
Open · checked 7d ago

About this role

The TPU Kernel Engineer position at Anthropic involves tackling complex performance bottlenecks across machine learning systems used for research, training, and inference. Based in San Francisco, CA, New York City, NY, or Seattle, WA, this role is designed for professionals skilled in low-level optimization and large-scale system problem-solving.

Role Overview

As a TPU Kernel Engineer, the primary focus centers on designing and optimizing custom kernels specifically for Tensor Processing Units. Professionals in this position address performance challenges across diverse machine learning workflows while collaborating closely with researchers to evaluate how model adjustments influence overall system speed and efficiency. Team members frequently engage in pair programming and tackle complex projects such as implementing low-latency text sampling, designing custom collective communication protocols, and debugging performance down to the assembly level.

Key Responsibilities

Identify and resolve performance challenges across various machine learning architectures and workloads.
Design, build, and refine specialized kernels targeting TPUs and alternative hardware accelerators.
Provide technical feedback to research teams regarding the performance implications of model modifications.
Construct quantitative models to evaluate system performance accurately.
Adapt existing machine learning models to support low-precision inference effectively.

Requirements

Demonstrated history of resolving complex, large-scale systems issues combined with low-level optimization skills.
Relevant educational background at the bachelor's degree level or an equivalent blend of formal training and professional experience.
Alignment with a hybrid work policy requiring staff presence in a designated office at least 25% of the time.

Nice to Have

Extensive background in optimizing machine learning workloads specifically for TPUs or GPUs.
Deep technical understanding of computer architecture and accelerator internals.
Practical familiarity with machine learning framework internals and transformer-based language modeling.

Skills and Tools

TPU and accelerator optimization techniques
Low-level kernel design and assembly-level debugging
Large-scale machine learning systems infrastructure
Quantitative system performance modeling

Salary and Benefits

The annual salary for this position ranges from $280,000 to $850,000 USD. Anthropic provides competitive compensation alongside benefits including optional equity donation matching, generous paid time off, parental leave, flexible working hours, and visa sponsorship support where applicable.

About the Company

Anthropic operates as a public benefit corporation dedicated to developing reliable, interpretable, and steerable artificial intelligence systems that benefit society. The organization treats artificial intelligence research as an empirical science, functioning as a single cohesive team focused on large-scale initiatives that advance long-term safety and trustworthiness.

How to Apply

Interested candidates should visit the official Anthropic careers page to submit an application and review open positions.

Questions candidates ask

What locations are available for the TPU Kernel Engineer role?

This position can be performed from San Francisco, CA, New York City, NY, or Seattle, WA, adhering to a hybrid policy requiring at least 25% time in the office.

What is the salary range for this position?

The annual base salary ranges from $280,000 to $850,000 USD.

Does Anthropic offer visa sponsorship for this job?

Yes, visa sponsorship is available, and the company partners with an immigration lawyer to assist candidates who receive an offer.

What are the core technical duties of a TPU Kernel Engineer?

Responsibilities include designing and optimizing TPU kernels, debugging performance issues at the assembly level, building quantitative performance models, and collaborating with researchers.

What education or experience level is required?

Candidates need a bachelor's degree in a relevant field or an equivalent combination of education, training, and experience, along with a track record in low-level optimization.

Ready to apply?

The application is on the official Anthropic careers page.

Apply on company site