Research Infrastructure Engineer Jobs USA 2026 | Thinking Machines Careers

Do you want to build the systems that power the next big AI models? AI models are getting bigger and bigger. Some now have trillions of parts (called parameters). Old-style software design cannot handle this size anymore. Today, top machine learning engineers need strong skills in three areas: high-performance computing (HPC), building custom GPU code, and managing many computers working together (called clusters).

If you are a senior infrastructure engineer or an AI researcher, and you want to grow your career in 2026, Thinking Machines Lab has some of the best Research Infrastructure Engineer jobs in the USA. The company is based in San Francisco, California. This top research lab is now hiring skilled engineers. These engineers will help build systems for training models, improving models after training, and using low-precision computing methods.

Why Choose Thinking Machines Lab?

Thinking Machines Lab was started by well-known researchers and builders. These are the same people who helped create ChatGPT, Character.ai, Mistral’s open models, and open-source tools like PyTorch, JAX, OpenAI Gym, and Segment Anything. Because of this, Thinking Machines Lab is seen as a leader in machine learning.

This job is different from a normal software engineering job. A Research Infrastructure Engineer at Thinking Machines Lab does not just keep servers running. Instead, these engineers design new ways for hardware and software to work together. They make training work better across thousands of GPUs. They also build fast pipelines for Reinforcement Learning (RL) systems.

Specialized Career Tracks at Thinking Machines Lab

Thinking Machines Lab has different job tracks for infrastructure work. Each track solves a different problem in building AI models.

Research Engineer, Infrastructure (RL Systems)

This job focuses on building fast pipelines for rollouts and rewards. It also helps make Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimization (DPO), and Proximal Policy Optimization (PPO) work well on large computer clusters. Engineers in this role often use tools like Prometheus, Grafana, OpenTelemetry, Slurm, and Kubernetes to watch how systems are performing.

Research Engineer, Infrastructure (Numerics & Acceleration)

This job focuses on making the math behind model training better and faster. Engineers build new low-precision number formats, such as BF16, MXFP8, and NVFP4. They also write custom code to help GPUs run faster, without hurting how well the model learns. Common tools used here include Triton, CUDA, PyTorch/XLA, Megatron-LM, DeepSpeed, and Block-Floating Point (MX) number tools.

Salary, Compensation, & USA Visa Sponsorship (2026 Data)

Thinking Machines Lab pays very well. This helps them attract the best talent from around the world.

Compensation Part	Package Details (San Francisco, CA)
Base Salary Range	$350,000 – $475,000 USD per year
Equity Grant	A large amount of company shares for early employees
USA Visa Sponsorship	Fully supported (H-1B, O-1, TN, and Green Card help)
Work Arrangement	Hybrid / In-office (San Francisco headquarters)
Perks & Benefits	Good healthcare, help with moving costs, and unlimited paid time off

Core Skill Matrix for Prospective Applicants

To pass the technical interview for a Research Infrastructure job at Thinking Machines Lab, you should be strong in these four skill areas:

Distributed Systems & Compute: You should have real experience training large models. This includes using Tensor, Pipeline, Sequence, and Data Parallelism across thousands of connected computers.

Numerics & Acceleration: You should understand the trade-offs of different number formats. You should also know how to write custom code and use methods that reduce the amount of data sent between computers during training.

RL & Post-Training Systems: You should be able to write clean and fast code in Python and C++. This code is used for testing models, connecting benchmark tools, and improving how models learn from feedback.

HPC Reliability & Observability: You should have hands-on experience finding and fixing problems. This includes broken hardware, slow networks, and memory issues in large GPU training systems.

Also Read: Test Lead Jobs UK 2026 | Scrumconnect Consulting Careers

How to Apply via Greenhouse & Interview Guidance

Thinking Machines Lab mostly hires people through its official Greenhouse job page and other trusted job websites. Here are some simple tips to help you apply: First, build a strong portfolio. Show your open-source work on well-known machine learning tools like PyTorch, DeepSpeed, JAX, Triton, or vLLM.

Second, apply directly through the Thinking Machines Greenhouse page. You can also check verified job posts on Ashby, Lightspeed Venture Partners, or LinkedIn.

Third, remember that many infrastructure jobs at Thinking Machines Lab stay open all the time. The hiring team keeps looking at new applications. They reach out when a new project needs more people. So you can apply again every six months as you learn new skills.

Frequently Asked Questions (FAQs)

Disclaimer: This article is written only to help people learn about careers. It is for information and search purposes only. It is not an official job ad. It is not made by, paid for, or approved by Thinking Machines Lab. Job details, salary numbers, visa rules, and how to apply may change at any time. This depends on the company’s own policies and hiring plans. Before you send any personal information, please check the real and current job details on the official Thinking Machines Lab Careers Portal (through Ashby or Greenhouse).