Infinity Hires Meta and CUDA Veterans to Automate AI Inference Software

Infinity, a San Francisco automated AI research company founded in August 2025, has hired Aditya Om from Meta Superintelligence Labs as Staff Inference Optimization Engineer and Elliot Arledge as Member of Technical Staff. The hires target one problem: the slow, manual work of preparing new AI chips to run models efficiently.

The appointments back development of Ignition, Infinity's autonomous research agent. Ignition generates, tests and optimizes the low-level software, or compute kernels, that determine whether a chip reaches competitive inference performance.

"We're hiring engineers to solve the greatest engineering challenges of our generation," said Jeremy Nixon, Founder and CEO of Infinity. He frames the goal plainly: automate inference software development and cut the time needed to ready new silicon for advanced models. A promise of automation, however, still needs public, reproducible evidence at scale.

Who are the new engineers?

Om arrives from Meta Superintelligence Labs, where he built software infrastructure for Meta's internal AI training and inference chips. Before Meta, he worked at Lightmatter on collective communications software for the Passage platform. He contributed to Lightmatter research on its Envise technology published in Nature. Earlier, he served as an Internet of Things systems design engineer at Intel. Om holds a master's degree in Electrical Engineering and Computer Science from the University of Michigan, Ann Arbor.

Arledge is a self-taught CUDA and kernel engineer. His educational courses on large language models and compute kernels have reached more than two million learners, according to the company. At Infinity he will work on automated AI infrastructure development. Reach is a distribution figure, not a verification of engineering output, and the company does not claim otherwise.

What problem does Ignition attack?

Emerging AI chip makers face a structural disadvantage. Established platforms ship with mature software ecosystems; new architectures require substantial engineering before models run well in production. Infinity's answer is AI agents that write, test and refine kernels using hardware performance feedback, reducing dependence on large human compiler and kernel teams.

What results has Infinity actually shown?

The company's strongest claim comes from a previously published case study with chip developer d-Matrix. Infinity reported reaching 92% of a chip's theoretical peak performance within 10 hours of initial hardware access. It also reported running three frontier AI models end-to-end within 10 days.

These are company-reported figures from a case study, not independent benchmarks. The company did not name the three models, specify the chip measured, or disclose comparison baselines. The 92% figure deserves scrutiny until third parties reproduce it.

Infinity maintains a design partnership with d-Matrix. The d-Matrix Corsair chip supports inference workloads through the Infinity d-Matrix Cloud.

The company has also announced a collaboration with HP Inc. to optimize NVIDIA's Nemotron 3 Super model on the HP ZGX Fury platform, powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip. No performance targets or timelines accompanied that announcement.

Who backs the company?

Nixon founded Infinity in August 2025. He is a former Google Brain researcher and co-founder of AGI House. Touring Capital backs the company, along with angel investors whose backgrounds span AI research and semiconductor technology. Infinity did not disclose its funding amount or valuation.

The company continues to recruit engineers in AI infrastructure, inference optimization and low-level software development as it scales the automated research platform.

What comes next?

Infinity is betting that automated kernel generation can compress chip bring-up from months to days. The d-Matrix numbers suggest early traction; broader adoption will depend on published, verifiable results across architectures beyond a single partner's silicon.