Universal AI Kernel
Infinity Secures $15M Funding to Power AI Across Diverse Chips, Challenging Nvidia's CUDA
Anonymous
Key Points
1
Infinity specifically targets Systolic Arrays in addition to SRAM, mobile chips, and GPUs, broadening the range of architectures its universal kernel aims to run on beyond typical GPU-based systems.
2
Infinity’s Omega algorithm enables the system to generate new algorithms and automatically evaluate them in a feedback loop, underpinning its vision of automatic invention across hardware platforms.
3
Infinity was launched last year by Jeremy Nixon, who also founded the AGI House hacker community, highlighting the founder’s long-standing interest in automated, meta-technologies for AI.
4
The article notes that PyTorch and TensorFlow are built on top of CUDA, underscoring Nvidia’s software ecosystem and reinforcing Infinity’s aim to abstract low-level details so AI workloads can run on non-NVIDIA hardware.
AI infrastructure startup Infinity has raised $15 million in seed funding at a $100 million valuation to build software that lets AI models run on nearly any chip — directly challenging Nvidia's grip on the AI industry, according to SiliconAngle and Pulse2.
The round was led by Touring Capital, with participation from Principal VC and researchers from OpenAI and Anthropic. AI chip company D-Matrix is already using the technology, Mezha reported.
CUDA Has a Stranglehold — Infinity Wants to Break It
Most AI software today runs on Nvidia's CUDA platform. Popular frameworks like PyTorch and TensorFlow are built directly on top of it, according to Mezha. That means companies buying non-Nvidia chips often face a painful, expensive process to get their software working on new hardware.
Infinity wants to fix that. Its software acts as a universal layer that sits between AI models and the chip. It aims to support a wide range of hardware — GPUs, mobile chips, SRAM chips, and even Systolic Arrays — without developers having to rewrite code for each one, TechBuzz reported.
An AI Agent That Writes Its Own Chip Instructions
At the core of Infinity's system is an AI agent called Ignition. It automatically generates, tests, and improves low-level "kernels" — small programs that tell a chip exactly how to process data. Writing these by hand is slow and highly technical work, according to SiliconAngle.
Infinity also uses an algorithm called Omega, which creates new algorithms and scores them in an automated feedback loop. The company calls this "automatic invention" — a way to discover better solutions without human engineers doing the work manually, TechBuzz reported.
Founder Jeremy Nixon Bets on Meta-Technology for AI
Infinity was founded by Jeremy Nixon, a former Google Brain researcher who also started the AGI House hacker community. Nixon launched the company last year with a focus on what he calls meta-technologies — tools that can automate other tools, according to Pulse2.
Nixon's vision goes beyond chip compatibility. He wants Infinity's system to autonomously reproduce and verify cutting-edge AI research results across different hardware platforms. That ambition is part of why researchers from OpenAI and Anthropic backed the round, TechBuzz noted.
Early Customer D-Matrix Shows Real-World Traction
D-Matrix, a startup building SRAM-based AI inference chips, is already using Infinity's software. Its Corsair chip is one of the emerging alternatives to Nvidia's GPUs. Getting software to run efficiently on chips like this has historically required months of manual engineering work, Digg reported.
Infinity says its system can match or beat vLLM — a widely used Nvidia-optimized inference tool — on performance. If true, that would be a significant milestone. It would mean chip makers could skip the painful CUDA migration process entirely and go to market faster, according to SiliconAngle.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
10
Articles
10
Reach
20
A
AnonymousAuthor