Infinity raises $15M to run AI inference on any chip, challenging Nvidia's software dominance

Inference startup Infinity raises $15M from Touring Capital, OpenAI and Athropic researchers

AI infrastructure company Infinity announced a $15 million raise at a $100 million valuation on Monday from investors including Touring Capital, Principal VC, and researchers from companies such as OpenAI and Anthropic. The startup is building software to make it easier for AI chips to run AI models. One big reason Nvidia became the top player is not just its high-performance chips, but also its CUDA software (Compute Unified Device Architecture), which allows its GPUs (originally designed to run graphics) to act as general-purpose processing CPUs. The largest AI development frameworks PyTorch and TensorFlow have been built on top of CUDA. This allows developers to write their apps in popular languages like Python, use those major AI frameworks and their apps will, by default, run on Nvidia chips. Most of these app-level startups wouldn't have the resources or know-how to write their own kernels -- the low-level software that operates chips -- and port their apps to other AI chips. So Infinity is trying to build CUDA-alternative kernel software that works with any type of chip, like SRAM, GPUs, phone chips, and Systolic Arrays. Infinity is part of a new wave of startups that are attempting, product by product, to chip away at Nvidia's market dominance. Infinity is attempting to build a universal inference library to run on all chips, allowing these chips to automate replicating state-of-the-art research results. Infinity was launched last year by Jeremy Nixon, once a researcher at Google Brain and creator of the hacker network community AGI House. Nixon told TechCrunch he decided to launch this company because he was obsessed with the idea of "automated invention" -- the belief that "AI systems can actually be a meta technology." He himself had invented a machine learning algorithm called Omega, he said, which essentially created new machine learning algorithms and automatically evaluated them in a feedback loop. That success got him thinking about other cases where this approach could work, and he turned to hardware, believing that automated systems could also generate the low-level code, like the kernels and so forth, needed to help run chips more effectively. Infinity's AI research agent Ignition is intended to write the low-level code needed for AI inference on Nvidia-alternative chips. It tests, debugs, and measures how fast the hardware performs with the code, and automatically rewrites the code if needed to improve performance. The system is self-optimizing, meaning it continuously learns and improves itself. It also adapts to different chip architectures, regardless of proprietary designs, Nixon says. The result is what Infinity claims is a CUDA-level software stack. Customers include the AI chip maker (and would be Nvidia challenger) D-Matrix, and Infinity is in talks with other big chip and cloud companies, Nixon said. Humans are in the loop, however, providing high-level direction while the agent does more of the tedious grunt work. In one case study, the startup found the agent works much faster than a human alone, reducing what could have been a years- or months-long process to hours or days. Infinity doesn't charge an upfront license fee; instead, it takes a cut of performance gains and cost savings, measuring changes in tokens per second. Right now, Infinity has 26 employees, including those in design, operations, and engineering.

Infinity raises $15M to make any AI chip inference-ready

Nvidia's grip on AI rests on more than fast chips. It rests on software. A San Francisco startup called Infinity has raised $15 million to loosen that hold. Its weapon is an AI agent that teaches any chip to run models in days, not years. The pitch is bold for a seed round. Infinity says it can strip away the one advantage that has kept Nvidia untouchable. It has just raised money to try. The company closed a $15 million seed round at a $100 million valuation on Monday, it announced. Backers include Touring Capital, Principal VC, executives at major chip firms, and researchers from OpenAI and Anthropic. The CUDA problem To understand what Infinity is chasing, look at why Nvidia won. Its chips are quick, but its real moat is CUDA, the software layer it has built up over nearly two decades. The big AI frameworks, PyTorch and TensorFlow, sit on top of CUDA. Write your app in Python and it runs on Nvidia hardware by default. That convenience is why Nvidia holds an estimated 80% of the data-centre AI accelerator market. Rival chips from AMD, Qualcomm, and AWS often match Nvidia on raw power. What they lack is the software. Porting models to new silicon means writing kernels, the low-level code that drives a chip, and few teams can afford the effort. An agent that writes the chip code Infinity wants to automate that work. Its AI agent, Ignition, generates, tests, and rewrites those kernels itself, and keeps tuning them as performance data comes back. Human engineers set the direction. The agent does the grind. The results Infinity reports are striking, though they come from the company itself. Working with the chip maker d-Matrix, it says Ignition hit 92% of a new chip's peak performance 10 hours after first touching the hardware. Within 10 days, three frontier models were running on it end to end. On another test, Infinity says it lifted a model's output from about 1,400 to more than 20,000 tokens a second in a single day. That would beat vLLM, a widely used open-source inference framework. The claims are not yet independently checked. Automating invention The founder gives the project its flavour. Jeremy Nixon is a former Google Brain researcher who created AGI House, a San Francisco hacker network that says it has spawned hundreds of startups. Nixon told TechCrunch he is obsessed with "automated invention," the idea that AI can be a kind of meta technology. He once built an algorithm, Omega, that invented other machine-learning algorithms and scored them in a loop. Ignition applies the same instinct to hardware. He announced the raise on X with a flourish, hailing the arrival of AI systems that "enable, optimize and invent" the next generation of AI systems. A crowded race, and the caveats Infinity joins a wave of startups trying to break the CUDA lock-in. It says it already earns millions in annual recurring revenue and employs 26 people. Its business model takes a cut of the speed and cost gains it delivers, rather than a licence fee. The caveats are real. This is a seed-stage firm with one public chip partner, and the headline benchmarks are self-reported. The prize, though, is large. As inference races to become two-thirds of all AI compute spending this year, cheaper ways to run it are worth a fortune. "The next era of AI will be defined not just by who makes the best chip, but by who can make any chip run state-of-the-art models at blazing speeds," Nixon said. If Ignition works as billed, the moat that made Nvidia untouchable starts to look shallower.

Infinity raises $15M to run AI inference on any chipset

Early-stage AI infrastructure research company Infinity Inc. said it raised $15 million in seed funding to develop software that automatically prepares new artificial intelligence chips to run inference workloads. The round values the company at $100 million on a post-money basis. Infinity will use the capital to expand its engineering team, scale its automated research platform and accelerate partnerships with chipmakers including d-Matrix Corp. The company said it is already generating millions of dollars in annual recurring revenue from chip design partnerships. The company's core technology is Ignition, an AI agent that generates, tests and optimizes the low-level software needed to run models efficiently on different processors. Infinity is targeting a longstanding obstacle for would-be Nvidia Corp. competitors: Even when their hardware is competitive, they lack a mature software ecosystem comparable to Nvidia's Cuda. The company said its technology can help loosen Nvidia's grip on the market for AI processors by generating what is effectively a Cuda-level software stack in hours or days.. Ignition automates "the creation of debuggers, profilers, compilers, kernels, the orchestration of kernels in inference, and many other steps," said Founder and Chief Executive Jeremy Nixon. Computing kernels are specialized software routines that perform core mathematical operations on AI chips. Traditionally, engineers must write and tune them for each combination of processor architecture and model, making it difficult for new hardware to keep pace with rapidly changing AI technology. Infinity said human engineers determine what tools Ignition should create, while the agent handles implementation. To work with proprietary instruction sets, the company builds decompilers, which translate executable code back into high-level source code that can be manipulated using standard programming languages. "Humans decide what tools the AI will build, such a memory layout compiler, while Ignition does the building," Nixon said. Rapid results The company said Ignition increased inference throughput for the Qwen3-8B model by 34% compared with the widely used vLLM framework after one day of automated optimization. Nixon said the test used a single Nvidia H100 graphics processor and has not yet been independently validated. In a project with d-Matrix, Infinity said its software achieved as much as 92% of the theoretical peak performance of the company's Corsair accelerator within 10 hours. It accomplished this by distributing matrix multiplication operations across all 32 computing units. Infinity said it had Qwen3, Qwen3.5 and Gemma4 running fully on the chip within 10 days. Infinity tests generated software with tools that enforce required memory and numerical behavior. "We build 'bit-for-bit' accurate checking systems, memory clobber detectors and more," Nixon said. The startup describes Ignition as a recursively self-improving system, meaning that it records successes and failures, builds tools to preserve successful techniques and constructs a representation of previously solved problems to make subsequent optimization runs more efficient. Infinity's revenue model links compensation to improvements it produces rather than software licensing fees. Nixon said optimization agreements generally give Infinity about 20% of the savings on additional computing purchases, calculated from gains in model throughput against an agreed baseline. The company was founded a year ago by Nixon, a former Google Brain researcher and co-founder of the AGI House Labs Inc. hybrid community hub, applied AI lab, and venture capital firm. Infinity said the new funding will also support discussions with additional chip companies. Touring Capital LLC was a significant participant, along with Principal Venture Partners LP and unnamed executives and angel investors from OpenAI LLC, Anthropic PBC and chip makers.