Infinity raises $15M seed to make any AI chip inference-ready
What's the deal? InfinityDealroom has a profile for this one. Try Dealroom →, an early-stage AI infrastructure company, has raised $15 million in seed funding at a $100 million post-money valuation. Touring CapitalDealroom has a profile for this one. Try Dealroom → and Principal VCDealroom has a profile for this one. Try Dealroom → led the round, joined by founder Jeremy Nixon, executives at major chip companies, and researchers from OpenAI and Anthropic.
What's the endgame? Infinity builds the software layer that makes any AI chip ready to run inference. Its autonomous agent, Ignition, writes and optimises the low-level compute kernels that determine how efficiently a chip runs AI models.
The funding will scale that research platform, expand Infinity's engineering team, and accelerate work with chip partners including d-Matrix.
By the numbers: Infinity is already generating millions of dollars in annual recurring revenue from chip design partnerships. In one test, it improved inference throughput on a Qwen3-8B model by 34% more tokens per second in a single day, and reached up to 92% of a new d-Matrix chip's theoretical peak performance within 10 hours of first accessing the hardware.
Why now? Nvidia holds an estimated 80% share of data-center AI accelerators, anchored by its CUDA software ecosystem. Rivals like AMDDealroom has a profile for this one. Try Dealroom →, QualcommDealroom has a profile for this one. Try Dealroom →, and AWSDealroom has a profile for this one. Try Dealroom → often match its hardware, but building production-grade inference stacks has been slow and manual.
The signal: Inference is on track to represent two-thirds of all AI compute spending in 2026. "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," said Nixon — a bet that software, not silicon, decides which chips reach their potential.
Read more: TechCrunch
Image credit: NeoSpire
