Eighty chip startups collectively raised over $8.5 billion in the third quarter of 2026, Semiconductor Engineering reported Monday in its quarterly startup-funding tally. Nearly $7.5 billion of the total went to just 22 startups that raised $100 million or more, including eight that saw investors return within roughly a year, some more than doubling the previous round.
The two headline rounds were both in AI inference. Etched raised $1 billion across two rounds, a $300 million round in July led by Sequoia Capital and another $700 million in August led by Jane Street, for its rack-scale inference chips built on TSMC N4P that the company says run trillion-parameter sparse models at 80 percent or more of peak FLOPs. SambaNova closed the first tranche of a $1 billion Series F at an $11 billion post-money valuation led by General Atlantic, with Seligman Ventures, T. Rowe Price and Capital Group as significant investors; it announced JPMorganChase as an inference infrastructure partner deploying its SN40 and SN50 systems. Positron AI, which builds inference systems on commodity LPDDR5X rather than HBM, took up to $875 million across a $375 million Series C and an up-to-$500 million Series C-1.
With the data center buildout driving demand for memory, investors also rewarded startups offering alternatives to capacity-limited HBM: chips that hold models in on-chip SRAM, designs built on commodity DRAM, and memory stacked directly onto compute. Data movement remained the most active area of the quarter, with interconnect, networking and photonics startups raising across approaches from co-packaged optics to millimeter-wave RF over plastic waveguides. Networking names were prominent: Xsight Labs raised over $300 million for DPUs and 800G Ethernet switching, Celero Communications took $275 million in Series C for coherent DSPs for AI optical interconnects, and Cornelis Communications raised $205 million for scale-out AI networking as an Intel HPC spin-out.