Ranked by newest founded · 25 of 40 on the roster
Positron AI builds energy-efficient LLM inference appliances (Atlas) as a US-made, memory-bandwidth-optimized alternative to GPUs.
EnCharge AI develops in-memory compute chips for ultra-efficient AI inference at the edge and in data centers.
Etched builds application-specific chips hardwired to run transformer architectures with extreme performance and efficiency.
MatX, founded by ex-Google TPU engineers, designs chips optimized specifically for large language models, with first silicon due in 2027.
Axelera AI is a European edge-AI startup whose Metis chips use in-memory computing to run computer-vision inference at low power and cost.
Rebellions is a Korean AI-chip unicorn behind the ATOM and REBEL inference accelerators, formed by its 2024 merger with SK Telecom's Sapeon.
d-Matrix builds in-memory compute chips optimized for efficient transformer inference at scale in data centers.
SiMa.ai builds a software-centric edge-AI 'MLSoC' that runs computer-vision and ML workloads at low power for the embedded edge.
Astera Labs builds high-speed connectivity silicon — retimers, CXL controllers, and fabric switches — for AI and cloud data centers.
FuriosaAI is a Korean startup building energy-efficient AI inference chips (RNGD); it rebuffed an ~$800M Meta buyout to stay independent.
Hailo builds edge AI processors optimized for computer vision and neural network inference in automotive and industrial devices.
Lightmatter builds photonic processors and optical interconnects to accelerate AI compute and reduce data-center power consumption.
SambaNova builds dataflow AI accelerators and full-stack software for enterprise AI training and inference at scale.
Cerebras builds wafer-scale AI chips with millions of cores optimized for large-scale model training and inference.
Groq builds Language Processing Units (LPUs) for ultra-low-latency LLM inference — the tech behind Nvidia's reported ~$20B Dec 2025 IP deal.
Tenstorrent designs open-source RISC-V AI processors and licenses chiplet IP for scalable AI compute.
Ayar Labs develops chiplet-based optical I/O solutions that replace electrical interconnects in AI and HPC systems.
SiFive licenses RISC-V processor IP and builds custom silicon solutions for AI, automotive, and data-center customers.
Navitas manufactures gallium nitride (GaN) power semiconductors that increase efficiency in data-center power supplies and AI systems.
Mythic develops analog compute-in-memory chips for edge AI inference, targeting industrial, defense, and automotive applications.
GlobalFoundries operates fabs focused on specialty and mature-node semiconductors for automotive, IoT, and RF applications.
Ambarella (NASDAQ: AMBA) designs low-power edge-AI vision SoCs — CV3-AD and N1 — for automotive ADAS, security cameras, and robotics.
Marvell designs custom AI silicon for hyperscalers and supplies data-center networking, storage, and interconnect chips.
Nvidia designs GPUs and AI accelerators that power the vast majority of frontier model training and inference workloads.
Arm licenses CPU architectures and IP used in nearly every mobile device and increasingly in data-center and AI chips.
