AI inference and training semiconductors with integrated thermal management
Vastai Technologies designs vision and AI compute semiconductors built on ARM and FPGA architectures, using PyTorch, DeepSpeed, and Megatron-LM in their development pipeline. The company is actively working on large-model training systems and liquid cooling solutions — a signal that they're scaling toward high-density AI inference and training workloads where thermal constraints are the limiting factor. Engineering-heavy hiring (8 roles) concentrated at senior and mid levels suggests active chip tape-outs and production ramp.
Vastai Technologies develops high-performance semiconductors for vision processing and AI compute, targeting real-time inference at edge and data-center scales. Founded in 2018 and based in Shanghai, the company operates across chip design, verification, thermal engineering, and production readiness. Their current focus spans AI solution deployment for industrial applications, large-model training acceleration, and reliability engineering — with active projects in chip verification, mass production hardware design, and FCT test environment setup. The company is privately held with 501–1,000 employees and currently hiring across engineering and product roles.
Vastai uses ARM and FPGA as primary compute architectures, paired with OpenGL/Vulkan graphics APIs and PCIe/Ethernet interconnects. The stack includes Synopsys PrimePower for power analysis and COMSOL/ANSYS Icepak for thermal simulation.
Vastai's development stack includes PyTorch, DeepSpeed, Megatron-LM, LLaMA, ChatGLM, and GPT frameworks — indicating focus on large-language-model training and inference optimization on their silicon.
Other companies in the same industry, closest in size