Home / Technology / Etched Challenges Nvidia With New AI Inference Chips
Etched Challenges Nvidia With New AI Inference Chips
2 Aug
Summary
- Etched secured $300 million, reaching a $10.3 billion valuation.
- Startup focuses on custom chips for AI inference, not general GPUs.
- New architecture uses custom chips, shared memory, and lower power.
Etched, an AI chip startup, has reached a significant milestone with a $10.3 billion valuation, following a recent $300 million funding round. This latest investment, which includes SK Hynix, will accelerate the production of their specialized inference systems. The company differentiates itself by focusing exclusively on AI inference, eschewing general-purpose GPUs.
Etched contends that conventional GPUs are often power-inefficient and lack adequate memory for demanding AI inference tasks. Their innovative approach involves custom-designed chips, Low Voltage Inference (LVI) technology, and Cluster Scale Memory (CSM). This architecture is engineered to provide greater FLOPs density and faster memory access than current GPU solutions.
The company has attracted substantial investment, with total funding now nearing $925.4 million across several rounds since 2023. Etched operates an 80,000 sq ft facility near Milpitas and has manufacturing operations in Taiwan. They report customer demand exceeding $1 billion and employ over 400 individuals, positioning them for global expansion.
Their systems are designed to support a range of AI models, including large language models and newer architectures like Mamba. This strategic focus on inference-specific hardware aims to set a new standard in AI processing efficiency and performance.
