🧠 Food for thought

Implications, context, and why it matters.

Dnotitia’s funding backs a hardware push to tackle AI’s memory problem

• Dnotitia’s main product is a custom chip called a Vector Data Processing Unit (VDPU), built to speed vector database searches for AI 1.
• The chip tackles a growing bottleneck in AI, the slow and expensive work of pulling data for Retrieval-Augmented Generation (RAG), a method that lets AI systems use outside information when answering questions 2.
• The company plans to offer the technology as a cloud service, a workstation, and licensable semiconductor intellectual property (IP) 3, 1. That IP includes chip designs other semiconductor companies can use in their own products 3, 1.
• Dnotitia has finished the VDPU design and tape-out, the step where a finished chip design goes to manufacturing 4. It is producing first silicon on Taiwan Semiconductor Manufacturing Co.’s (TSMC’s) 12-nanometer process, with full commercialization set for early 2027 4.

The AI hardware stack is splitting beyond the GPU

• Dnotitia’s funding marks a phase in AI where the bottleneck is moving from raw computing power to faster data retrieval 5. That shift creates demand for specialized accelerators that work alongside general-purpose graphics processing units (GPUs) 5.
• The company is building for the infrastructure layer that helps enterprises use AI in practice 5. Support for open-source databases such as Milvus and FAISS ties its bet to growth in the wider vector database software market 5.
• If companies like Dnotitia do well, South Korean chip intellectual property could gain ground in a hardware segment now led by U.S. and Chinese firms 4.