Oriental Compute Core, a Shanghai-based AI chip startup, launched its DF1000 software-defined near-memory 3D chip in Shanghai on July 13.

The company said the chip is designed to address AI computing bottlenecks in memory, bandwidth, and power consumption.

Oriental Compute Core said DF1000 delivers 520 TFLOPS at BF16, uses dynamic random-access memory (DRAM)-logic wafer-level hybrid-bonded 3D packaging, and reaches 6.4 TB/s of memory bandwidth, higher than Nvidia H200’s 4.8 TB/s.

Hybrid bonding is a 3D packaging method that directly bonds copper pads between stacked chips, allowing denser vertical interconnects and lower resistance than microbump-based designs.

The company also introduced server and cluster products based on DF1000 and said its software stack is compatible with mainstream deep learning frameworks.

Oriental Compute Core said it plans to release DF2000 in the fourth quarter of 2026 and DF3000 in the fourth quarter of 2027. It also said an A+ round completed in April 2026 gave it a post-money valuation of about 12.3 billion yuan (US$1.82 billion).

The launch comes as China’s AI chip market grows more crowded, with domestic graphics processing unit makers including Biren Technology, Iluvatar CoreX, Moore Threads, and MetaX reporting strong revenue growth as demand for domestic computing power rises.

🔗 Source: Security Times