Chinese chip IPOs hit high demand, oversubscribed nearly 3,000x

Two Chinese chipmakers saw heavy demand from retail investors ahead of their IPOs, after a strong trading debut by Moore Threads Technology.

MetaX Integrated Circuits Shanghai, which manufactures graphics processing unit chips, had its retail tranche oversubscribed 2,986 times, while Beijing Onmicro Electronics, a radio frequency chipmaker, was oversubscribed 2,899 times.

Investor interest increased after Moore Threads, viewed as a potential local competitor to Nvidia in AI chips, surged 425% on its first trading day.

Moore Threads’ own IPO was oversubscribed by about 2,750 times in the retail segment.

MetaX is seeking to raise US$585.8 million in its Shanghai listing at an offer price of 104.66 yuan per share, giving it a price-to-sales ratio of 56.4 times, below the peer average of 127.4 times for 2024.

Tight IPO approvals and recent risk aversion in China’s secondary market have contributed to strong demand for new listings this year.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

MetaX’s 56x price-to-sales (P/S) multiple hints at IPO hype over proven results

The article omits revenue and margins. It also leaves out customer mix and whether MetaX ships at volume or still develops products. A 56x P/S looks lifted by retail investors after Moore Threads’ 425% debut rather than by clear operating data.
The “below peer average” claim of 56.4x vs 127.4x for 2024 lacks context on scale, profits or position. Many Chinese GPU startups use external contract chip manufacturers (foundries), and U.S. export rules curb access to leading-edge manufacturing processes (advanced nodes), which can limit competitiveness.
Moore Threads’ MUSA stack and software development kit (SDK) have made progress 1 2 yet they trail Nvidia CUDA on maturity and support 1. There is no clarity on MetaX’s stack or developer uptake. The price may lean on hopes of state support for domestic chips over clear technical edge or adoption.

Cloud providers can build integration layers for domestic GPU workloads

Moore Threads’ torch_musa (a PyTorch compatibility package for the MUSA stack) lets PyTorch users swap device tags from “cuda” to “musa” 3. That suggests thin layers can bridge proprietary stacks. Chinese clouds can provide wrappers for MetaX and other local GPUs to offer mixed training and inference.
MUSIFY (which helps convert CUDA code to run on MUSA) likely translates CUDA code at runtime 1. That gap creates room for stronger translators, debuggers and profilers. The SDK supports Intel and domestic CPUs 2, so vendors can tune cross-platform systems that pair CPUs with GPUs.