ChainThink reports that on July 24, Citrini analyst Jukan cited recent remarks from DeepSeek founder Liang Wenhong, stating that AI-driven code generation and advanced programming languages like TileLang are lowering the barriers to entry for the CUDA ecosystem.
Although DeepSeek used NVIDIA GPUs to train the V3 model, it has reduced its dependence on NVIDIA's software ecosystem through its proprietary compiler and TileLang environment.
Liang Wenfeng previously estimated that if TileLang and the DeepSeek compiler were ported to Huawei chips, China’s chip ecosystem issues could be largely resolved within about a year, with the main bottleneck shifting to production capacity.
He also stated that the chip gap between China and the U.S. includes a hardware efficiency difference of approximately four times and a time lag of about two years.
Jukan stated that DeepSeek is closely collaborating with Huawei and is expected to receive approximately 16,000 Huawei AI chips, with the Huawei 950 SuperNode capable of replacing workloads handled by GB200/GB300.
It summarizes this trend as "the CUDA moat is coming to an end" and is therefore bearish on NVIDIA.
Jukan also noted that this is one of the reasons he is bullish on AMD, as advancements in AI coding will accelerate the growth of the ROCm ecosystem and narrow the gap with CUDA.
When AMD recently invested in Anthropic, it stated that it would actively use Claude Code for chip design and software engineering.