Citrini says coding AI is weakening CUDA’s moat, backs AMD and turns bearish on Nvidia

News Editor

2026-07-23 16:56:56

Citrini analyst Jukan, citing recent remarks from DeepSeek founder Liang Wenfeng, argues that AI-powered code generation and higher-level programming tools such as TileLang are cutting into the barriers that once protected Nvidia’s CUDA ecosystem. The argument is not that Nvidia hardware has already been displaced—DeepSeek still used Nvidia GPUs to train its V3 model—but that the software lock-in around CUDA is becoming less absolute as model builders rely more on self-developed compilers and newer programming layers. According to the view relayed by Jukan, DeepSeek has already reduced its dependence on Nvidia’s software stack through its own compiler and the TileLang environment. Liang had also said 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, leaving capacity as the main remaining bottleneck. He described the China-U.S. chip gap as roughly 4x in hardware efficiency and about two years in time lag, and said DeepSeek is working closely with Huawei and expects to receive about 16,000 Huawei AI chips. Jukan framed that as a sign that CUDA’s moat is eroding, a negative for Nvidia and a possible tailwind for AMD’s ROCm software stack.

NvidiaAMDCUDAROCmDeepSeekHuaweiAI codingSemiconductors

BlockBeats reported on July 24 that Citrini analyst Jukan, citing recent core views from DeepSeek founder Liang Wenfeng, said AI-driven code generation and higher-level programming languages such as TileLang are rapidly lowering the entry barriers around the CUDA ecosystem.

In Jukan’s telling, DeepSeek still used Nvidia GPUs to train its V3 model, but the company has already cut its reliance on Nvidia’s software ecosystem through a self-developed compiler and the TileLang environment.

Porting to Huawei chips is central to the argument

Liang had previously said that if TileLang and the DeepSeek compiler were ported to Huawei chips, China’s chip ecosystem problem could be largely solved within about a year, with manufacturing capacity left as the main bottleneck.

He also quantified the gap between Chinese and U.S. chips at roughly 4x in hardware efficiency and about two years in time lag. Liang added that DeepSeek is working closely with Huawei and is expected to receive about 16,000 Huawei AI chips, while Huawei’s 950 SuperNode can take on workloads handled by GB200 and GB300.

Jukan says CUDA’s moat is nearing its end

Jukan described the shift as a sign that “CUDA’s moat is ending,” and said he is highly bearish on Nvidia on that basis.

He added that the same logic is one reason he is bullish on AMD. Progress in coding AI, in his view, should also speed up the development of the ROCm ecosystem and help narrow the gap with CUDA.

Jukan also noted that AMD, when it recently invested in Anthropic, said it would actively use Claude Code in chip design and software engineering.

Software changes are being read as pressure on Nvidia’s edge

Taken together, the views cited by Jukan suggest that advances in AI coding tools are eating into Nvidia’s competitive barriers from the software side. They also point to the possibility that China’s chip ecosystem issues could be resolved faster as code generation improves. In that framework, AMD could benefit as ROCm gains ground, while CUDA—the software moat that has long helped Nvidia retain developer loyalty—faces pressure from both Chinese chip adaptation efforts and AMD’s software progress. On Liang’s view, catching up on hardware efficiency and capacity is ultimately a matter of time.

This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan.

500

Disclaimer:

The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.

Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.