Isn’t it the worst that Huang Renxun is gone? Gao Canming explodes Huida's AI chip hegemony may collapse

Gao Canming warned that Huida's chip hegemony may collapse. (Reuters)

[Financial Channel/Comprehensive Report] The resignation of Huang Renxun, the founder of Huida, as CEO is probably not the biggest crisis facing Huida. Well-known analyst Tim Culpan warned in his latest comments that the real threat to Huida is not the loss of Huang Renxun, but the erosion of its AI chip monopoly. As technology giants such as Amazon, Microsoft, OpenAI and Anthropic actively develop their own chips, the competitive advantages that Huida was proud of in the past may gradually collapse. He even bluntly said that even without Huida, the AI ​​industry can continue to develop.

Tim Culpan published a comment in Culpium, an independent media he founded, titled "Huang Renxun is the rudder, but Huida's engine has problems." He pointed out that when Huang Renxun made a public appearance in Taipei a few months ago, he was still wearing his signature leather jacket, like a rock star, confidently promoting the concept of "the more you buy, the more you save." At that time, global companies were rushing to buy Huida chips, and the AI ​​craze seemed to be inseparable from the founder.

However, with Huang Renxun leaving the CEO position, the outside world began to question Huida's ability to maintain its past growth momentum. Gao Canming admitted that he originally underestimated the possibility of Huang Renxun leaving. He once accepted an interview with CNN and believed that Huida's board of directors should face up to the "key person risk", but it is expected that the board of directors may deal with it by strengthening supervision and restricting Huang's power rather than letting him leave.

Gao Canming described Huang Renxun's departure as a ship suddenly losing its rudder, but he emphasized that Huida also faced a more cruel reality: even if there is a rudder, if there is a problem with the engine that propels the ship forward, it will not be able to continue sailing at high speed.

Gao Canming pointed out that Huida has been the biggest promoter and beneficiary of the AI ​​revolution in the past few years, holding more than 90% of the AI ​​computing market share. However, this near-monopoly advantage is facing challenges. Large technology companies such as Amazon, Microsoft, OpenAI and Anthropic are actively developing their own chips, and AMD is gradually catching up, making Huida no longer the only choice in the market.

He believes that many people still believe that Huida's GPU technology is leading and it will be difficult for competitors to shake its position. This judgment may still be correct in some fields, but it does not mean that Huida will always maintain its advantage, nor does it mean that all AI applications must use Huida chips.

What is truly noteworthy is the rise of custom chip ASICs. Gao Canming emphasized that every AI server system that uses self-developed ASIC chips by technology giants means that there is one less system that may use Huida GPUs. In the past, the AI ​​market expanded rapidly, and all players had the opportunity to share the growth dividends. However, even if the overall market continues to grow in the future, Huida's proportion may still gradually shrink.

In addition to hardware competition, Huida's long-established CUDA software ecosystem is also facing challenges. In the past, CUDA made developers highly dependent on the Huida platform, increasing the cost for customers to switch chip suppliers. However, as large technology companies invest more resources in developing alternatives and some customers have begun to try to reduce their dependence on CUDA, Huida's most important competitive moat may be gradually eroded.

Gao Canming further pointed out that the AI ​​industry is facing an important business model change: from investing heavily in training models to gradually turning to generating revenue through AI inference.

He compared AI training to building a factory, and inference to a delivery vehicle that delivers products to customers. Training models requires huge computing resources and capital expenditures, but what really allows AI companies to continue to make profits is the inference stage of providing services after the model completes training.

If AI companies slow down training, they may directly reduce chip purchases, or they may shift more budgets to inference. Gao Canming believes that both situations may be detrimental to Huida, especially that many ASIC chips being developed by technology giants are designed for inference workloads.

Since only a small improvement in inference efficiency can significantly increase the profitability of AI services, technology companies have strong incentives to design their own chips rather than continue to purchase Huida's standardized GPUs.

Gao Canming even cited Apple's abandonment of Intel and its shift to self-developed chips as an example to explain why technology giants want to master chip design capabilities. For large enterprises, self-developed chips can not only reduce dependence on suppliers, but also help optimize performance and cost for their own product needs.

Taking Anthropic as an example, Gao Canming pointed out that Broadcom will be its important infrastructure partner in the next few years. In addition to its chip design capabilities, Broadcom also assists large cloud companies in developing customized ASICs. Anthropic's other data center partners, including Amazon, Google and Microsoft, are also moving toward reducing chip dependence.

In comparison, although emerging cloud computing players such as Nscale and Lambda still rely heavily on Huida, their proportion in Anthropic's overall infrastructure layout is relatively limited. This means that Huida’s largest customer base is gradually developing the ability to provide AI computing power on its own.

Gao Canming also mentioned that Huida has been actively deploying new fields such as robotics, autonomous driving and digital twins in recent years, trying to find growth momentum beyond AI computing, but there is currently not enough evidence to show that these businesses can become the main revenue pillar in the next few years.

He emphasized that the AI ​​revolution does exist and will be one of the most important technological changes of this generation, but this does not mean that Huida can always maintain the amazing growth rate of the past few years. In the future, more economic benefits created by the AI ​​industry may gradually flow to large cloud service providers and AI model development companies.

At the end of the article, Gao Canming raised a thought-provoking question: Can Huida continue to develop without Huang Renxun? This is certainly worthy of attention, but more importantly, can the AI ​​industry continue to develop without Huida? His answer was: "Yes."