Nvidia vs Huawei – No Contest

Huawei has no chance but will gain domestic share anyway

• When it comes to performance, Huawei’s chips have no chance of beating Nvidia in a straight fight, but with a focus on inference and the thumb of the CCP on the scale, there is little doubt that Huawei will gain share at home.
• Huawei is focusing on dominating the domestic market for AI chips and is wisely focusing mostly on inference as the hill to climb in training is much higher.
• Together with the strong preference of the Chinese state for homegrown products, Huawei has a much greater chance than anyone else of overcoming the barriers to entry that Nvidia has so effectively erected over the last 20 years.
• This is because the key to Nvidia is not its GPU processors but its CUDA development platform as well as the fact that it is always at least one generation ahead of its competitors.
• This is how it can truthfully claim that it is cheaper to train your model on Nvidia silicon even though its high prices allow it to earn gross margins of 70% or more.
• However, in China, the situation is completely different.
  • First, availability: where Nvidia’s latest chips are not available and Huawei is competing against the H20 chip from Nvidia whose performance and economics are a far cry from the cutting edge.
  • However, even against this, the Huawei Ascend chips are having difficulty because of developer preference for the CUDA development platform which is only available for Nvidia silicon.
  • However, this is greatly helped by the Chinese state.
  • Second, domestic preference: where in China it is very clear to everyone that business carries on with the permission of the CCP which can be withdrawn or altered at any time and with no warning.
  • Outside of China, CUDA is a major barrier to entry but with the state clearly preferring that everyone uses homegrown products, there is a clear incentive for developers to switch away from Nvidia.
• Furthermore, Huawei is also focusing on inference where Nvidia is also very strong, but it does not have the same kind of power that it has in training.
• This is a wise move by Huawei as RFM Research has long concluded that as AI rolls out across the world’s economies, the market for AI chips for inference could be much larger than for training.
• Here, Huawei will also have the support of the Chinese state and so I think it is very likely that Chinese companies who are rolling out AI services will elect to run their inference on Huawei.
• The downside here is that Chinese AI will be more expensive to train and run both in terms of buying silicon chips and running data centres which is why the focus in China is on doing more with less.
• I think that this strategy is going to bear fruit as it is already clear that with a focus on efficiency, smaller models can be made to perform better than larger ones which will also be less costly to run and maintain.
• However, outside of China, Huawei has very little chance of gaining share against Nvidia or any of its competitors.
• This is because it is competing against more advanced silicon which is cheaper to purchase and run relative to the delivered performance.
• Many countries are also wary of Huawei given its history as well as its designation as a company related to the Chinese military by many Western countries.
• However, China is a very large market meaning that there is plenty of opportunity for Huawei at home and so I don’t see its inability to gain traction overseas as a huge problem.
• This is yet another sign of the increasing decoupling of China from the rest of the world and we are likely to see different technology standards also start to emerge.
• This will result in separate and incompatible networks and a Balkanisation of the Internet.
• This is good for no one as growth of the global technology sector in the long term will be lower than it otherwise would have been.