China vs. USA – Chips for Free?

Buying Huawei makes no sense unless they are almost free.

• Huawei is having another go at selling its AI chips overseas but as it is faced with an increasing number of competitors who are all substantially cheaper to deploy, Huawei will have to virtually give the chips away.
• Huawei is bidding on an Egyptian government tender to supply around 1400 Ascend (950 presumably) chips for a training cluster and another 600 for inference.
• This is a very small deployment and looks like a trial of the system which, if successful, could lead to a much larger order.
• As far as I am aware, this would be the first real successful attempt by Huawei to export chips, as a bid made to Malaysia in 2025 does not seem to have gone anywhere.
• The White House is reportedly concerned about a successful deployment by Huawei and is encouraging US companies to cobble together a competitive bid.
• This should not be very difficult, as Huawei’s chips are so uncompetitive that Huawei will have to price them at almost nothing to achieve a sale.
• Furthermore, they are so expensive to operate that any owner concerned with the cost of electricity would think at least twice before purchasing them.
• This opinion is based on Huawei’s own data from September last year when it published the specifications and roadmap for the Ascend series and RFM’s research into the economics of AI data centres (see here and here).
• Here, RFM concluded that Huawei’s 2026 Ascend chip, compared to Blackwell, would cost more than 3x to build and consume 9x more power per unit of compute (FLOPs) produced.
• Furthermore, based on Huawei’s roadmap, this gap is going to widen in the coming years as Huawei stays stuck at 7nm and everyone else continues forward.
• Hence, it should not be very difficult for Nvidia or anyone else to win this business, but as usual, the reality is much more complicated.
• I suspect that everyone is so busy with the massive data centre roll-out that no one is paying much attention to Egypt, which, combined with the supply constraints, means that there are not many chips available.
• President Xi is currently making a state visit to Egypt as part of a regional tour, and so I expect that this will be used to push Huawei’s case.
• The net result is that I don’t think that it will be very difficult to win Egypt’s business away from Huawei, but there are other factors that may allow Huawei to get its foot in the door.
• Even if it does win a trial order, I still don’t think that Huawei is going to get very far, as once Egypt realises how much the chips cost to run, it is unlikely to take this pilot order to scale and will go somewhere else.
• This is the main battleground where the ideological struggle is being fought through the technology sector, as countries like Egypt and Malaysia will have to choose which way to go.
• At the moment, the USA has a very large advantage because, for once its products are both technically superior and far cheaper to buy and operate, even with Nvidia’s 70%+ gross margins.
• This is the main reason why UAE, Saudi Arabia and to some degree Qatar have pivoted towards the US and away from China.
• Hence, I would not be surprised to see Egypt buy Western hardware and then use it to run Chinese open source models which are very competitive in performance and cost-effective to run.
• China is doing extremely well further up the technology stack, but when it comes to hardware, its disadvantage, which is likely to remain for the foreseeable future, hands the advantage to the USA and its allies in the AI race.