Huawei’s announcement of a new chip design philosophy for AI marks a further step towards Beijing’s decoupling from Western technologies, which risk remaining trapped by their structural limitations by Andrea Monti – Initially published in Italian by Formiche.net
Huawei’s announcement of a new chip design philosophy for AI marks a further step towards Beijing’s decoupling from Western technologies, which risk remaining trapped by their structural limitations.Western media are making a big fuss over Nvidia’s recent announcement regarding the forthcoming availability of RTX Spark, a native superchip installed on next-generation personal computers and laptops to run AI models – and in particular LLMs, those that interact with and via language – ‘locally’, i.e. without necessarily relying on external services.
Nvidia’s announcement is aimed at expanding its presence – already very strong in the sector of high-performance processors for Big Tech – into the market for ‘off-the-shelf’ hardware production as well. This is how Huang’s company would manage to further expand its control over substantial parts of the AI supply chain, attracting not only those who build the ecosystem but also those who operate within it.
Huawei responds with an alternative to brute force
By contrast, Huawei’s virtually simultaneous announcement went largely unnoticed: on 25 May in Shanghai, at the IEEE International Symposium on Circuits and Systems, it presented the Tau Scaling Law, a new approach to improving transistor density and the overall performance of electronic systems that promises to significantly narrow the gap with US technologies.
US restrictions and the unexpected effect of decoupling
As is well known, the US administration uses export bans and restrictions on certain products and technologies to curb China’s progress in the field of artificial intelligence.
For its part, China has made a virtue of necessity and has begun to develop the entire AI supply chain in-house, from chips to processors to AI models, transforming (or seeking to transform) its weakness—namely, the lack of access to the most modern technologies—into a strength. In particular, it has focused its strategy on two main areas: hardware optimisation (hence the Tau Scaling Law), and the performance and energy optimisation of models and development environments. All with the aim of maximising the integration of individual components.
In other words, whilst the US pursues development through increased computing power, energy consumption and the number of parameters used by models, China designs its infrastructure by reducing resources—or, rather, by ensuring it uses only the bare minimum and maximising the integration between hardware and software.
China is not playing catch-up, but running in a different race
Setting aside the outcomes of their respective marketing strategies – the market will decide whether RTX Spark is a success, just as actual implementation will demonstrate whether the Tau Scaling Law actually works – the approaches of the two tech giants suggest some reflection on the current state of AI geopolitics.
The US and China are playing two games: one for technological control of the AI production chain and the other for strategic and geopolitical supremacy.
China is the challenger and is objectively at a disadvantage if it agrees to play on the opponent’s turf, with a timetable and rules it has not defined and must therefore accept. Hence the decision to shift the contest to a terrain under its own exclusive control. This is the significance—not only technological but also geopolitical—of Beijing’s choice.
The American limitation: data centres, energy and industrial gigantism
Obviously, the issue of frugal AI or, in any case, operational efficiency is not unknown to the US approach. However, it is conditioned by industrial, economic and financial choices that drive technological gigantism due to what could be the data centre bubble and related infrastructure.
US AI requires enormous amounts of energy, computing power and physical space; their growth cannot continue indefinitely, but – and this is the problem – nor can it be halted. The investments made by the financial market are too substantial and there are too many economic interests at stake. Furthermore, the long lead times for building this infrastructure are hardly compatible with the need to maintain a strategic advantage over China.
Although the outlook is objectively uncertain, it would be simplistic to claim that, in the long term, the US will lose the AI race. What is clear, however, is that the so-called ‘philosophical’ choices made at the outset of the competition have irrevocably shaped subsequent developments.
Betting on gigantism in the belief that there would be no competitors or that it would be possible to keep them at bay was an ill-advised choice, which is now forcing the US to grapple with strategic gambles that have not proved entirely successful.
The lesson for Europe: less regulation, more industry, more autonomy
This situation traps the EU between Scylla and Charybdis.
Its Atlanticist geopolitical stance precludes the Union from forming an axis with Beijing by accessing its AI technologies (some of which are even open source and open weight), even if only as part of a strategy to reduce dependence on or American technological influence.
At the same time, the US does not seem particularly willing to cede the use of its infrastructure to European partners. However vast they may be, US technological resources are already entirely absorbed by Big Tech; moreover, even if there were excess capacity, the US would be unlikely to cede it to Europe, given the risk of creating a third AI hub.
All this provides very clear indications as to the direction the European Union should take to secure a role in the AI market. It could, in fact, capitalise on this, but only if it decided to focus genuinely on the creation of a third hub, rather than getting lost in the regulatory labyrinths it has built and from which it can no longer escape, even if it wanted to.