In over two hours of conversation on Lex Fridman’s podcast, Jensen Huang repeatedly focused on the characteristics of the semiconductor supply chain and its main bottlenecks, which affect the speed of development of artificial intelligence infrastructures, even in the face of demand that remains insatiable for now.
Huang explained that his management of the supply chain is based on constant and transparent communication with all partners, both upstream and downstream of the production chain. For example, he spends a lot of time talking with partners like ASML, and obviously TSMC, to share growth trajectories and NVIDIA’s future needs, explaining market dynamics so they can properly plan capital investments and thus enable the sector’s acceleration. In response to repeated questions from Fridman, the NVIDIA leader said he has no concerns about ASML’s ability to scale production and provide the necessary tools for NVIDIA’s system growth, and took the opportunity to praise the capabilities of the technology giant from Veldhoven.
While ASML continues to break sales and order records (also thanks to the latest agreement with Sk Hynix on memory), startups and initiatives seeking to challenge its de facto monopoly have grown over the past year.
Among them is Substrate, founded by Briton James Proud and supported among others by Peter Thiel and the CIA’s venture capital vehicle. Another entity that has developed in recent months is xLight, founded in Palo Alto, which does not aim to replace ASML’s machines but to improve a critical component, the light source, and is supported by Pat Gelsinger, former CEO of Intel.
Another startup in this wave is European: the Norwegian Lace Lithography, led by Danish-Norwegian physicist Bodil Holst, which wants to use beams of helium atoms instead of light.
These projects are still at an embryonic stage and will hardly be able to compete with the power of ASML’s supply chain, founded on very tight integration with the capabilities of suppliers and customers. ASML, which I have tried several times to describe in detail (for example here on Substack), has decades of know-how in optics, precision mechanics, chemistry, metrology, maintenance, and very advanced software in its system. The giant from Veldhoven does not produce a prototype but an industrial machine of enormous complexity, and in this field no approximation is allowed nor will be allowed.
In any case, these initiatives represent investment theses that suggest a political reading: despite ASML’s supply chain having essential nodes in the United States (due to the company’s history and the US research programs I recalled in my 2022 book “The Dominance of the 21st Century”), it is in Washington’s interest to support some difficult bets on diversification.
It remains to be seen whether these challengers will survive the current moment of political attention, or if their funding window will close before they manage to approach a real challenge to ASML. After all, the specialized press in the Netherlands had good reason to recall the case of Lyncean Technologies, a Californian startup that proposed itself as an alternative to ASML and went bankrupt in 2022.
Certainly, Chinese initiatives will survive, because they are part of the existential effort of the People’s Republic of China to create a less vulnerable semiconductor supply chain in competition with the United States. The microelectronic supply chain in China continues to grow, but even Chinese announcements about alternatives to ASML (such as Shenzhen’s “Manhattan Project” from the Reuters investigation) seem, for now, far from concrete production capacity.
