Jensen Huang, the CEO of NVIDIA, has been comfortably ensconced in the spotlight for several weeks now. And is not for less. His company just passed a market capitalization of more than a trillion dollars thanks essentially to the very high demand for its GPUs for data centers. Currently, close to 80% of the chip market for artificial intelligence belongs to him, and nothing seems to indicate that his dominance will be altered in the medium term.
NVIDIA engineers are working on the microarchitecture of their next chips for graphics and artificial intelligence applications, and a fundamental ingredient of their recipe is the lithography with which they will be manufactured. This company works with both TSMC and Samsung, which currently have the most advanced photolithographic nodes, and it is very likely that some of their GPUs will be made by Intel in the future.
This information does not come from any leak. In fact, Jensen Huang himself has confirmed it during one of the conferences that he has just given at Computex, which is being held these days in Taipei (Taiwan). The NVIDIA leader has acknowledged that recently have received the first chips tests made by Intel using its next-generation lithography, and according to Huang, the result is good.
NVIDIA wants to diversify the origin of its chips. Intel wants the best lithography
The very probable entry of Intel in the short term into the orbit of NVIDIA does not have to imply at all that TSMC will stop manufacturing its graphics processors. According to the consultancy AMR (Allied Market Research), in 2031 the market for chips for artificial intelligence applications will have a turnover of more than 263,000 million dollars. It is a real outrage, especially if we bear in mind that in 2021 his business amounted to just over 11,000 million dollars.
NVIDIA is interested in simultaneously having the semiconductor manufacturing capacity of TSMC, Intel and Samsung
Over the next few years, NVIDIA will have to cope with a very high demand for artificial intelligence chips, so it is interested in simultaneously having the semiconductor manufacturing capacity of TSMC, Intel and Samsung. This strategy will allow you respond to demand with more agility, and, as a bonus, it will also help protect you if one of the companies responsible for manufacturing your chips has problems with their integration technology.
What we cannot foresee right now is which GPU will be manufactured by each of these companies. It is possible, although it is only a conjecture, that TSMC will produce the chips of the future GeForce RTX 50 family, and that the GPUs for artificial intelligence will be commissioned simultaneously by TSMC, Samsung and Intel. Any other organization is also possible. In all probability NVIDIA will assign its orders depending on the development of the lithographic nodes of its semiconductor suppliers.
TSMC and Samsung are already making chips in their 3nm nodes, but Intel is currently lagging much further behind. Pat Gelsinger, the CEO of the latter company, assured during an interview with The Wall Street Journal that his medium-term strategy in the field of the chip industry involves having the best transistors and the world’s most advanced integration technology by 2025. Clearly this statement is a very strong statement of intent.
Intel has confirmed that during the second half of next year it plans to have the 18A (1.8 nm) lithographic node ready.
Intel expects to have the Intel 3 node ready to start manufacturing during the second half of this year, as well as to start chip production on the Intel 20A (2 nm) node during the first half of 2024. And, what if possible is More surprising, it has also confirmed that during the second half of next year it plans to have the 18A (1.8 nm) lithographic node ready. In any case, the most important thing is that we have reason to take his intentions very seriously.
The statements just made by Jensen Huang suggest that the development of Intel’s next lithograph is on the right track, but we also have something else. At the beginning of last March Wang Rui, the president of Intel’s subsidiary in China, stated that her engineers have already completed the development of their 2 and 1.8 nm integration technologies. This does not mean that Intel can already start manufacturing chips with these lithographs, but it reflects that the itinerary that has been marked is credible. We’ll see what happens eventually, but everything seems to indicate that in 2024 the competition between the main chipmakers is going to be unusually fierce.
Cover Image: Intel
More information: Tom’s Hardware
In Xataka: TSMC negotiates hard: the German Government will have to accept its conditions to host its new factory