Friends, have you paid attention to the hot news of "Nvidia spent $45 billion to buy chips" recently? As a self-media practitioner, I am very concerned about this matter!
We all know that NVIDIA, as the world's top GPU manufacturer, has been sparing no effort to improve the performance and quality of its products. And this time they plan to spend a huge amount of money on HBM3e memory and advanced packaging technology, apparently to ensure that flagship GPU products can be the first to seize the market in 2024.
You say, will the sky-high price of $45 billion hollow out Nvidia? Moreover, will this huge investment trigger a sense of crisis among domestic GPU manufacturers, thereby accelerating the localization process? Honestly, I'm curious and looking forward to it! So, let's talk about it today.
First, let's take a look at what the $45 billion is. According to a report by Tomshardware, a well-known hardware website, in order to ensure the supply of GH200 and H200 series GPUs in 2024, Nvidia has spent $1.3 billion to book part of Micron and SK hynix's HBM3e memory production capacity.
However, the figure of $1.3 billion seems a bit petty, because according to the current supply and demand situation, the budget is obviously not enough to buy. I estimate that NVIDIA's real HBM memory procurement cost should be around $6.8 billion, and with the cost of CoWoS advanced packaging, the entire investment may reach $45 billion.
You may ask, why is Nvidia investing so much money? In fact, the main reason is that they want to ensure that their GPU products can be launched before their competitors in the first half of 2024, so as to seize a large wave of market share.
As we all know, the time window for the first half of the year is extremely critical. Once this opportunity is missed, the product will have to wait until the second half of the year to officially hit the market, by which time it will already lag behind the competition. Therefore, Nvidia made a desperate bet this time and spent a huge amount of money to ensure the smoothness of the supply chain, which is nothing more than to gain a firm foothold in the industry and continue to occupy a leading position.
So the question is, Nvidia has invested so much money to get its own product line, will it be a pressure on domestic GPU manufacturers? Personally, I think so.
Because judging from the current data, the global HBM3e memory and CoWoS packaging production capacity will be in short supply in the first half of 2024. Nvidia has taken the lead in occupying the mainstream supply channels, which will undoubtedly have a certain impact on some manufacturers with slightly inferior strength.
However, crises often mean opportunities. This pressure from NVIDIA will instead promote domestic manufacturers to accelerate independent innovation and break through key technologies, so as to truly realize the independent and controllable domestic GPU.
For example, domestic manufacturers can bypass the bottleneck of HBM3e memory and CoWoS packaging, purchase traditional DRAM chips from the spot market, and then stack and package them into low-specification HBM memory. Of course, the ultimate goal is to develop a real HBM chip and achieve a complete domestic substitution.
In addition, there are already some domestic packaging factories that are laying out CoWoS production capacity, such as through UMC, GF foundry interposer, and then entrusting packaging to OSAT manufacturers such as Amkor. Although the yield and process nodes of this model are still lagging behind TSMC, with the continuous improvement of the process in the future, the strength of domestic CoWoS packaging will also increase greatly.
What's more noteworthy is that giants like Samsung and Intel are also stepping up the layout of CoWoS full-stack solutions. Once their productization plan is implemented, the global CoWoS production capacity pattern may undergo new changes, which will undoubtedly bring new opportunities for domestic manufacturers.
So to sum up, I think NVIDIA's heavy investment of 45 billion US dollars, although it has brought some pressure to domestic GPU manufacturers, but at the same time, it has also forced us to accelerate the pace of independent innovation. With the continuous progress of domestic HBM memory and CoWoS packaging technology, I firmly believe that one day, the strength of domestic GPUs will compete with top foreign manufacturers and dominate the global market!
Okay friends, about Nvidia's $45 billion core purchase, I will briefly analyze it here. I believe that everyone has a clearer understanding of the development status and future trend of domestic GPUs.
However, I still want to emphasize once again that although there are indeed some shortcomings in the key technologies of domestic GPUs, as long as we persist in independent innovation, we will eventually achieve a breakthrough. Just like the pressure brought about by NVIDIA's huge investment this time, it will push us to work harder and be more focused, so that we can eventually keep up with or even surpass the advanced level of foreign countries.
Therefore, there is absolutely no need to be pessimistic about the prospects of domestic GPUs. In the long run, with the strong support of national policies and the unremitting efforts of technical workers, I firmly believe that China's chip industry will be able to catch up and the day will eventually come!
Finally, I'm happy to share with you some of my thoughts on this matter. If you have any other thoughts or questions, please let me know in the comments section and we'll have an interactive discussion. Looking forward to your wonderful views!