Reporting by XinZhiyuan
EDIT: LRS
Nvidia CEO Huang Jenxun recently said that it is considering using Intel foundries, and once the cooperation is reached, the graphics cards produced can be called the "double Ying" of the chip industry, but some analysts said that this is just a polite remark.
What the? Nvidia is actually going to work with Intel?
At a recent online conference, Huang Renxun revealed that they will consider using Intel's foundry services, and said that Intel is also very interested in letting Nvidia use their foundry services.
However, as far as the news released so far, Nvidia will not soon hand over the chip to Intel FOUNDC, and the negotiations between the two sides around chip foundry services may take a while.
Lao Huang said that after all, it is not as simple as going to the supermarket to buy a bottle of milk. The two sides need a lot of time to conduct in-depth negotiations, including the integration of the supply chain and so on. NVIDIA's cooperation with TSMC and Samsung in the past few years has also undergone years of development, and they are also open to cooperation with Intel, and they are very happy about Intel's efforts.
As soon as the news came out, Intel's stock price rose 2.5% at one point, and the stock price fell back at the close.
However, the outside world also expressed great concern about this cooperation, after all, Nvidia and Intel are competitors in many fields.
Whether it is Nvidia's efforts in the development of Grace CPU or Intel's upcoming DG2 graphics card, it can be seen that Shuangying intends to seize the territory in the "enemy camp".
In response, Lao Huang responded that they have always worked closely with Intel. They shared the roadmap with Intel long before they announced it, and Intel knew a lot of their secrets, and AMD knew it, but Nvidia was mature enough to realize that they needed to work together.
Lao Huang also said that Intel needs to learn to adjust itself to meet the needs of customers, because chip foundry is not Intel's specialty, but entering the foundry field is also what Intel must do.
Some analysts believe that Huang's reference to the cooperation with Intel may be "polite".
Bernstein analyst Stacy Rasgon said, "Choosing Intel is an option for Nvidia, and it doesn't come at any cost. But now you still can't know anything about how the two sides will work together."
Gartner analyst Sheng Linghai said, "Any possibility exists, for Intel, it now urgently needs to use the x86 architecture to attract foundry customers, otherwise it will soon be overtaken by arm."
So does Nvidia really have feelings for Intel?
Intel TransformationAL FOUNDI
Throughout history, Intel has not been a "serious" foundry.
Toothpaste Factory once had the most advanced technology in the world, single-handedly competing with the world's simple instruction set, using its market size and cooperation with Microsoft to dominate the global computer CPU market.
In 2008, Intel publicly declared that it wanted to open the foundry business, and in 2013, it also said that it would open the chip foundry business to all chip companies, and it was indeed for LG and Spreadtrum and other manufacturers to produce chips, but there were many restrictions, and it could not oem all types of products, and the price was higher than the market price, so the volume has not been very large.
For the huge Intel, foundry has always been a dispensable state. Intel's own chip demand is already very large, the production line of the light of its own business capacity has been full, while Intel's own chip research and development, production of one-stop service, the larger the production capacity, the more can dilute the cost, is conducive to Intel to earn more profits.
Because Intel itself has always adhered to the IDM (VerticalLy Integrated Manufacturing) model, that is, its own design, its own production, its own packaging and testing, it does not need others to help its own OEM, and relying on x86 architecture Intel has been able to lie down and make money, why do oem hard work, there is not much interest in helping others, self-sufficiency.
After entering the era of mobility, Intel, which is accustomed to squeezing toothpaste, has been thrown down a lot.
With the rise of AMD, the server market and desktop processors have been hammered by agricultural enterprises, sold cheaply, and have more cores.
And the ARM camp on Intel's market is also eyeing the tiger, in the past Apple computer has been dependent on Intel's chips, but the new M1 chip performance is good, but also shows that ARM can not only occupy the mobile end, for the PC side can also be a great achievement, and Microsoft also expressed support for ARM.
At the same time that Intel was suffering from the enemy, the bad news came again, the process was not going well, the market was still the old 14nm, the difficult to produce 10nm chip was not released until 2021, and the chip produced using 7nm technology would not be available until 2023.
In contrast, AMD has already released a 7nm Zen3 in 2020, and will release a 5nm Zen4 this year.
However, Intel's technical background should not be underestimated, Intel's process has reached a transistor density of 106 million / mm at the current 10nm node, slightly exceeding the density of 0.95 billion / mm and 0.97 billion / mm on 7nm respectively. In the next generation of 7nm nodes, Intel aims to increase the transistor density to 180 million / mm, if the plan is achieved, its density will not only exceed the current TSMC's latest 5nm process, but also greatly shake off Samsung's 5nm.
But it is undeniable that Intel's market position is declining year by year, last year backed by its own factory, relying on the advantages of spot and AMD can still compete, but what to do next?
In 2021, Intel's new CEO Pat Gelsinger took office and announced intel's IDM 2.0 strategy, which plans to become the world's first chip manufacturing again. It intends to invest $20 billion in the United States to build two chip factories and set up an Intel foundry service division to be responsible for oem chips for other manufacturers.
The chip factory officially started construction on September 25 last year, built on Intel's Ocotillo campus in Chandler, Arizona, and is scheduled to be operational in 2024.
At The Inel Accelerated conference last July, Kilsinger revealed that it had signed a contract with AWS, and Amazon would be the first customer to use the Intel Foundry Services (IFS) packaging solution. At the same time, Qualcomm will also use Intel 20A process technology, which is expected to be launched in 2024.
In early 2022, Kirsinger struck hard: announcing that the investment could increase to $100 billion, building eight factories, and planning to build the world's largest chip manufacturing base in Ohio, USA. However, whether the other 6 fabs will continue to be built depends on the demand for Intel's foundry services. Intel also poached Jeff Wilcox, the leader of Apple's M1 development team, to be responsible for all SoC architecture designs.
If Nvidia's order is won this time, it will not only "steal" some graphics technology, but also may also change Intel's passive situation in the field of wafer manufacturing.
TSMC is not enough
At the interview, Huang Jenxun mentioned that NVIDIA's current major foundries include TSMC and Samsung, and did not mention Intel.
But he also mentioned that building a foundry the size of TSMC is not suitable for the faint of heart, and TSMC works with more than 300 companies around the world.
Is it possible for Intel foundries to grow to the scale of TSMC?
TSMC is a pure foundry, and it does not design complete chips themselves, although their ability to make many modules may be much stronger than that of design manufacturers. However, due to TSMC's commercial positioning of "only doing OEM", customers will not have worries and will be assured to give orders to TSMC.
At that time, Intel was to prevent its own technology from leaking, and only then did it use the IDM model to produce chips. Nowadays, Intel on the one hand FOUNDRY, on the other hand to produce its own brand of chips, some large manufacturers are difficult to rest assured to hand over the design files to Intel, so it is difficult to say whether Intel foundry can develop to the scale of TSMC.
However, TSMC's hot market also shows that customers need to queue up to produce chips.
In order to compete for TSMC's 5nm production capacity, as of the fourth quarter of last year, NVIDIA has paid a high advance payment of $9 billion to ensure production capacity, and will pay $1.79 billion in the first quarter of this year, with a total amount of more than $10 billion.
Nvidia's next-generation graphics card, the GeForce RTX 40, will have a built-in Ada Lovelace GPU, and amd using the same TSMC 5 nm process, and Nvidia has invested heavily to ensure RTX 4080, 4090 and other RTX 40 series 5nm production capacity.
Some analysts believe that NVIDIA invested money to grab TSMC's production capacity, because GPUs are its main source of revenue, and if sales are poor, it will be a major blow to operations. And according to data from analysts, NVIDIA accounted for about 5.8% of TSMC's revenue last year, making it the seventh largest customer.
High costs and insufficient capacity may also be a reason for Nvidia to embrace Intel.
However, whether the two Ying can be combined is still far from being unknown.
Resources:
https://www.bloomberg.com/news/articles/2022-03-23/nvidia-ceo-to-look-at-intel-for-foundry-says-shift-will-be-hard
https://wukong.toutiao.com/question/6834331881677259023/