news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

[Text/Guacha.com columnist Pan Gongyu]

In the vast desert plains of Arizona, more than 300 days of sunlight shine on the factories of semiconductor giants such as Intel, TSMC and NXP every year. The factories are equipped with the most expensive equipment used in semiconductor manufacturing - lithography machines.

Intel employees once discovered that the throughput of the lithography machine would suddenly drop between 1 and 2 a.m. After a long investigation, Intel finally found the "murderer", their lovely neighbors - hundreds of cows.

The cows in the nearby dairy farm fart a lot every day. The methane produced by the cows farts a lot every day. Due to the change in wind direction between 1 and 2 a.m., it enters the clean room through the air purifier, affecting the efficiency of the lithography machine. Intel came up with a perfect solution: relocate the farm. And remember the lesson: don't build a factory in a place where there are a lot of cow farts.

As a giant in the global chip industry, Intel, like Coca-Cola and Ford, is a symbol of the "American spirit of the times". They have too many stories with lithography machines. Over the years, they have been constantly solving the "bullshit problem" of old machines and have been constantly purchasing new machines to maintain their core competitiveness in the chip manufacturing industry.

In April this year, Intel announced that the assembly of the first High NA (high aperture) EUV was basically completed, and the company also mobilized its most elite propaganda forces to shoot a set of exquisite videos showing the transportation and assembly process of this EUV lithography machine, which cost more than 350 million euros and can be used for chip manufacturing below 3nm, from ASML to Intel's Oregon factory. The momentum was so great that a large number of industry peers and US officials formed a unified trend of forwarding on social media.

Intel, the world's first customer for High NA EUV

From relocating the dairy farm to buying High NA (high numerical aperture) EUV, there is a consistent behavioral logic: the "personality" of the leader of American semiconductor manufacturing must be established - Intel used to have a team of disciplined and hardworking engineers who could always troubleshoot various equipment failures in a timely manner to ensure yield and production capacity; in 2024, Intel will also maintain its personality as a pioneer, and they are still the most solid force and loyal executor of the US "chip manufacturing reshoring" plan. However, the glamorous appearance and proud demeanor can hardly stop the external bleeding of internal organs.

Intel's Bleeding History and the New CEO's Demolition

On August 1, Intel's Q2 financial report was a striking torrent in the avalanche of US stocks. The stock price fell 20% that day, and the market value evaporated by 32 billion US dollars. At first glance, revenue only fell 1% year-on-year, but the losses of key departments such as foundry and autonomous driving business were increasing. What shocked the outside world was the announcement of layoffs of more than 15% (more than 15,000 jobs) and the decision to suspend dividends starting in the fourth quarter, which soon triggered a class action lawsuit against the company by small and medium shareholders.

However, just nine months ago, the "first son of American chips" was still looking forward to the beautiful vision of "four years and five nodes" and IFS, the foundry business, becoming the second largest in the world by 2030. Quietly, Intel's market value is getting further and further away from 100 billion US dollars, only less than 40% of AMD and less than one-tenth of Nvidia.

Stock prices, market capitalization and corporate growth are closely related. In the eyes of the bigwigs in the capital market, Intel's long-term plans for the next four to five years have become ridiculous in the face of the current dismal performance. It is no wonder that those who filed a class action lawsuit against the company believe that the company's management did not treat investors honestly in the first half of this year and suspected financial fraud to deliberately raise the stock price.

In January 2020, Nvidia surpassed Intel in market value and then left in the dust

Intel's massive layoffs are not without reason. As early as 2022, Intel encouraged up to 2,000 employees in Ireland to take three months of unpaid leave to cut costs. It should be noted that Ireland is a "heavy position" of Intel's overseas manufacturing department. Before Germany launched a large number of projects, Ireland was Intel's main export port for foundry work, compared with Israel, which focuses more on research and development.

Looking back, there is only a thin wall between the falling petals and the three-foot-thick ice. If we ignore Intel’s series of decision-making mistakes in areas such as data centers and high-performance computing, and how they have missed the mobile processor and AI booms, and only observe the process of Intel’s “foundry business” in the past three years, we can get a glimpse of the depth and danger of its problems.

In March 2021, the current Intel CEO Pat Gelsinger succeeded Bob Swan as the company's new head. The two can be said to be a miniature history of Intel during the period of global integrated circuit development after the 1980s. Gelsinger was the company's chief architect of the 80486 processor, personally instructed by industry legend Andrew Grove, and was the company's youngest vice president and CTO.

He seemed to have enough confidence to announce "setting things right" right after taking office, and he heavily disparaged his predecessor Swan on the company's development plan, believing that his own background as a professional engineer was more in line with the company's background, and that Swan's advancement path from CFO to CEO was tainted with too much "money smell" in the capital market and deviated from the development route based on manufacturing.

Kissinger's "house-demolition" journey after taking office

From the timeline, we can see that Kissinger basically reshuffled the top-level structure 18 months after taking office, and started a new round of business integration. At the investor conference in the summer of 2022, he divided all the company's business lines into "traditional" and "emerging."

The Data Center and Artificial Intelligence (DCAI) Division, the Client Computing Group (CCG) and the Network and Edge Group (NEX) represent traditional businesses; the Foundry Business (IFS), the Accelerated Computing Systems and Graphics Group (AXG) and the Mobileye autonomous driving business represent the company's emerging business direction.

What surprised the outside world was that within three months after Kissinger announced the three emerging businesses, he almost single-handedly split and reorganized them again. First, he reclassified GPUs into consumer graphics cards and data center AI accelerators, splitting them into two different departments; secondly, in the second half of 2022, he announced a major separation from Mobileye, allowing it to go public independently and no longer consolidate its financial statements. Intel's collapse was foreshadowed in its trade-offs in dealing with emerging businesses.

Intel has redivided its GPU business map in the past two years, and its "divorce" with Mobileye seems to be due to its own reasons. The former has been forced to retreat by Nvidia and AMD, and the latter's main "black box" model of autonomous driving is also extremely expensive. Moreover, the acquisition of Mobileye was the proud move of its predecessor. While L3 and L4 are still hot "concept stocks" in the assisted driving market, Mobileye was allowed to go to the capital market to raise more funds alone. This can get rid of the burden of burning money and clean up the predecessor's team. This approach is quite reasonable.

Compared with these two business lines, Intel's IFS, or foundry business, has become the darling of the spotlight since the plan was announced in March 2021, and it also embodies Kissinger's efforts in his second venture.

So, how has Intel's foundry business (IFS) performed over the past three years? In a nutshell, it is currently a scar on Intel's torso.

Intel is willing to "fake" to discredit its OEM reputation

Compared with the 1990s or more than a decade ago, the global semiconductor foundry industry has now crossed over the barbaric growth period that was more suitable for the law of the jungle and entered a relatively stable high-end era with relatively clear division of territories among various lords.

Looking back, perhaps 2018 and 2019 were watershed years. Not to mention SMIC, which is currently concentrating on its development, Globalfoundries and UMC were unable to support the exponentially rising capital expenditures and announced their withdrawal from the competition for advanced processes in the past two years.

The world below 7nm is a paradise for alternative adventurers, and the competitive relationship between TSMC, Samsung and Intel has become increasingly subtle. Pat Gelsinger realized that the best way to get rid of the reputation of "toothpaste factory" that hesitated on 10nm process is to split the design and manufacturing departments. Only by making the foundry department (IFS) equally open to major global design companies outside and splitting design and manufacturing internally can the traditional Tick-Tock strategy be further upgraded (one year to upgrade the process and one year to optimize the architecture).

At first glance, Intel's strategy of fighting each other conforms to the trend of division of labor. Under the politically correct background of maintaining the trend of manufacturing return, it can also make the financial statements look better. Although reducing the quality of IDM is not an "Aladdin's lamp" that can solve all problems, AMD's rapid rise after ceding its foundry department to GlobalFoundries has proved that this move is quite effective.

At the end of last year, a piece of news shocked the global semiconductor foundry circle: Intel IFS's revenue in Q3 of that year reached US$311 million, making it into the world's top ten for the first time in history. Then Intel officially announced that it would strive to rank second in the world in foundry business by 2030.

The top ten narrative seems like a farce at the moment

Kissinger kept telling all major media that "Intel has secured outsourcing contracts for 11 of the top 15 design companies in the world," but Intel's annual financial statements four months later instantly dispelled all the high-quality narratives.

Intel's foundry losses for the whole year of 2023 will be $7 billion, with losses of $2.5 billion in the first quarter of this year and $2.83 billion in the second quarter. In the eyes of Intel's new generation of management, the ever-expanding foundry losses are more serious than the delayed release of the Ultra "Arrow Lake", Xeon 6 server processors, and Gaudi 3 artificial intelligence accelerators.

Just after the news that IFS briefly broke into the top ten of the global foundry industry at the end of last year, Intel announced that it would cooperate with Taiwan's UMC (currently ranked in the top four in the world in terms of market share of pure foundries) to develop a 12nm process, and used UMC as a lobbyist to make MediaTek, another important member of the "United Family Gang", the first important external customer of IFS. In this regard, Samuel Wang, a semiconductor analyst at Gartner, a world-renowned consulting firm, once told the author at an exchange event: "Compared with 2022, Intel IFS's revenue in 2023 suddenly increased by a hundred times, but the capital expenditure on related equipment did not increase, but basically remained the same, which is very strange. After further research, I found that Intel adjusted the base of the automotive chip packaging business, increased the multiple, and misappropriated data from other business departments, creating a 'lie' that IFS could break into the top ten."

Since 2020, the US government has launched a whirlwind of chip manufacturing repatriation, from the "Endless Frontier Act" to the "Chips Act". Intel, as an important customer of the US Department of Defense, is the only local company capable of assuming the role of frontier foundry industry. However, the IFS created by Kissinger is unable to shoulder the heavy responsibility of the times. What is the reason?

Intel cannot truly separate its “internal” and “external” businesses

The unrealistic narrative of "IFS has made it into the top ten in the world" is constantly facing more cruel reality tests. Claus Aasholm, a well-known Danish semiconductor analyst who was once interviewed by Mind Observer, wrote an article titled "Intel's Death Spiral" on the social media semiwiki, pointing out that the main dilemma of IFS is that it has not yet formed a real customer service ecosystem.

Intel's external foundry business (blue) accounts for a very small proportion of the entire manufacturing department

Moreover, Intel's choice to cooperate with UMC and MediaTek has made their competitive and cooperative relationship with TSMC more complicated.

Kissinger talked about the company's product line and technology evolution at the Deutsche Bank Technology Conference last September. Whether it was an unintentional mistake or deliberate rhetoric, he said this very casually: "We are TSMC's customers. So I know their wafer costs, wafer average selling prices, and the N2 budgets they show to their N5 customers and N3 customers. We know what our goals are. They (TSMC) used overly expensive technology at different market price points before. Therefore, we used a bad economic model to present costs instead of market prices, resulting in inherent inefficiencies."

It should be noted that Intel's two advanced PC processors code-named Arrow Lake and Lunar Lake are indeed manufactured by TSMC. Kissinger actually publicly said that he understood TSMC's specific costs in the field of cutting-edge processor foundry and then "prescribed the right medicine", which obviously violated the most basic principles of commercial confidentiality.

Kissinger's remarks at least show that Intel is still in the initial exploration stage in the foundry competition and cooperation relationship. Without success in the "internal", it will not know about the "external". How can it further win the trust of big customers like NVIDIA?

(High-NA) EUV lithography can't save Intel

According to the "Rayleigh formula" of lithography machines, the improvement of lithography technology has been a comprehensive attack in multiple dimensions over the past few decades, that is, the continuous optimization of exposure wavelength, numerical aperture and process factors. However, the current shortening of exposure wavelength (13.5 nm for EUV) and the increase of numerical aperture (NA) (currently increased from 0.33 to 0.55) have already approached the limits of comprehensive physical and cost considerations.

An important part of the "turnaround" discourse system of Intel's new CEO after taking office is that Intel's lagging behind in cutting-edge technology many years ago was the result of not using the most advanced equipment. So Intel spent a huge amount of money to buy the industry's first commercial high numerical aperture (High-NA) EUV lithography machine, and there was news that Intel had "contracted" all of ASML's high-NA EUV lithography machine production capacity this year.

In Intel's view, this (High-NA) EUV lithography machine is like Sun Wukong obtaining the golden hoop, relying on huge force bonus to go to the sky and into the earth. Intel plans to use it for the first time in its node Intel 20A (equivalent to 2nm) in the first half of 2024, and plans to use it in Arrow Lake platform in future mass production.

Intel's four-year, five-node plan

However, the cost of chips manufactured using (High-NA) EUV lithography machines has increased dramatically. Although more chips can be cut from each wafer, more chips need to be sold to make up for the investment. It is still impossible to determine whether the market demand for the most advanced process is sufficient to absorb the cost. In other words, whether (High-NA) EUV lithography machines can continue to be mass-produced and the maturity and stability of the machines have not been sufficiently verified by the market. Because of this, TSMC did not rush to purchase from ASML, but chose to optimize computational lithography capabilities, mask manufacturing and computational infrastructure, including new materials, step by step.

More importantly, TSMC has a very large and high-quality customer base to help them debug various equipment bugs, which is exactly what Intel lacks. Jay Goldberg, president of the US semiconductor consulting firm D2D Advisory, also pointed out in a conversation with the Mind Observer: "The real challenge facing Intel's foundry is that they must have more customers in their economic model to support the research and development required to continuously advance their manufacturing processes. For many years, they only needed Intel as a customer. But now they need external revenue of the same scale and must double the scale of revenue to support continued progress on the track of Moore's Law."

Intel is currently caught in a dilemma. On the one hand, they need the most advanced (High-NA) EUV lithography machine as a production and marketing tool to declare their R&D and manufacturing capabilities below 3nm and expand their customer base. However, as a new machine, the (High-NA) EUV lithography machine makes Intel a guinea pig-like experimental product, and they have to force it to launch it despite the pressure of equipment depreciation and mass production amortization costs to quell external doubts.

Conclusion: Maybe it is really just a game for East Asians

Intel IFS has set a goal of becoming the world's top two by 2030, seemingly assuming that TSMC can only be the first. However, TSMC's plan to build a factory in Arizona, the United States, has not progressed smoothly. The 5-nanometer wafer factory originally scheduled to be put into production in 2024 has been postponed many times and is currently expected to be postponed to 2025. TSMC is known for its harsh working environment. Not only do they often work overtime on weekends, but they even urgently recall employees in the middle of the night. The management of Taiwan is known for its tough style. Employees may face severe penalties or even dismissal for the slightest mistake. However, this management model is not suitable for the United States.

TSMC's strategic decision to build a factory in the United States even led to the resignation of former executive Liu Deyin, and the "dual head" became Wei Zhejia alone. I believe TSMC must have evaluated the issue of workplace culture integration on the other side of the ocean.

Of course, Kissinger can sneer at TSMC boss Morris Chang's assertion that "cutting-edge chip foundry is a natural product of East Asian culture", but the repeated delays of TSMC's Arizona plant are enough to sound the alarm for Intel. In the foundry world below 5nm, hunger, sensitivity to danger perception and speed of action are all indispensable. They need to find the root cause of the problem like tracking cow farts in the past - a problem that seems metaphysical but has a solid causal chain.

This article is an exclusive article of Guancha.com. The content of the article is purely the author's personal opinion and does not represent the platform's opinion. It cannot be reproduced without authorization, otherwise legal liability will be pursued. Follow Guancha.com WeChat guanchacn to read interesting articles every day.