news

after delivering its disastrous second-quarter report, intel's stock price dropped 26% the next day, and its market value shrank to less than $100 billion, barely 1/3 of amd and less than 3% of nvidia.

ceo pat gelsinger immediately launched three plans to increase revenue and reduce expenditure: first, transfer 15,000 employees to the society; then sell asset packages to raise funds. the programmable chip unit altera and the autonomous driving chip company mobileye were put on the shelves; and then intel plan to spin off the chip manufacturing department and get rid of this burden once and for all.

rumors of qualcomm acquiring intel would have been considered onion news five years ago, but when public opinion began to seriously analyze the feasibility of the acquisition, the seriousness of the problem became apparent.

in 2020, gelsinger returned to intel with a technical background. the market expects him to set things right as ceo and lead intel back to the growth track. however, the drastic reform plan has reached the mid-term acceptance stage, and the former semiconductor overlord can only feel sorry for itself in the triumphant song of magnificent 7.

the collapse of a building never happens overnight, and the dusk of big blue does not come unprepared.

the failed pendulum

the backwardness of chip manufacturing processes is the core factor behind intel's rapid decline in the past few years.

intel is one of the few companies in the semiconductor industry that still adheres to the idm model. the so-called idm can be simply understood as the design, manufacturing, packaging and testing of chips are all handled by oneself. the advantage is that it has strong production capacity and can implement its own strategy in an all-round way. the disadvantage is that it has a long business front and high investment costs.

based on the idm architecture, intel formulated atick-tockproduction model: that is, taking two years as a unit,the "tick year" focuses on chip manufacturing and updating chip processes, while the "tock year" focuses on chip design and updating the architecture.

for example, sandy bridge in 2011 and ivy bridge in 2012 used the same architecture, but the process was upgraded from 32nm to 22nm (tick). haswell in 2013 continued to use the 22nm process, but updated the architecture (tock).

intel relied on this strategy to conquer all the invincible players in the world and gained a near-monopoly market position. but in 2014, the “tick-tock” pendulum began to fail.

although the 14nm process was successfully mass-produced in 2014, it has been two and a half years since the previous generation of 22nm. it took a full four years for the 10nm process to be mass-produced.

in the past few years, intel has been squeezing toothpaste, just in time to catch up with samsung and tsmc, which have pushed the process to 5nm in one go. relying on tsmc's production capabilities, amd began to fight back, and the company's market value historically surpassed intel's in 2022.

when kissinger took office in 2021, intel's stock price rose by 7% that day. but what the former received was actually a mess:its core business has been cannibalized by amd, and chip manufacturing has been left far behind by tsmc. intel is completely unprepared for the upcoming explosion of the artificial intelligence market.

kissinger's solution is an ambitious idm2.0 plan - breaking the tradition of "making your own chips". the design department can find foundries like tsmc to produce chips, and the manufacturing department can take oem orders from other chip companies. .

but the particularity of intel is that intel's manufacturing business only serves its own chips. the chip process is constantly being upgraded, and the old process is idle and eliminated.

for example, when samsung advances its own chips to 7nm, it will let its eliminated 14nm production line take on external orders. intel had no foundry business before, resulting in the chip manufacturing department not having mature process capacity. by 2021, intel's 28nm and above production capacity will be almost eliminated.

based on this, intel proposed to acquire tower semiconductor, the ninth-ranked israeli chip foundry. the latter not only has a complete and mature process production line, but also has deep roots in automotive chips, industrial chips and other fields, perfectly complementing intel.

however, god's calculations are not as good as those of humans. geopolitical factors caused governments of various countries to tighten supervision of cross-border mergers and acquisitions, and the acquisition ultimately failed. although the orders in hand are not small, they have been unable to become the company's revenue.

in 2023, intel's foundry department contributed us$7 billion in losses. this year, it will continue its efforts and has already lost us$5.3 billion in the first half of the year. the radical "four-year, five-node" plan is still at the level of the ppt pie, and the heavily invested wafer fab has also stopped. in the "three connections and one leveling" stage.

it is difficult for a company to make mistakes in the trough, but it is easy for a company to make mistakes in its peak years. intel's various shortcomings were buried in their heyday.

audience in the center of the stage

the 2010s were a glorious decade for intel, but also a lost decade. intel struggled to find competition across its product lines, at the cost of missing out on almost everything there was to miss.

the refusal to supply mobile phone chips to apple is an epic stain on intel, but when looking at the technological revolution in the semiconductor field after 2010, from smartphones to autonomous driving to artificial intelligence, intel has always been the awkward audience in the center of the stage.

in the artificial intelligence track, intel had at least four opportunities to play, but without exception they ended in failure.

in 2009, intel launched a project called larrabee, aiming to launch a truly universal gpu based on the x86 architecture [3].

at that time, nvidia's gpu had attracted the attention of academic circles because of its super parallel computing capabilities. the person in charge of the larrabee project happened to be current intel ceo pat gelsinger, but gelsinger jumped to emc in the same year that the project was launched, and larrabee could only give up halfway.

xeon phi is the successor of larrabee and is still positioned as a general-purpose gpu. but intel's idea changed from designing a new gpu to "giving the cpu computing power comparable to that of the gpu" [3]. due to high costs, xeon phi had almost no customers during its extremely limited product cycle. coupled with the difficulty in producing 10nm, it was eventually canceled in 2017.

before axing xeon phi, intel completed two major mergers and acquisitions. one is to buy the "second fpga" altera, and the other is to acquire the ai ​​chip company nervana. the former represents a potential hot track for ai inference chips, and the latter is a technical route similar to google's tpu - tpu drives alphago. the hero behind the scenes.

unfortunately, before being spun off or terminated, altera only contributed an average of us$500 million in revenue per quarter, and nervana only delivered one product, nervana-npp[4].

in the two emerging technologies of mobile chips and autonomous driving, intel is repeating similar stories:

the atom series, which focuses on the mobile terminal market, was born in the second year after the iphone was launched. in a period when it should have been pursuing victory and iterating rapidly, atom waited for five years but stayed put. because of concerns that atom, which focuses on cost-effectiveness, will squeeze the production capacity of other products, intel's internal resource feeding priority has been repeatedly ranked lower.

mobileye was the king of autonomous driving chips at the time, but after being acquired by intel, it stalled. its chip performance fell out of the first echelon, and it could only rely on l1/l2 low-level autonomous driving to barely maintain its respectability.

unlike the stories of ibm and cisco, intel has a very keen sense of emerging fields and has not slackened in its investment in new technologies. at its peak, intel's r&d spending alone accounted for 30% of the semiconductor industry.

opening intel's "technology toolbox", there are also the baseband business acquired from infineon and later sold to apple, the high-speed storage product optane (which will be shut down in 2022) that has been losing money, and even unmanned aerial vehicles. machinery business (sold to musk in 2022).

but throughout the 2010s, this "try hard but only a little bit" r&d and investment style continued throughout, ultimately causing intel to miss out on almost all emerging technologies after the pc.

corresponding to these passes is the frequent turnover of company ceos - no snowflake is innocent in the avalanche.

let the ceo work for a few more years

since 2000, intel has had a total of five ceo changes. this frequency is considered normal for a football team, but it is taboo in high-tech companies.

on the matter of missing the technology wave, every ceo of intel can be criticized. the first is paul otellini:

under otellini, intel successfully pulled apple into the x86 camp, but because it refused to design chips for apple, it missed the best opportunity to enter the mobile terminal market.

after looking back, otellini tried to remedy the situation, first launched the atom series of processors, and entered the mobile market with netbooks. although the performance is mediocre, the vitality of this chip is not bad. some old tesla models are equipped with atom processors.

subsequently, otellini began to invest heavily in euv lithography machines, investing us$4.1 billion in asml's euv project, which was more than tsmc (us$1.4 billion) and samsung (us$974 million) combined [6].

but both investments were shelved after otellini resigned in 2013. his successor krzanich did not believe in the scale effect of euv lithography machines, and insisted on using more mature and cheaper duv lithography machines at the 10nm process node. as a result, the 10nm yield rate has been delayed and was cornered by samsung and tsmc. overtake.

as early as 2017, intel acquired the first euv lithography machine, but it has been gathering dust until 2021. in the meantime, intel has produced seven generations of products at the 14nm node, and the name of the toothpaste factory has become increasingly famous.

frequent personnel changes have caused instability in various business strategies, which in turn has affected intel's long-term investment in emerging technologies.

krzanich led the acquisition of nervana during his tenure, which was intel's fourth attempt in the field of artificial intelligence. however, in 2019, krzanich resigned due to his failure to control his lower body in the office, and nervana died as expected.

after robert swan, who had a financial background, took over, he acquired the ai ​​chip company habana labs and transferred all the resources that should have been invested in nervana.

the helmsman of a high-tech company needs a keen sense of smell, a long-term vision, and a strong approach to implementation, but what carries everything is a long enough time in office, so that long-term technology investments can maintain stable operation and be able to obtain a steady stream of benefits. resource support.

intel's biggest competitor amd has also experienced turmoil. in 2014, su zifeng also took over a lot of mess and served as ceo for ten years. it was not until after 2020 that her series of strategic plans were gradually translated into actual market share and company performance.

when compared horizontally with technology companies such as microsoft and google, intel's leadership changes are too frequent. in contrast, huang renxun, who just celebrated his 60th birthday, has no intention of retiring, saying that he wants to follow zhang zhongmou and work for another 30 years. zuckerberg just turned 40 this year and is in his prime.

another problem for intel is that after struggling alone in the x86 market, its next strategic direction has always been unclear.

for high-tech companies, the successor is positioned as the "executor of the next strategy." therefore, most companies willchoose the person who handles the strategic business of the future, rather than the person who handles the traditional business of the past.

current ceo of googlebefore sundar pichai took over, he managed the chrome and android businesses and had almost no contact with the search business. bezos' appointed successor is andy jassy, ​​who founded the aws cloud computing business and has almost no control over e-commerce. coincidentally, before taking office, satya nadella had never touched windows and office, but relied on cloud computing to build his career.

intel has keenly captured almost all revolutionary emerging technologies. unfortunately, compared to the desperate efforts in "only the paranoid survive", they have only failed a little bit on every track.

references

[1] Intel used to dominate the U.S. chip industry. Now it's struggling to stay relevant，CNBC

[2] Intel’s Lunar Lake is actually made at TSMC，PC World

[3] Project Larrabee: How Intel's First Attempt at GPUs Failed，How-to Geek

[4] on the fall of intel nervana, semiconductor industry observation

[5] unwilling to wait for 450mm and euv technology, intel invested us$4.1 billion in asml, eet-china

[6] pat gelsinger: intel’s native technology believer, techsugar

[7] only the paranoid survive, andy grove

editor: li motian

visual design: shu rui

editor in charge: he luheng