news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

for the storage industry, mixed blessings and sorrows are the truest portrayal now.

fortunately, dram and nand are recovering, as well as hbm, which has emerged as a new force. what is worrying is that the industry prospects are not clear, and the growth has slowed down before it lasted long.

this is true for korean manufacturers, american manufacturers, and japanese manufacturers. although the continued sales of ai chips have brought many opportunities, the continued downturn in the consumer market has also caused headaches for several manufacturers.

some are happy and some are sad, which is undoubtedly reflected in the recent trends of major storage plants.

sk hynix

sk hynix is ​​certainly the one with the least worries.

although morgan stanley was bearish on sk hynix in september, sold its shares, and lowered its investment rating by two levels, causing panic in the korean stock market for a while.

fortunately, at present, sk hynix’s prospects are relatively optimistic. compared with morgan stanley, most brokerages have maintained relatively stable expectations for sk hynix.over the past month, third-quarter and annual operating profit forecasts for sk hynix have been revised only slightly downward or remained stable.

what needs to be focused on is that recently, sk hynix announced the start of mass production of 12-layer high-bandwidth memory 3e (hbm3e), becoming the world's first supplier to mass-produce this memory. it plans to start delivering this advanced chip later this year , which is the first 12-layer version of advanced memory it sells to nvidia.

in march this year, sk hynix began to supply the 8-layer version of hbm3e to customers such as nvidia. the maximum capacity of each hbm3e chip is 36gb, exceeding the 24gb upper limit of hbm3. the data transfer rate is 9.6 gb per second, which means using four the gpu of a 36gb hbm3e chip can read 70 billion parameters 35 times per second.

the stacks use sk hynix's large-scale reflow molding underfill packaging technology, which is said to reduce heat dissipation by "up to 10%", thereby reducing the thickness of the hbm stack. sk hynix said it has been able to create products with 8 with a 12-layer hbm stack of the same layer stack height, the dynamic random access memory modules in each stack are 40% thinner than previous generations.

sk hynix president justin kim said in a statement: "sk hynix once again broke through the technological limits and demonstrated our industry leadership in the field of ai memory. we will continue to maintain our position as the world's largest ai memory supplier. at the same time, we are steadily preparing next-generation memory products to overcome the challenges of the ai ​​era.”

affected by the news of mass production of 12-layer hbm3e, sk hynix's stock price soared 9.44% on thursday to close at 180,900 won ($136).

but hynix is ​​not without worries, and the first one to bear the brunt is nand.

for hynix, before it joined zhizhi memory and acquired intel's nand department, its presence in the nand market was extremely weak and it was difficult to compete with its fellow competitor, samsung.

however, after the shareholding and acquisition, hynix's nand division began to climb rapidly in market share. according to trendforce's latest quarterly report, samsung ranked first with a market share of 36.9%, while sk hynix and solidigm (formerly intel nand division) )’s market share reached 22.1%, firmly occupying the second position, which was unimaginable a few years ago.

however, compared with the rapid growth of market share, it is the burden brought by gradually increasing expenditure. sk hynix acquired intel's nand and ssd departments through a two-stage transaction. in the first phase, sk hynix acquired intel's ssd business and nand manufacturing plant in dalian, china, for us$7 billion. in the second phase, sk hynix will pay intel an additional us$2 billion in 2025 to acquire intellectual property related to nand flash wafer manufacturing and design, r&d employees, and dalian factory employees.

this means that hynix will have to pay an additional us$2 billion in the next year. however, since being acquired by sk hynix in october 2020, solidigm has been in a state of loss. it was not until the second quarter of this year that it turned a profit for the first time. the future is not ruled out. there is also the possibility of losses.

this is also the reason why hynix wants to spin off solidigm and list it in the united states. the public offering of solidigm can enable sk hynix to obtain a considerable amount of cash and make up for part of the capital expenditure. as the hbm factory continues to expand production, solidigm is like it's a burden.

samsung

unlike hynix, samsung is at a crossroads of destiny.

samsung is scheduled to release a provisional third-quarter earnings report in the second week of october, but securities firms have been lowering the company's profit forecasts.

according to fnguide, samsung's third-quarter operating profit forecast was lowered to 11.23 trillion won (range 9.7 trillion to 14.79 trillion won) in september from 13.66 trillion won (range 11.72 trillion won to 15.2 trillion won) in august. korean won), showing a downward trend. 17.78%. the consensus for samsung's full-year operating profit was also lowered from 45.32 trillion won in august to 40.82 trillion won in september, a decrease of 9.93%.

this month, 15 south korean brokerages have lowered their target prices for samsung electronics, lowering the consensus target price to 99,560 won, below the "100,000 won" mark. it seems that no one is optimistic about the development of samsung semiconductor. why has the former storage overlord fallen to this point?

in june this year, nvidia ceo jen-hsun huang said in response to questions about hbm used in his company's ai accelerator: "samsung electronics, sk hynix, and micron will all supply hbm to nvidia." "we are also working hard to arrange (samsung electronics, micron) pass the test as soon as possible and put it into use in the ai ​​semiconductor process. nearly a year after hynix passed the test, samsung has not solved this problem. the dram department that samsung was once proud of has now become "achilles." heel".

although samsung subsequently passed the relevant tests, it was already far behind hynix. this is also related to samsung's previous neglect of hbm technology.samsung has been actively investing in hbm development since 2015 and successfully completed the first mass production of hbm commercial products, beating hynix on the third-generation product hbm2e. however, in 2019, out of due to the slow growth of market demand, samsung reduced hbm investment and reduced the size of the team, which ultimately resulted in significantly lagging behind hynix in hbm3 and hbm3e.

under such circumstances, samsung itself has fallen into "desperation" and is making life-and-death efforts to ensure hbm's leadership position. on september 25, the head of samsung electronics’ storage division told executives and employees, “we must do our best to work together,” “we have determined the direction we must take to restore competitiveness,” and “now we need execution capabilities.”

each department of samsung electronics holds an all-hands meeting every quarter, with senior executives including heads of departments participating, and this meeting was held specifically to recognize the semiconductor crisis.regarding the question raised by employees about "ensuring excellent talents and countermeasures for personnel loss," the person in charge said, "please work hard," and "please protect the employees around you who are about to leave."

although the cold winter has passed or has not yet arrived, samsung's storage department has been feeling cold.

micron

micron is also a typical example of ups and downs.

micron's stock price has experienced roller coaster fluctuations this year. the memory giant soared in the first half of 2024 and hit a record high in june. however, the stock price fell by nearly 40% since then, making it a bearish position for morgan stanley. object.

but recently, micron has swept away the gloom and announced unexpectedly strong sales and profit forecasts, recording the largest gains in the past 13 years. the company said in a statement wednesday that fiscal first-quarter revenue was about $8.7 billion. that compares with analysts' average estimate of $8.32 billion. excluding certain items, profit per share was about $1.74, compared with expectations of $1.52.

micron's performance beyond expectations is mainly due to the artificial intelligence boom. it is understood that hbm demand has always exceeded supply, which allowed micron to increase prices and obtain long-term guaranteed contracts. it said on wednesday that hbm products in 2024 and 2025 already sold out.

the strong performance drove the stock up 15% to $109.88 in new york trading, the largest one-day rise since december 2011. micron has risen 29% this year.

manish bhatia, micron's executive vice president of operations, said in an interview that micron's advantage is that it is the first chip manufacturer to reliably provide more advanced memory in high volumes. he said micron is well-positioned as companies race to enhance their ai software and hardware (and use more memory in the process).

micron has also shrugged off slowing demand for pcs and smartphones, the two largest memory markets. micron said device shipments are now starting to grow again, and bhatia said those devices will increasingly be equipped with artificial intelligence features that require more memory chips to work properly, which brings another benefit.

ceo sanjay mehrotra said in a statement: "strong artificial intelligence demand is driving strong growth in our data center dram products. we will enter fiscal 2025 with the best competitive position in micron's history."

executives expect the hbm chip market to grow from $5 billion this year to $25 billion in 2025, and increased demand for hbm chips will bring in billions of dollars in revenue next year.

but for micron, there are still many problems, not only technical but also market-based. although there is strong demand for artificial intelligence systems that use specialized and expensive memory, most of micron's business still comes from pcs and smartphones. . these are mature markets where sales have picked up slightly this year, but the growth rates are in the single digits at best.

in the hbm market, micron is still the weakest one. according to brokerage forecasts, micron's hbm sales may reach us$4 billion by 2025, which will allow the company to occupy about 16% of the market share. if it wants to rely on hbm's idea of ​​​​boosting the overall business is good, but it is not enough to improve micron's overall situation.

kioxia

for kioxia, the biggest problem is that his own destiny is not in his own hands.

in 2018, toshiba, eager to sell its storage business, attracted a bunch of foreign companies to bid. after a fierce battle, the bidders were reduced to four, namely hon hai precision, sk hynix, western digital and broadcom. .

hon hai had the highest bid, but was eliminated in the first round. the reason was nothing more than its chinese background. broadcom, on the other hand, was also deprived of the qualification to purchase because of possible layoffs after the purchase. the remaining western digital and sk hynix they are all part of the us-japan-south korea alliance, but hynix, which is backed by sk group, is richer and has a higher bid. although western digital strongly opposes it without the lucrative western digital, it does not help.

in the end, toshiba sold it to a consortium headed by bain capital for us$18 billion. the consortium was participated by sk hynix, apple, dell, seagate, kingston, etc. sk hynix occupied a pivotal position in it. sk hynix 4 trillion won (approximately us$2.9 billion) was invested in kioxia, 2.7 trillion won was invested in private equity funds led by bain capital, and 1.3 trillion won was used to acquire the kioxia convertible bonds issued by toshiba.

through this investment, hynix indirectly obtained about 15% of toshiba storage's shares. hynix had certain control over the major changes made by kioxia, which also laid the foundation for kioxia's current situation. .

kioxia originally planned to list on the tokyo stock exchange in 2020, but due to the escalation of sino-us trade tensions, the listing plan was postponed. in 2023, kioxia tried to merge with the memory division of western digital to better compete with samsung electronics in the nand flash memory market. however, this merger negotiation was stalled due to the opposition of sk hynix. the reason is nothing more than the merged company it will pose a threat to hynix's nand business.

in the second half of this year, there was news in the industry that kioxia was preparing to conduct an ipo in october, which was expected to become the largest ipo after softbank. however, recently kioxia canceled this plan. it is understood that bain capital has been aiming to set the market valuation of kioxia at 1.5 trillion yen ($10.39 billion), but the recent sharp drop in the share prices of listed peers has made this valuation target difficult.

even if kioxia can successfully go public, it will lose the opportunity to merge with western digital. it will be difficult to compete with samsung and hynix in both market and technology. this is a dilemma.

western digital

western digital, which wants to merge with kioxia, is in an even worse situation.

western digital has been studying a split since last year. in the future, it will be split into two different business units, one producing hard disk drives (hdd) and the other producing nand flash memory and ssd. current western digital ceo david goeckeler will be responsible for the nand and ssd businesses, while irving tan, executive vice president of global operations, will become ceo of the hdd business.

wedbush analyst matt bryson said he calculated that western digital's nand and ssd businesses (based on its $19 billion acquisition of sandisk nine years ago) could be worth $10 billion to $23 billion as stand-alone businesses between. he told subscribers: "our analysis shows that western digital's hdd segment is worth almost all of the company's total valuation (just over $20 billion), while sandisk's valuation is actually close to zero.

"our sum-of-the-parts valuation indicates that the market simply does not value western digital's nand business." this can be seen if you compare the annual revenue and market capitalization of western digital and seagate. seagate's business is mainly hard drives, with revenue of us$6.6 billion in the last fiscal year. its current market capitalization is $21.4 billion. western digital's revenue in its most recent fiscal year was $13 billion, twice that of seagate, and its market capitalization was $21.9 billion, almost the same as seagate's. so half of western digital's business is effectively worthless compared to seagate.

of course, in fact, the nand business cannot be completely worthless, but this also proves to a certain extent why western digital wants to split up, because the hdd business has been dragged down by nand to a considerable extent, and after the split, they went in different directions. hello, i’m hello, everyone.

although western digital's performance is not too bad due to the wave of artificial intelligence, the split is already a certainty. perhaps merging with kioxia is still the best solution, and the nand market will eventually move towards grouping for warmth.

write at the end

for the storage industry, the fluctuation cycle is an inevitable disaster. those who can survive it can gain more market share, while those who cannot survive it can only go bankrupt and liquidate and wait for the price to sell.

for giants, operating dram and nand at the same time can obtain higher revenue, but it also bears higher financial pressure. while focusing on nand or dram can breathe a sigh of relief in terms of expenditure, it will also weaken it in disguise. the ability to resist risks.

who can be happy until the end?