Samsung Backstabs The DRAM Cartel – Capital Expenditure Cuts, 130 Manufacturing Projects, 60 Companies Report
Samsung swims against the tide as the industry cuts capital expenditures and delays manufacturing ramps
As the world heads into a recession, semiconductor companies have been thrust into uncertainty. Uncertainty leads to change, and foundries and integrated device manufacturers (IDMs) alike have found themselves caught in the crossfire. Capital expenditures have been in turmoil as companies rush to cut buildout plans, and this will have many impacts on the industry into the near future. In this report, we will analyze what these changes mean for the industry. In particular, Samsung seems to be in a Mexican standoff with their competitors. Samsung seems like they are looking for a pyrrhic market share victory.
Capital expenditure is a good gauge for future growth. It shows that companies are willing to spend in order to grow their market share and increase volumes. However, fabs are complex multi-billion dollar investments that take many years. If the environment gets rocky, cutting down on these just makes sense. Little known is Rock’s Law, sometimes referred to as Moore’s Second Law, states that the cost of a fab doubles every four years. This has limited many players’ abilities to advance to the latest nodes, with many players dropping out, most recently GlobalFoundries and UMC.
Fabs require extremely high utilization to just break even on the immense capital investment. If the market is facing even a minor down-cycle, it’s likely that many fab ventures become unprofitable. After all, every tiny bit of optimisation is essential to their profitability.
Previously, we have shared our tracking sheet of 106 capital projects by semiconductor companies. In the subscriber-only section, we have an update on it. The updated data tracks $420B of expansion projects across 130 sites, 70 fabs, 12 advanced packaging facilities, 17 assembly & testing facilities, 18 wafer substrate plants, 3 organic substrate plants and 10 R&D facilities. It also contains the historical annual capital expenditure of 59 public companies, with 58 of them on a quarterly basis.
A quick summary of the biggest players is shown here, but we will discuss more than these firms.
Samsung – $SSNLF
As the world’s largest DRAM and NAND manufacturer, Samsung has historically had a very high level of CapEx. In 2022, Samsung has record CapEx in Korean Won. However, weak exchange rates have left it relatively lackluster in US$, even dropping a small amount year-on-year. In its latest Q3 earnings call, Samsung finally announced its long-awaited CapEx Guidance for 2022, ₩47.7 Trillion, or roughly $37.5 Billion. This puts it on the higher end of previous estimates of between ₩37.5 and ₩49 Trillion. Samsung’s guidance calls for a gargantuan expenditure in Q4, almost double that of Q2 or Q3. Although Samsung has historically loaded its CapEx in the last quarter, but this seasonality is larger than any of the previous instances.
We believe most of Samsung’s spending is on memory. In 2023, Samsung is moving its entire flagship smartphone SoC to Qualcomm’s S8G2 built on TSMC N4. With Qualcomm and Nvidia moving large portions of the product stack to TSMC, Samsung Exynos flagship cut, and Google Tensor Pixel sales lackluster, there aren’t really logic customers for them to spend on.
With regards to memory, Samsung is spending a decent bit on both NAND and DRAM. They have cut wafer starts somewhat for both these fields, but nowhere close to as much as their competitors. In NAND, Samsung is finally ramping their 176L, which will catch them up to Kioxia and WDC, but leave them behind Micron, YMTC, and SK Hynix newest nodes
Samsung’s spending seems more extreme on DRAM. The DRAM market is a bit of an oligopoly where Samsung, Micron, and SK Hynix have moved their Capex in the union over the last ~7 years with very little market share shift. This now ends.
The results will be pure carnage. Samsung will finally ramp their 1A DRAM node, whereas their highest volume node is 1X, which will increase supply massively, at a time when there is already large oversupply. The oligopoly is broken as Samsung’s new chairman wants to eat market share.
Furthermore, Samsung smells blood in the water, if they can gain a few points of share from Micron and SK Hynix in this downturn plus drive consolidation in the NAND market with Western Digital and Kioxia already in turmoil, they could notch a permanent reduction in the number of competitors. The US government blocked YMTC and CXMT’s rise in NAND and DRAM, and Samsung wants to pick up all their slack
While intense spending may help Samsung regain its footing in memory, its cultural issues will still hamper such efforts, especially on logic.
TSMC – TSM 0.00
TSMC is the world’s largest and most profitable logic foundry. They have unprecedented insight into the purchases and demands of their customers. This allows them to predict future demand more accurately. In Q1, TSMC announced that they would be spending $40-44 Billion for the full year, which we covered earlier this year. In Q2, they announced that their CapEx would be closer to the bottom end of the estimate in Q1, at around $40 Billion. This was still a very substantial amount, over 30% higher than in 2021. This showed TSMC’s commitment to spend as much as it needed to maintain leadership.
As we first reported, weeks before TSMC’s announcement, utilization rates of TSMC’s golden nodes, N7 and N6 plummeted. Some of TSMC’s biggest customers like AMD and MediaTek migrated to the newer N5 family of nodes. Furthermore, the recession has began to strike its customers. TSMC, too, realized this, and in their Q3 call, dropped their guidance a 2nd time, now to $36 Billion.
The reasons cited were the utilization and delays in equipment delivery. As stated previously, we expect this to impact fab buildout quite significantly, especially Fab 22, which has it’s N7/N6 buildout cancelled. In the first 3 quarters, they have dropped their overall guidance by about $6 Billion. It is worrying for the rest of the industry that even TSMC, with its limitless insights, failed to see this coming. We expect CapEx to drop significantly in the near-term while the industry recalibrates.
SMIC – $980.HK
SMIC’s CapEx changes, while not the largest, are certainly the most interesting for Q3. They had initially guided 2022 CapEx up about 10% year-on-year to $5 Billion. However, in Q3, China and the USA officially waged economic war. As China’s national semiconductor champion, this has undoubtedly proven existential to SMIC. SMIC raised its guidance for the year to $6.6 Billion, or a 50% rise over the previous year. This was despite them guiding Q4 revenue down by 15% in anticipation of export restrictions and weak macroeconomic factors. SMIC is desperate to spend and get as much as it can before even more restrictions are imposed.
SMIC has been heavily reliant on US technologies for manufacturing. Tightening restrictions mean it is feeling the crunch as it is unable to receive machines and parts for maintenance from a large portion of its supplier base. SMIC would need to redevelop all their technologies to avoid the heavy lock-in. Even if SMIC was content with abandoning their current expansion projects in Shenzhen, Beijing and Shanghai, their machines would inevitably stop working without maintenance and spare parts. Of course, this assumes the China restrictions are actually enforced, but based on what we have heard on the ground, it seems enforcement is very light.
SemiAnalysis is an ad-free reader-supported publication. To receive new posts and support our work, consider becoming a subscriber.
This bump in CapEx is a clear attempt to hoard as much as they can before restrictions are tightened even further. In an economic war, every small advantage matters. Only time will tell if this saves SMIC. Will Chinese equipment manufacturers like SMEE and Naura catch up? Or will SMIC fade into irrelevance?
Next we will break down Intel, the other Tier 2 foundries, and the other 2 major DRAM players and also we share our dataset.