The delivery lead time of semiconductor equipment has doubled, sending huge ripples across the entire industry.
The semiconductor industry has once again entered an era of "waiting for equipment". In late July, the main equipment delivery lead time of the five major equipment suppliers, namely Applied Materials, ASML, Lam Research, Tokyo Electron and KLA, has been extended to 1.5 to 2 times the original level, and equipment that normally takes six months to deliver now requires a wait of about one year.
South Korean local equipment manufacturers are in a similar situation: a company that supplies front-end and back-end equipment to TSMC and Micron has seen its delivery lead time extended from 3-4 months to 6-8 months, with individual categories exceeding one year. The procurement departments of Samsung Electronics and SK Hynix have begun to place orders in advance to lock in future production capacity ahead of schedule.
Delivery lead time is the most sensitive prosperity indicator in the semiconductor industry. The mid-year forecast released by SEMI on July 14 gives a corresponding macro annotation: global semiconductor equipment sales will reach 165.9 billion U.S. dollars in 2026, a year-on-year increase of 23.2%, hitting a record high, and will continue to grow to 229.5 billion U.S. dollars in 2028. However, what deserves more attention than the prosperity itself is that the extended delivery lead time is practically rewriting the industry's cost structure, competition pattern and production capacity schedule, and the bottleneck causing all this is not in the hands of equipment manufacturers themselves.
Extended Delivery Lead Time: What Price Is the Industry Paying
The most direct cost is the systematic increase in chip manufacturing costs. Equipment price hikes and extended delivery lead times occur simultaneously. To lock in production capacity, chip factories have to accept higher equipment prices and longer prepayment cycles, and the capital expenditure per unit of production capacity rises accordingly. TSMC has raised its 2026 capital expenditure to 60-64 billion U.S. dollars. In addition to the shortage of production capacity, another reason is equipment price inflation — this part of "inflation" will eventually be amortized into the depreciation cost of each wafer, and TSMC has accordingly launched a new round of foundry price increases. Amazon also revealed in its Q2 earnings report released on July 31 that the rise in memory prices is one of the reasons for pushing up its capital expenditure expectations, and it has once again raised its full-year capital expenditure to 220 billion U.S. dollars. All these point to a point that needs attention when interpreting the capital expenditure data of various companies: the nominal growth rate contains price moisture, and the actual production capacity purchased is not as much as the figures show.
In addition to costs, the overall schedule for production capacity release is moving backward. TrendForce judges that the mass production time points of this round of newly built wafer fabs mostly fall between the second half of 2027 and 2028, and the situation of insufficient DRAM supply will be difficult to reverse before that. SK Hynix CEO Kwak No-Jung expects 2027 to be the tightest year in supply in the history of the memory industry. Amazon CEO Andy Jassy said bluntly that even with such a large expenditure scale, the company still cannot meet all the demand in 2026, the state of supply shortage will continue until 2027, and the demand in 2028 will be staggering. If the equipment delivery lead time continues to extend, there is a risk that this schedule will be further delayed. The slower the supply arrives, the longer the shortage lasts. This is the underlying basis for Goldman Sachs' judgment that the DRAM and NAND supply-demand gap in 2026 will hit a new high since 2011, and the supply shortage may continue until after 2027.
The backward shift of the schedule is reshaping the competition pattern. The sequence of equipment delivery directly determines the sequence of production capacity implementation. In the memory market where prices are rising rapidly, there is a huge difference in returns between starting production half a year earlier and half a year later — the DRAM contract price in Q3 2026 is expected to rise by another 13% to 18% month-on-month, Samsung plans to raise prices for the third time within the year, and all of SK Hynix's HBM production capacity this year has been sold out. Manufacturers that have grabbed equipment at present will release new production capacity exactly at the window of the tightest supply and highest price; those that fail to grab equipment may miss the price peak when their production capacity is put into use. Head manufacturers with abundant capital that can lock positions in advance thus gain first-mover advantages, while second-tier chip factories and small and medium-sized equipment manufacturers are squeezed at both the equipment and component levels. The shortage cycle has always been an accelerator for the strong to get stronger. Samsung Electronics' Q2 earnings report released on July 30 shows that its semiconductor division's profit contribution rate exceeds 99%, HBM4 has been shipped on a large scale and HBM4E samples have begun to be delivered, and the revenue proportion of server memory chips has hit a record high, further consolidating its leading position in the AI memory field.
The end of the chain is consumers. Memory original manufacturers tilt their advanced production capacity to high-margin products such as HBM, the supply of general-purpose DRAM and NAND has shrunk and prices have continued to rise, and the cost of mid-range domestic mobile phone models has generally increased by 300 to 800 yuan. The price increase chain of "components — equipment — production capacity — chips — complete machines" is ultimately shared by the end market.
The Reason Why Equipment Cannot Be Manufactured Is That Components Cannot Be Manufactured
To understand why these costs are difficult to eliminate in the short term, we need to figure out where the bottleneck is stuck. A counter-intuitive fact is that equipment manufacturers are not unprepared: ASML has given a clear capacity expansion roadmap. Based on the production capacity plan of about 65 Low-NA EUV units in 2026, it will increase by 30% in 2027, and study another 30% increase in 2028; for immersive DUV, based on about 130 units in 2026, capacity will be expanded at a rate of 30% per year, which basically corresponds to doubling in three years. The production scheduling of Applied Materials and Lam Research is also at full load. The whole equipment link is not short of willingness to expand production or capital, what it lacks is its own "raw materials".
Semiconductor equipment is a typical long-chain precision manufacturing. An etcher or thin-film deposition equipment is composed of thousands to tens of thousands of components, among which the supply of a number of high-precision parts is highly concentrated: ultra-high vacuum valves are almost monopolized by Switzerland's VAT, with a high-end market share of over 90%; the mass flow controller (MFC) market is occupied by the top five manufacturers such as Japan's HORIBA and the U.S.'s MKS, accounting for about 85% of the share; dry vacuum pumps are concentrated in the hands of Japan's Ebara, the UK's Edwards and Germany's Pfeiffer; for the precision ceramic structural parts necessary for etching and CVD equipment, their high-end aluminum nitride powder and ultra-precision processing have long been controlled by Japanese companies such as Kyocera and Sumitomo. These links have small market size, extremely high technical thresholds and very few players. If any link cuts off supply, the complete equipment cannot be delivered.
It is precisely these links that snapped first. According to industry research, the market originally expected the component shortage to appear at the end of this year, but since April this year, many component manufacturers have received notices of extended delivery lead times from overseas suppliers, and some categories are directly out of stock. The testing equipment supply chain provides more granular evidence: according to South Korean media reports, the delivery lead time of FPGAs required for testing equipment has been extended from 8-10 weeks to a maximum of 52 weeks, drive ICs for equipment have to wait at least 10 weeks, and the price of some x86 processors has risen from 1 million won to 3 million won; a South Korean testing equipment manufacturer has signed a supply contract worth over 10 billion won with Samsung Electronics, but has to postpone the delivery period by three months due to component delays. Equipment manufacturers are waiting for components, and chip factories are waiting for equipment, thus forming a complete delay transmission chain.
The poor production capacity elasticity in the component link has structural reasons. The capacity expansion of high-precision components involves complex processes such as materials, heat treatment and ultra-precision machining, and the construction period is generally 2 to 3 years, with slow yield ramping, which is naturally slower than that of complete equipment manufacturers, and even slower than the 1.5 to 2-year construction pace of wafer fabs. The willingness to expand production is also insufficient — most overseas component companies have just experienced the industry downturn around 2023, when they were clearing production capacity, and have doubts about the sustainability of AI demand, so they are unwilling to make large-scale investments for a round of demand that may be falsified. The restocking demand of equipment manufacturers is further amplifying the gap: the component inventory consumed during the downturn needs to be replenished, coupled with safety stock, the actual growth rate of component demand is higher than that of the equipment itself. With the slowest supply, the weakest willingness and the most amplified demand, components have become the most rigid link in the entire industrial chain.
Under the shortage, procurement behavior itself is also pushing up the delivery lead time reading. The more large manufacturers lock orders in advance, the less available production scheduling capacity of equipment manufacturers will be, and the longer the delivery lead time faced by other customers, who will then be forced to place orders in advance. ASML's Q2 earnings report reveals a rare detail: CFO Roger Dassen revealed that EUV orders have been accumulated two years in advance, "a situation we have not encountered for many years". Customers placing orders in advance and paying prepayments for equipment to be delivered two years later means that chip factories are providing financing for the capacity expansion of equipment manufacturers, and the bargaining power is obviously biased towards the equipment side. SK Hynix also revealed at its Q2 performance meeting that the conventional cycle of its long-term supply agreements with core customers is about 5 years, with supporting guarantee mechanisms such as deposits to strengthen performance guarantees and improve the predictability of long-term demand. This also means that part of the current "1.5 to 2 times" delivery lead time reading is caused by pre-positioned locking positions, not all corresponding to real immediate demand — this vigilance is needed when interpreting delivery lead time data.
Deformation of the Cycle
Putting the bottleneck in the framework of the cycle, it is changing the shape of this round of semiconductor upward cycle. The reason why the traditional memory cycle is drastic is that the supply response is fast: once the price rises, original manufacturers increase capital expenditure, equipment is in place and production capacity is launched, supply will be oversupplied within one or two years, and the price falls. But this time, every link of the supply response has been extended — the maximum delivery lead time of components is 18 to 24 months, the equipment delivery lead time doubles, and the construction of wafer fabs takes 1.5 to 2 years. The production capacity decided today will not form effective supply until around 2028. The systematic decline in supply elasticity means that the price rising period will be longer than that of historical cycles, which is the logical source of Citi's aggressive prediction that the average DRAM price will rise by 88% in 2026, and also the basis for some securities researchers to judge that the equipment supply-demand tension may last until 2030.
But looking at the same mechanism the other way around, it is the shape of risk. The slow supply response means that once there is a problem on the demand side, the brake on supply will also be slow. At present, the equipment orders locked by chip factories and the component orders locked by equipment manufacturers are all multi-year commitments made based on the premise that AI infrastructure investment continues to grow rapidly; the production capacity announced by TSMC, Micron, Samsung and SK Hynix will be released intensively in 2027-2028. If the growth rate of AI computing power investment slows down by then, the industry will not smoothly transition from shortage to equilibrium, but will swing directly from extreme shortage to surplus under the bullwhip effect — advance orders amplify demand readings during the upward period, and will be doubled back in the form of order cancellations during the downward period. Every time the equipment delivery lead time doubles in history, it is accompanied by a demand narrative that seemed unshakable at the time; and every time the delivery lead time falls, it is more sudden than the industry expected.
It is worth alerting that even with the "explosive" earnings reports of head manufacturers, the market's concerns about whether the capital expenditure cycle of AI chip investment is approaching its peak are also rising. For example, after Samsung Electronics released its record Q2 earnings report on July 30, its market value has fallen by 40% compared with its peak in June; South Korea's KOSPI index fell by nearly 29% cumulatively in July, hitting a record since the Asian financial crisis, reflecting the market's concerns about the slowdown of the AI capital expenditure cycle. This "buy the rumor, sell the fact" market, and the deviation between performance and stock price, are important signals for interpreting the current cycle stage.
The leading indicator really worth tracking at the moment is not equipment sales, but when the component delivery lead time peaks and when the prepayment terms of equipment manufacturers loosen — the link with the tightest bottleneck has always been the first link to send a signal when the cycle turns. For the moment, the conclusion can be summed up in one sentence: The pricing power of this round of semiconductor shortage is shifting upstream along the industrial chain, from chip factories to equipment manufacturers, and then from equipment manufacturers to those little-known manufacturers of vacuum valves, flow meters and ceramic parts. Whoever controls the delivery lead time controls the rhythm of this cycle.
This article is from the WeChat official account "Semiconductor Industry Review" (ID: ICViews), author: Jun Xi, authorized for release by 36Kr.