Most industry analysts and pundits focus on foundry/logic as the driver of the semiconductor industry because they are on the bleeding edge of Moore’s Law pushing the boundries of transistor shrinkage.
Many in the industry, including investors, have missed the fact that the industry has returned to its origins as a memory centric industry and that this massive shift has had many far reaching impacts on Moore’s Law, profitability, valuations, capital spending, market share and who is going to dominate the industry going forward.
The “gestalt” of the semiconductor industry has undergone a fundamental change as its component parts of NAND, DRAM, foundry and logic have changed in relative value and importance. Industry importance and therefore value cannot be assigned solely based on where a company or product is on Moore’s Law roadmap.
What are the new valuation drivers and where should investors and the industry put their money?……
Follow the Money….
If we use capex as a proxy of where semiconductor manufacturers are spending their money and placing their bets, its clearly the memory industry and specifically 3D NAND. Spending on 3D NAND has been the primary driving force behind the current long term upswing in semiconductor equipment.
While its true that DRAM spending has picked up again after a drought, this is more of a recent phenomenon. Spending on logic has been down over the long run, just look at Intel and the stretching out of capex and delays of 10NM. Intel is the poster child of this shift as its clear that Intel is spending an ever increasing percentage of its capex on memory and if we were able to split out logic is highly likely to continue to be down. Intel’s Dalian fab is the prime example here, even the delayed & under delivered XPoint points to more memory spend.
If we look at Samsung, the most profitable part of all of Samsung is semiconductors and Samsung is the leader in 3D NAND and continues to double down by spending oodles of money in the process of trying to bury the competition. When we look at The myriad of Chinese fab projects, the one that sticks out is a monster 3D NAND project which is rumored to be similar in size to a Samsung fab (something like $8B worth)
Infinite 3D NAND price elasticity drives the market and capex
As we have said many times, the 3D NAND market is almost a textbook example of a price elastic market. As pricing continues to drop through lower manufacturing costs enabled by 3D, demand increases to suck up any excess capacity and keep us in good supply demand balance (and a continued slight shortage…). Solid State drives are the bottomless pits of demand along with increasing storage on mobile devices. This is unlike the relatively price inelastic DRAM market where most mobile devices have a couple of gigs of DRAM because thats what the OS needs no matter what the price. We see no real slowing of NAND demand as long as pricing continues its long term technology driven downward trajectory….there are billions of rotating magnetic drives to displace.
Logic, CPUs are headed in the opposite direction, down…
Its no secret that Intels processor business is in a long term decline not only due to the slowing of the PC industry but also due to lower value of the CPU’s function. Device manufacturers, such as Apple & Samsung, make their own, lower cost, lower value CPUs. ARM, NVIDIA & QCOM are out to replace CPU’s in servers and other devices.
AMD recently rejoined this “race to the bottom” as it came out with a competitive series of processors at half of Intels values forcing a price lowering response from Intel. Its little wonder that Intel is slow on 10NM and slowing spend on CPU capacity.
Samsung passing Intel to be #1 in semis
Just in case you didn’t believe there is a monumental shift in the semi industry, you only need to wake up and look at Samsung versus Intel in semi sales. Samsung is well on track to surpass Intel as the world’s biggest maker of semi’s in a quarter or two (maybe even the June quarter). If you look at the revenue charts, the curves look like a “X” and when they cross, there’s no going back…its over, done.
Even though Samsung has foundry and logic, its memory thats driving Samsung’s semi train, and 3D NAND is the engine. The number one semi company will again be a memory centric company. This is obviously ironic for those who remember that Intel was founded in 1968 as a maker of SRAM and DRAM and even though it developed the first microprocessor in 1971 it wasn’t until 1981 that microprocessors became the majority of business. Intel eventually dropped out of the DRAM business, along with 5 other U.S. DRAM makers leaving Micron as the sole survivor on the island.
Moore’s Law was always about transistors per square inch not gate pitch/nanometers
Maybe we need to change the definition to transistors per cubic inch due to 3D….
The industry has always focused on gate pitch , measured in microns and now nanometers since the beginning of the semi industry, as the definition of Moore’s Law and who was winning the race. Moore’s Law has always been more of an economic law of cost per transistor not transistor size. The reason that transistor size mattered is that wafer real estate had been the primary driver of cost.
Transistor size also matters to CPU speed and power consumption but that doesn’t matter as much in memory devices, which are becoming the dominant piece of the pie.
Now we can get more transistors per square (cubic) inch through 3D stacking of transistors so we don’t need gate pitch improvements as much as we used to, especially in memory. We get the same or better cost effect of Moore’s Law by going vertical rather than going smaller.
Many investors and industry insiders have missed this subtle but monumental disturbance in the force of the semi industry as they remained transfixed on nanometer measurements measuring gate pitch as the yardstick of the industry.
The impact of Moore’s Law 3D transistor density versus gate pitch density on capex
The shift of the industry’s focus from shrinking gate pitches to achieve density increases to 3D stacking to achieve density goals has clearly shown up in equipment sales. The primary driver of gate pitch increases is lithography and litho equipment sales have lagged the industry average over the last few years. Etch and deposition are the primary tools to add more layers (ie 3D) and have grown faster than average over the last few years and have exploded as 3D NAND has exploded.. The shift to 3D NAND has had a direct impact on shifting more dollars to LRCX and AMAT and away from ASML (part of this is self inflicted as the delay of EUV has delayed gate pitch improvements). KLAC which historically had been logic and foundry centric is now more evenly balanced between memory and logic as memory yield and complexity gets worse.
Thus the shift to a memory (3D NAND) centric semi industry has also shifted the fortunes of the equipment industry very significantly.
Memory’s new valuation shows up in Toshiba sale process
Who would have thought that Toshiba’s semiconductor operations which manufacture memory devices, primarily NAND, could be worth a reported $27B. A few short years ago you could buy defunct memory operations for pennies on the dollar for very good equipment and facilities. Micron had a very good business model of picking up assets on the cheap at the bottom of every memory cycle when someone always went out of business, most recently Inotera.
Now we are talking about Toshiba selling its memory business for something like $27B which is enough money to buy 3 , large, bleeding edge 3D NAND fabs with some spare change left over (and Toshiba doesn’t have 3 big new fabs so the business has significant enterprise value). If Toshiba is worth $27B to an industry buyer then Micron is cheap at at its current valuation of $31B to public investors. This perhaps also begs the question if Intel is really worth $172B even if we account for the non semi business they have and are buying into.
Does 3D NAND bring with it a longer up cycle??
Given the long and deep seated demand for NAND it appears we have a relatively long and strong cycle. At some point supply could overwhelm the industry’s ability to absorb all the capacity eventually coming on line but we think the elasticity will dampen most oversupply and put a bit of a floor in pricing, certainly more than the bottom dropping out that DRAM sees.
Exactly how long is anybody’s guess but we are already in a long up turn. We don’t see a crash in 3D NAND pricing given this elasticity but we could see an amplified impact on the equipment market as it is a typical second order derivative market that over reacts as compared to the semi market it serves.
Right now we don’t see anything on long range radar.
Investors need to look at the impact of the shift in Moore’s Law as the semi industry shifts to a memory centric model and as the torch of leadership is passed to a memory driven company. We need to rethink the value of memory companys and products. We also need to look at and invest in the longer term trends as we see the Chinese focusing on memory in the semi industry for the long run. We can easily imagine the Chinese doing to Samsung what Samsung is about to do to Intel.
We need to look at the second and third order derivative impact on semi equipment companies and their suppliers as we see the continued shift in Moore’s Law and continued shift towards memory.