For decades the business of making digital memory chips has been among the most volatile within the IT sector, with periods of rising demand, prices, and profits inevitably followed by collapse. These recurring booms and busts were, as such cycles always are, rooted in the nature of the product and the nature of competition among its producers. However, according to Igor Tishin, an IT analyst at Harding Loevner, there is reason to think that industry competition has evolved to a new state of balance, where cyclicality has moderated and profitability can be higher and steadier.
With each successive leap in computing power, memory manufacturers like Idaho-based Micron have periodically needed to make substantial investments in new processes and equipment. Here, Micron silicon dies, having already been imprinted with transistors and circuitry, have their micro circuitry wired for connectivity right before the unit is packaged into a finished “chip.”
Source: Crucial.com, Micron Technology, Inc.
The coronavirus appears to have demand for memory at a new cyclical inflection point. Unemployment and falling household and corporate income, like that which has occurred during the COVID-19 recession, can dent demand for computers, smartphones, and other devices containing memory chips. But the pandemic, by accelerating changes in how people live and work, is offsetting the impact of those traditional demand factors. Remote working, online learning, and streaming entertainment increase the need for not only personal devices but also cloud-computing services, whose data centers are vast users of memory chips. “There is no shortage of positive or negative commentary on memory these days,” Tishin says. “All the commentators are using harbingers of demand recovery or decline as signs of the next turn in the cycle.”
But while recovering demand would be a promising signal of memory chipmakers’ earnings prospects in the near term, Tishin says, “Even this signal lacks durability.” What the pundits are overlooking, he thinks, are signs of an important—and potentially more durable—development on the supply side of the market: the emergence among makers of a new restraint with respect to manufacturing capacity and product pricing. Industry-wide discipline on these dimensions—while tricky to achieve in practice—is crucial for sustained profitability in making a commoditized product like digital memory.
A History of Ups and Downs—and Losses
From the 1970s through the 2000s, the memory industry was among the world’s fastest growing, with growth rates reaching up to 70% a year. Yet, despite creating incalculably large value for society, from 1997 to 2012 memory-chip manufacturers collectively destroyed US$9.5 billion of investors’ capital across multiple boom-bust cycles, according to the consulting firm McKinsey & Company.
From 1997 to 2012, memory-chip manufacturers collectively destroyed US$9.5 billion of investors’ capital across multiple boom-bust cycles.
How did this capital destruction come about? The key culprit was the cyclical dynamics that sprang from the nature of the memory chip itself. For all its technical intricacy and criticality to the digital age, memory, like wheat or airline seats, is effectively a commodity.¹ Unlike logic chips (the “brains” that carry out computational tasks), memory merely stores data. Dynamic random-access memory (DRAM), which stores the data actively drawn upon by a device’s operating system, and NotAnd (NAND) chips for longer-term storage, come in a relatively limited variety of configurations, with few differences from one producer’s chip to the next.
Prior to the 2000s, memory chips were used primarily in servers and PCs. The small number of dominant PC and server makers exercised tremendous bargaining power over the dozens of companies producing memory chips. Faced with stiff competition themselves, computer makers such as Hewlett-Packard and Dell tried to keep a lid on memory prices by continually threatening to switch suppliers, and chipmakers routinely capitulated. Moreover, while these chips are relatively simple in function, they are not easy to make or keep up to date. The enormous capital required to make memory chips and short product lifecycles thus sapped chipmakers’ profits. Over time, continuous innovation resulted in more powerful and less expensive chips, making possible ever-more powerful and cheaper electronics. But it also caused capital expenditures to skyrocket. In the early 1990s, a new fabrication plant capable of producing the most advanced memory chips cost about US$300 million. By 2016, the price tag had risen to US$10 billion.
A Micron silicon wafer undergoes part of the fabrication process, after being imprinted with memory transistors and circuitry, and before being diced up into dies, tested, wired, and packaged.
Source: Crucial.com, Micron Technology, Inc.
Chipmakers also had to contend with sharp and unpredictable shifts in demand brought on by the reverberation of business cycles and the uneven pace at which new electronic devices are adopted. Lacking any real ability to differentiate their products, when demand for memory fell, most chipmakers responded the way sellers must in any commoditized market: they accepted whatever price they could get and kept their production humming to continue to squeeze some value out of their equipment. A handful of leading producers increased production and cut prices even more aggressively, apparently figuring, as Tishin says, “they might as well get some mileage out of the downturns by making things even worse for weaker players with leveraged balance sheets in the hope that they don’t survive.”
Eventually, the unbridled competition drove even stronger players to desperation. From 1999 through 2002 nine companies—including South Korea’s Samsung and SK Hynix, Germany’s Infineon Technologies, Japan’s Elpida, and the US’s Micron—colluded to fix the price of memory chips, according to plea agreements later reached with US and EU prosecutors. Micron escaped serious penalties in return for its cooperation with authorities, but fines levied against the others totaled US$730 million, and a number of people, including Samsung’s senior executive for DRAM, went to prison. “It shows just how brutal the industry was,” says Tishin. “These people were willing to risk everything to get themselves some relief.”
An End to Cycles?
In the first decade of the new century, the memory industry underwent upheavals that reduced the number of chipmakers, allowing hopes for a more orderly industry finally to come to pass. In the wake of the dotcom crash of 2000-02, the formerly prominent memory operations of Texas Instruments, NEC, and Hitachi all fell victim to mergers or bankruptcy. A passel of new manufacturers emerged to capture a share of the new demand driven by smart phones, but the global financial crisis soon wiped them out along with a few former survivors, including Infineon’s memory subsidiary. After 2012, when Micron bought Elpida out of receivership, only three companies were left controlling over 90% of the global DRAM market.²
Atop the triumvirate was Samsung. The South Korean conglomerate had spent years of heavy investments in R&D to supply the memory for its vast line of electronic devices, such as mobile phones, tablets, TVs, and consumer electronics, leaving it with a business that by 2012 accounted for over a quarter of its annual revenues, and the largest global market share (roughly 45%). Next was Hynix, a company entirely devoted to making memory chips, followed by Micron, another dedicated memory manufacturer, with 20% share.
Source: Alliance Bernstein
The industry’s customer profile was also evolving. The market no longer revolved around servers and PCs alone. In addition to smart phones, smart appliances and automobiles were produced by many customers, diluting the bargaining power of any one. Entirely new sources of memory demand were coming online quickly, including AI, big data, and the internet of things (IOT).
The industry experienced extraordinary growth starting in late 2016, with DRAM chip prices and profits overall rising for nine consecutive quarters amid strengthening demand. The chipmakers extraordinary profitability gave rise to new allegations of price fixing, with the filing of a class-action lawsuit in the US and news reports that the recent price increases had caught the attention of Chinese authorities. Then, at the end of 2018, with the global economy weakening, chip inventories began to swell, and prices collapsed yet again. By the end of 2019, memory chip prices had fallen 33%, the most since the dotcom bust, and the fourth-quarter profits of the three leading memory chipmakers dropped by as much as 50% year over year. As of that fall, a federal judge in California threw out most of the claims in the suit for lack of evidence, and the Chinese had failed to press their allegations further.
Heading into 2020, evidence mounted that demand for memory was rebounding, thanks to spending at data centers and increased interest in smartphones designed to capitalize on emerging 5G, or fifth-generation, cellular networks. Still, Tishin was unimpressed. “These signs were positive, but they did not form a solid investment thesis,” he says. “The reason is, nobody, including the chip makers themselves, knew for sure what would happen with demand.”
Rather, Tishin has been more focused on another shift. During the most recent downturn, memory chipmakers for the first time actually reduced capacity and kept prices from a freefall, this time without crossing the line in the eyes of now-watchful authorities. And, while their profits plunged, none of the three lost money. Memory was still a commodity, but it was starting to act like a better-behaved one. “If these dynamics hold,” Tishin says, “they could make memory an attractive area for long-term investment, regardless of how the cycles play out.”
A New Discipline?
When an industry has many participants, it is all but impossible to cooperate on prices except through collusion. But when only a handful are involved, they can (though don’t always) through their actions and inactions reach an implicit understanding that it is in their mutual interest to hold the line on prices and production. In game theory, this is known as a Cournot competition, and is a good description of the state of the memory market today. “You get a good sense of these dynamics on their earnings calls,” Tishin says. “During a call, Samsung, for example, would be discussing its expectations for demand in the next quarter or two and how it is planning to provision supply, and how they are not trying to gain share with excessive supply. We have three players now, each with positive cash flow and very high positive cash balances and virtually no debt, who all seem to understand that each of the other two is perfectly capable of surviving even the most vicious downturn. And part of keeping the peace, I think, is just periodically reassuring one another through the forums legally available to them that they are no longer out for blood.”
“Part of keeping the peace,” says Harding Loevner’s Igor Tishin, involves the three remaining dominant players “periodically reassuring one another through the forums legally available to them that they are no longer out for blood.”
The economics of memory production have changed as well. Much of this has to do with a slowing pace of innovation. There are two ways to increase memory production. As recently as five years ago, most of the increases came from “shrink”—that is, from using finer manufacturing technologies to cram more memory bits (short for binary digits) onto a unit of silicon surface, a relatively cost-effective improvement. But “shrink” has finally reached the point where each doubling of memory density takes longer.³ Now, to maintain the rate of increase in output, manufacturers must also use more silicon at the cost of substantial capital investment, rather than simply put more “bits” onto each unit. While any player could still make a grab for larger market share, their incentive to do so has been reduced.
Source: Alliance Bernstein
Finally, the rate of growth in demand for memory has shown signs of slowing, as it now occurs off a much larger base. With sales projected at US$140 billion in 2020, memory is now the largest part of the US$450 billion semiconductor market. Even with transformative developments such as 5G and IOT continuing to drive the appetite for additional DRAM, most analysts, including Tishin, believe memory demand simply can no longer grow by 70% annually. Long-term growth rates around the 20% or so that has been projected for 2020 now seem more realistic, Tishin says.
Ironically, this slowdown, too, he says, should have a constructive effect on profitability. Greater predictability of demand should make it easier for Samsung, Hynix, and Micron to plan accurately for capacity expansion, reducing the risk of their overshooting the mark and then having to slash prices. “Compared to other industries, these are still very healthy rates of growth,” Tishin says. “Are they more boring than the potential for a burst of 70% growth? Yes, and that’s the point. As long-term investors, we are less interested in trying to time the cycle than we are in finding companies that can sustain and compound healthy levels of growth and profitability over long stretches of time. These are the kinds of dynamics we look for in an industry. Boring is how we like to make money.”
Analyst Igor Tishin, PhD, contributed research and viewpoints to this piece.
¹So close are the parallels between memory and more traditional commodities that average memory chip prices are described in terms of “spot” prices, and at one point in the early 2000s, the now-infamous energy company turned trading emporium Enron ran a fledgling memory chip futures desk.
³When Intel founder Gordon Moore famously observed in 1965 in what became known as “Moore’s Law” that the number of transistors on a microchip would roughly double every two years, he was speaking about logic chips, but the number of bits on a memory chip has shown essentially the same rate of growth over the years.