Latest Post

Films from 2024 give us a sense of déjà vu: Do recycled films really work? Woman stabbed “multiple times” and almost thrown from balcony at Miami International Airport; man arrested
Citywire Selector | The secret tech investor: Memory chip makers are having their moment

Anyone who has been fortunate enough to read Chris Miller’s Chip War knows that the memory chip sector is notorious for its cyclicality and brutal downturns.

From the 1970s to the 2000s, the storage industry was one of the fastest growing sectors in the world, with growth rates of up to 70% per year.

And although memory chip manufacturers have created incalculable value for customers and technological progress for society, McKinsey estimates that between 1997 and 2012 they destroyed a total of $9.5 billion in investor assets over several boom and bust cycles.

But this time – yes, really – it could be different, as rising demand for ever more powerful AI hardware and software is fueling a bull market for memory chips, with sell-side analysts now revising their forecasts.

Boom and bust

But first a brief review: How did this destruction of capital come about?

The main culprit was the cyclical dynamics that resulted from the nature of the memory chip itself. For all their technical complexity and importance to the digital age, memory chips are practically a commodity product. Unlike logic or AI chips (the “brains” that perform computational tasks), memory simply stores data.

There are two types of memory chips.

Dynamic random access memory (DRAM) stores the data that is actively accessed by a device’s operating system, while non-volatile memory (NAND) chips, used for longer-term storage, are available in a relatively limited variety of configurations and vary little from manufacturer to manufacturer.

Price is often used as a weapon and South Korean players, especially Samsung, have a reputation for investing throughout the economic cycle, even in times of low utilization.

However, a look at the factsheets of the major UK technology investment funds shows that Polar Capital Technology, Allianz Technology and Manchester & London Micron Technology are among the top 10 holdings. In addition, M&L and Allianz both own Western Digital, with Allianz ranking it among the top 10 stocks.

Beat for the cycle

This is interesting because technology fund managers are more likely to be attracted to predictable, annuity-like revenue streams rather than joining a cyclical rollercoaster ride. It is also rare for these three portfolios to overlap. Moreover, it is odd that none of them have Korean company Samsung or SK Hynix – the first and second players in this market – in their top 10.

What’s going on? And why are these three fund managers suddenly attracted to this sector?

There is no doubt that the industry is now more rational, as the reduction in the number of players to the three large corporations – the two Koreans and Micron – means a more oligopolistic market structure.

Proof of this is that these three have publicly stated that they will remain disciplined on capacity and price. In other words, the Koreans have decided that it is better to each own 40% of a profitable industry with higher margins than to completely dominate an industry with low or even negative margins.

In addition, the memory industry has just come out of a post-Covid downturn, so we may now be entering the upswing phase of the cycle. Prices have recovered since the bottom of the last cycle, with spot prices for NAND and DRAM now said to have risen to 191% and 26%, respectively.

Profiling growth

While these two points make investing in this industry more attractive, I suspect the real reason for these investors’ involvement is the changing growth profile of this subsector.

In the past, memory has been used in PCs and servers, but in the future there could be a huge need for memory from AI in all formats – server, edge and mobile. This has led to sell-side analysts raising their memory growth forecasts to 15-20% per year for the next five years and starting to talk about long-term rather than cyclical growth in the subsector.

Memory is even more important for AI chips – graphics processing units (GPUs) and specialized AI accelerators – than for the simpler central processing units (CPUs) for several important reasons:

  • While CPUs have only a handful of cores, AI chips have a parallel architecture with thousands of cores designed to perform matrix multiplications in parallel on large data sets. This parallel processing requires extremely high memory bandwidth to feed all cores with data.
  • AI workloads, especially training large neural networks, require storing and quickly accessing massive amounts of data. Memory bandwidth and capacity requirements are orders of magnitude higher than traditional CPU workloads – the Nvidia A100 GPU chip can utilize 2 terabytes per second of memory bandwidth, more than 20 times the 90 gigabyte requirement of a high-end CPU.
  • Today’s largest AI models have billions or even trillions of parameters that must be stored in memory during training. Attempting to train these massive models on CPUs with limited memory bandwidth would be prohibitively slow.
  • The memory subsystem accounts for a significant portion of the chip area, power consumption and cost of modern AI accelerators, reflecting its critical importance. Optimizing memory bandwidth and capacity is a key design priority to reduce power consumption and is therefore of paramount importance to the total cost of ownership of the semiconductor.

Now the sell side is waking up to this drastic change in demand. Micron shares rose 5.5% today to a record high of $156 after brokers Cantor Fitzgerald and Susquehanna Financial raised their price targets yesterday, citing price trends for DRAM and NAND memory. The upgrades also sent SK Hynix shares to a record high in Seoul.

A higher level

Then we have the rise of High Bandwidth Memory (HBM) with even higher bandwidth, capacity and performance at lower power consumption by vertically stacking up to 12 DRAM memory chips to shorten the data transfer distance while enabling smaller form factors.

HBM is increasingly being used for machine learning, high-performance computing, and more recently, generative AI models, causing memory manufacturers to reallocate more than 20% of total DRAM wafer supply here. This has led analysts to forecast double-digit price increases in both second-half quarters of this year and further increases in 2025.

In other words: things can hardly get better for the “big three” of memory.

The fact that these investment funds prefer the Idaho-based company Micron to the Koreans could be due to geopolitical tensions and the US government’s efforts to strengthen the domestic semiconductor industry.

So if you feel like you missed the AI ​​chip boom (which we doubt), this rebound is at an earlier stage and with better valuation metrics.

Yes, Micron shares may have doubled in value in the last year, but they still trade at 10 times EBITDA compared to Nvidia’s 35 times EBITDA.

The Secret Tech Investor is an experienced professional who has been managing technology stocks for more than 20 years.

Leave a Reply

Your email address will not be published. Required fields are marked *