Memory Oligopoly's Payday

Programming note: ARPU will be off next week, back on Dec 29.

The Seventh DRAM Cycle

If you have tried to buy computer memory recently, you may have noticed something strange. Prices have gone haywire. A kit of G.Skill 32GB DDR5 RAM that cost $125 in September is now listed for nearly $400. Framework, the modular laptop maker, just jacked up its memory prices by 50% and rewrote its return policy to stop customers from scalping the RAM out of its machines. This isn't just a niche problem for PC builders; it's a supply shock that is rippling through the entire consumer electronics market, with Dell, HP, and Xiaomi all warning that your next laptop or smartphone is about to get more expensive.

The official story is that AI ate your RAM. The insatiable demand from AI data centers for high-margin, specialized memory like HBM has forced manufacturers to reallocate their limited production capacity away from boring old consumer chips. As one Gartner analyst told The Verge, "If you are not a server customer, you will be considered a second priority."

This is a neat and tidy narrative. But it misses a much more interesting story. This isn’t just an accidental supply crunch caused by a new technology. It is also a story about market power, a familiar cyclical drama where a handful of powerful players have found the perfect opportunity to make a fortune.

To understand what is happening, you have to look at the market structure. Just three companies—Samsung, SK Hynix, and Micron—control 94% of the global DRAM market. It is a classic oligopoly. And as Dell's COO Jeff Clarke reminded investors recently on a earnings call, the industry is simply playing out a familiar script:

This isn't our first DRAM cycle. There have been seven, I think, in the last 40 years.

Look, we're in a very unique time. It's unprecedented. We have not seen costs move at the rate that we've seen. And by the way, it's not unique to DRAM. It's NAND, it is hard drives, leading edge nodes across the semiconductor network... Demand is way ahead of supply... First rule of our supply chain is to get the parts. Supply matters. Mix matters... We've learned a great deal since COVID, since the previous cycle of this last super cycle of this magnitude was 2016 through 2017...

No product category is not going to be impacted in terms of the aggregate cost basis moving. I don't see how this will certainly not make its way into the customer base. We'll do everything we can to mitigate that.

Some of those previous DRAM cycles were, to put it mildly, less than accidental. In court, they had a different name: "price-fixing."

In 2002, the U.S. Department of Justice launched a probe that resulted in Samsung, SK Hynix, and Micron all pleading guilty to their involvement in an international conspiracy to illegally inflate DRAM prices between 1998 and 2002. Executives went to jail. The companies paid hundreds of millions in fines. In 2018, they were hit with another class-action lawsuit for similar behavior that allegedly tripled DRAM prices between 2016 and 2018. This is an industry with a well-documented history of using its concentrated power to manipulate the market.

Which brings us back to today's shortage. After getting burned by an oversupply cycle in 2023, the memory giants are now enjoying a spectacular reversal of fortune. Take Micron: its net income more than doubled in a single year, jumping from $2 billion in the first quarter of 2025 to $5.5 billion in 2026. Or look at SK Hynix, whose net income just hit an all-time high of 12.6 trillion won, a figure up 119% from the year before.

Are they scrambling to increase production to meet this "unprecedented" demand? Not exactly.

A Samsung representative recently told investors that the company plans to "pursue a strategy of maintaining long-term profitability" instead of "rapidly expanding facilities." Meanwhile, SK Hynix is reportedly seeking only short-term contracts to prevent customers from locking in any long-term pricing stability.

This is the quiet part out loud. They are choosing higher prices over higher supply because it is fantastically profitable.

The decision from Micron to kill off its consumer-facing brand, Crucial, is the exclamation point on this strategy. The signal to the market is unambiguous: the small-margin business of selling RAM to everyday people is over. The real money is in selling high-margin server memory to a handful of desperate hyperscalers.

This is where the AI narrative becomes so useful. The "insatiable demand" from AI provides the perfect, unassailable cover for a highly profitable supply squeeze. It is just good business sense to prioritize the customers willing to pay the most. No one can blame Samsung for chasing the margins that AI servers offer, even if it means the price of a Chromebook goes up by 20%.

The AI boom isn't the fundamental cause of the memory shortage; it is the narrative catalyst for a market correction that the oligopoly has been waiting to engineer. The cost of building Jensen Huang's five-layer AI cake is not just being paid by venture capital and private credit. It is also being paid by anyone who wants to buy a new laptop, a new phone, or a new gaming console.

More on Semiconductor Shortage:

• The AI-fueled chip shortage could raise smartphone prices — new research spells out by how much (CNBC)
• Chip Shortage Lingers as Honda to Halt Output in Japan, China (Bloomberg)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

The PyTorch Alliance

• The Headline: Google is partnering with Meta on a new initiative, "TorchTPU," to make its AI chips fully compatible with the popular PyTorch software framework, a direct move to challenge Nvidia's CUDA software dominance. (Reuters)
• ARPU's Take: This is the first truly credible, large-scale assault on Nvidia's CUDA software moat. Google is finally admitting that forcing customers to use its internal tools is a failed strategy. By partnering with Meta to bring the industry-standard PyTorch to its custom hardware, Google is making a pragmatic and aggressive move to break Nvidia's software lock-in.
• The Product Question: This initiative confirms that the primary battleground in the AI chip market is the software ecosystem, not just the hardware itself. If PyTorch runs as well on TPUs as it does on GPUs, Nvidia's CUDA lock-in evaporates for a massive segment of the market, turning the competition into a pure price/performance battle where Google's vertical integration gives it a margin advantage.

The Conversational Commerce Race

• The Headline: DoorDash is launching a grocery ordering plug-in for ChatGPT that allows users to create shopping lists from recipes, a direct competitive response to a similar move by Instacart. (Bloomberg)
• ARPU's Take: The race to become the default shopping cart inside AI has officially begun. This is a necessary defensive move by DoorDash, which cannot afford to let its chief grocery rival, Instacart, own this powerful new customer acquisition channel. The key difference—DoorDash's redirect to its own app for checkout versus Instacart's fully in-chat experience—reveals two competing philosophies for the future of AI-driven commerce.
• The GTM Question: This move establishes conversational AI platforms as a new top-of-funnel for e-commerce. The key question for go-to-market leaders is whether to treat these platforms as a simple lead-generation channel or as a fully transactional, end-to-end point of sale.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.