Samsung’s 2026 operating profit is now forecast to exceed the company’s total profit over the past 40 years. SK Hynix alone is expected to report a Q2 profit near 64.4 trillion won. Combined, the two Korean memory giants are on track to generate close to 150 trillion won in a single quarter. These numbers are not just staggering—they expose a fragile reality that most of crypto refuses to discuss: our entire decentralized infrastructure sits on a wafer-thin foundation of centralized silicon.
Let me be clear: I’m not here to rain on the AI parade. I’ve spent years in this industry, from my early days as an Ethereum Foundation community advocate translating EIPs for non-technical users to my current role as a decentralized protocol PM auditing governance models. I’ve seen hype cycles come and go. But the current bull market’s euphoria is masking a structural risk that could unravel everything we’ve built.
The HBM Super Cycle: A Blessing and a Curse
The engine behind these jaw-dropping numbers is High Bandwidth Memory (HBM)—specifically, the HBM3E chips that power NVIDIA’s H100 and B200 AI accelerators. Both Samsung and SK Hynix have bet their future on this single technology. SK Hynix currently holds a 50% market share in HBM, while Samsung trails at 30%, struggling with yield issues around its latest 1b (12nm class) process. The technical challenge lies in 3D stacking: TSV (through-silicon vias), micro-bumps, and hybrid bonding create multi-deck architectures that are as much art as science.
During my time auditing DeFi protocols, I learned that complexity is the enemy of security. The same principle applies here. HBM’s advanced packaging—TC-NCF, MR-MUF, X-Cube—creates layers of dependency that are opaque even to semiconductor insiders. One manufacturing defect, one contamination event, and the entire supply chain stalls. We saw this in 2022 when a power outage at a Samsung fab caused a 10% drop in global DRAM output. Now imagine what a single HBM yield bug could do to the entire AI compute layer.
But the deeper issue is not technical—it’s relational. NVIDIA is the sole meaningful customer for HBM today. That means 90% of the revenue generated by these two Korean giants flows through a single corporate bottleneck. From hype cycles to hydraulic stability, this is the opposite of decentralization. It’s a waterfall model, not a mesh network.
The Contrarian Angle: Why This Super Cycle Might Be a Mirage
The articles celebrating Samsung’s “40-year profit” are missing a crucial point: that number is largely a PR artifact. Forty years ago, the semiconductor industry was a fraction of today’s size. Comparing absolute profit numbers across decades is like comparing a 1990s flip phone to an iPhone 16—semantically true but fundamentally misleading. More importantly, the high capital expenditure required to maintain HBM leadership (Samsung alone spends 40 trillion won annually on CapEx) means that much of this “super profit” is immediately reinvested into machines and cleanrooms. It’s not free cash flow; it’s an arms race.
And here is where my contrarian side kicks in. I believe the market is pricing these stocks as if the AI boom will never end. But history tells us that every storage super cycle—from DRAM in the 90s to NAND in the 2010s—was followed by a brutal correction. The only thing that might break this pattern is if AI demand proves structurally different. Yet the signs of froth are everywhere. NVIDIA’s lead is being challenged by AMD and self-designed chips from hyperscalers like Google and Amazon. If the next generation of AI training shifts to alternative memory architectures—or if HBM4 adoption is delayed—the entire profit thesis collapses.
From Chips to Chains: What This Means for Crypto
We are not just users; we are the protocol. And the protocol relies on hardware. Every validator, every miner, every zk-proof generator consumes silicon. Right now, that silicon is overwhelmingly produced by two Korean factories that are themselves dependent on a single Dutch company (ASML) for lithography equipment and a single Japanese consortium for critical chemicals. This is a centralized supply chain for the decentralized world.
During the 2021 bull run, I launched a DAO for digital art curation, managing a treasury of $200k in ETH. The lesson I learned was painful: when the market turns, everything that depends on continuous compute demand—including mining, staking, and even rollups—feels the squeeze. The 2022 Terra collapse taught us that infinite leverage is a fairy tale. The 2023 FTX scandal taught us that trust in centralized entities is misplaced. Now, 2024 is teaching us that the physical foundation of our digital dreams is fragile.
The code is cold, but the community is warm. We cannot let the coldness of silicon dictate the fate of our warm, human-powered networks. That means we must invest in hardware diversity. It means supporting projects like Akash or Golem that decentralize compute, but also pushing for open-source chip designs and alternative memory technologies. The ultimate solution is not to rely on Samsung or SK Hynix to rescue us—it’s to build systems that can function even if they fail.
The Takeaway: Build for Resilience, Not for the Hype
My final thought is a question for every builder reading this: If your protocol depends on a single HBM supplier—or even a single GPU vendor—what happens when that vendor hits a yield problem? What happens when geopolitical tensions cut off the supply of EUV photoresist from Japan? Your decentralized dream becomes a centralized nightmare.
We are entering a new era where the lines between AI, crypto, and semiconductor manufacturing blur. The winners will not be those who ride the super cycle the highest, but those who design systems that can survive its inevitable bust. From hype cycles to hydraulic stability—that’s the engineering challenge of our generation.
Let’s not waste it on a single-chip bet.
— Abigail White Decentralized Protocol PM, Rome Former Ethereum Foundation Community Advocate Author of "The Sentient Ledger" series