Hook: The Liquidity Mirage at Scale
In the first quarter of 2024, Samsung Electronics reported its highest semiconductor profits in over a decade. The headlines screamed success: AI demand fueled record DRAM and HBM revenue, driving the DS (Device Solutions) division to an operating profit of approximately $5.3 billion. But as I dissect the numbers—sifting through the earnings release, the supply chain filings, and the whispered tensions between Seoul and Washington—I see not a triumph of technological mastery, but a carefully constructed illusion. An illusion that liquidity, whether in crypto markets or in chip supply chains, is the same as settlement.
I’ve spent years auditing liquidity pools in DeFi, watching billions of dollars in TVL evaporate when the underlying economic moat proved fragile. The semiconductor industry today is no different. Samsung is the world’s largest memory maker, the second-largest foundry, and a leader in HBM (High Bandwidth Memory) packaging. Yet beneath the surface of record profits lies a structural dissonance that mirrors the very flaws I exposed in Uniswap V1 during the 2019 audit: 80% of the liquidity was fleeting “fat token” manipulation. Today, 80% of Samsung’s profit surge is fleeting, propped up by a single AI-driven product cycle. The rest is a slow bleed.
Context: The Global Liquidity Map of Chips
To understand the illusion, we must map the global liquidity of semiconductors. The industry is not a monolith; it’s a fragmented web of capital flows, manufacturing dependencies, and geopolitical fault lines. Samsung sits at its center as an IDM (Integrated Device Manufacturer), controlling everything from DRAM design to advanced packaging. Its DS division contributed over 60% of Samsung Electronics’ total operating profit in Q1 2024, a stark reversal from the 2023 trough when the division was barely profitable.
But this liquidity—the billions flowing into Samsung’s coffers—is not distributed evenly. The profit centers are clear: HBM and DRAM for AI servers, where Samsung commands roughly 40% of the HBM market and 42% of all DRAM. These products are sold to hyperscalers like NVIDIA, AMD, and Google, who are themselves in a frenzy to build AI infrastructure. The demand is real, but it is concentrated. It is a narrow pipeline feeding a single end-use case: AI training and inference.
Meanwhile, Samsung’s foundry business—the division that manufactures logic chips for external clients—remains a persistent loss leader. Despite investing over $170 billion in new facilities (including the Taylor, Texas plant and the Pyeongtaek clusters), the foundry’s operating margin hovers around -5% to 0%. The capacity utilisation rates tell a grim story: DRAM fabs at 90%+, NAND at 75%, but advanced logic fabs at only 60–70%. This is not scaling; this is fragmentation. As I wrote in my 2021 DeFi Summer analysis, “When liquidity is sliced into too many pools, each pool starves.” Samsung is starving its own foundry to feed its memory business.
Core: The Architecture of the Illusion
Let’s drill into the mechanics. The “record profits” are a function of two factors: the cyclical rebound of memory prices and the structural explosion of HBM demand. The DRAM price index rose over 50% from Q4 2023 to Q1 2024, driven by AI server procurement. HBM3e, the latest generation, sells for three to five times the price of standard DDR5. Samsung’s HBM revenue in Q1 2024 was approximately $3.5 billion, representing over 20% of its total semiconductor revenue. This is the core of the illusion: a single product category, HBM, is propping up an entire empire.
But HBM is not a technological moat; it’s a packaging dependency. The true bottleneck is not memory fabrication but advanced packaging: the 2.5D and 3D stacking techniques that integrate HBM with compute chips. Samsung’s Cubed technology (I-Cube for 2.5D, X-Cube for 3D) is world-class, but it relies on a fragile supply chain. The TC-NCF (Thermal Compression Non-Conductive Film) process, critical for HBM stacking, uses materials sourced almost exclusively from Japan and the Netherlands. The machines for high-volume manufacturing—the ASML TWINSCAN EXE:5000 series High-NA EUV lithography—are single-sourced. Samsung has one of the first units, but its deployment for 2nm logic production is still in R&D.
From my experience auditing the liquidity of DeFi protocols, I learned that the most dangerous positions are those with high apparent returns but low underlying settlement finality. In crypto, settlement is the moment a transaction is irreversibly recorded on-chain. In semiconductors, settlement is the moment a chip is physically fabricated and tested. Samsung’s HBM business has high settlement because the chips are being produced and shipped. But the liquidity that funds its expansion—the capital allocated to foundry, to new fabs in Texas, to the massive R&D budget—is settling into assets that may never yield profits.
Let’s look at the capital expenditure numbers. In 2024, Samsung’s semiconductor capex is expected to be $35–40 billion, roughly 50–57% of its semiconductor revenue. For comparison, TSMC’s capex is 35–45% of revenue. Samsung is spending a higher proportion of its income on building new capacity, yet its foundry revenue is only $10–12 billion (versus TSMC’s $70 billion). This is not investment; it’s subsidy. Samsung is using its storage profits to subsidize its foundry ambitions, in the hope that it can eventually catch up to TSMC in advanced logic. But the timeline is uncertain. The Texas plant, originally slated for 2024 production, has been delayed to 2025. Ramp-up to full production will take until 2027. By then, TSMC will have moved to 1.4nm and Intel will have shipped 18A. The liquidity is being poured into a pool that may never reach positive settlement.
I recall my 2019 audit of Uniswap V1, where I manually tracked 50 high-frequency trading wallets and discovered that 80% of the liquidity was ephemeral—money that would vanish at the first sign of volatility. Today, Samsung’s foundry business is that ephemeral liquidity. The billions flowing into it are not creating lasting value; they are masking a structural deficit. The company’s DS division as a whole has a return on invested capital (ROIC) of 5–8%, below its weighted average cost of capital (WACC) of 8–10%. In other words, Samsung is destroying value with every dollar it invests in its semiconductor empire. The only reason it continues is because the storage cycle provides temporary oxygen.
Contrarian: The Decoupling Fallacy
The prevailing narrative among semiconductor bulls is that AI demand will decouple the industry from historical cyclicality. The argument goes: AI infrastructure is a structural, multi-year investment cycle that will sustain high memory prices and drive foundry utilisation. But this narrative ignores two critical counterpoints: the concentration of demand and the fragility of supply chains.
First, demand concentration. Over 80% of Samsung’s HBM output goes to a single customer ecosystem: NVIDIA and its GPU supply chain. If NVIDIA’s next-generation architecture (Rubin, expected 2026) shifts to custom memory solutions or if SK Hynix gains further market share, Samsung’s HBM revenue could collapse. This is not a diversified base; it’s a single pool of liquidity. I’ve seen this pattern before in DeFi: a yield farm that offers 500% APY attracts massive TVL, but when the underlying farm token crashes, the TVL evaporates in hours. Samsung’s HBM revenue is a 500% APY yield farm in disguise.
Second, supply chain fragility. The semiconductor industry is currently undergoing a geopolitical “decoupling” driven by US-China tensions. The CHIPS Act and the US export controls on advanced chipmaking equipment have forced Samsung to bifurcate its supply chain. Advanced fabs (for 3nm and below) are now built exclusively in Korea and the US. The Xi’an NAND plant in China is restricted to mature nodes. This has increased costs and complexity. But the real risk is the dependence on a single supplier for critical equipment: ASML for High-NA EUV, and Japanese firms for high-end photoresists and etch tools. In a worst-case scenario—a Taiwan blockade, an escalation of the chip war—Samsung’s ability to produce HBM or advanced logic could be severely impaired. The “profit” is settlement only if the chips are produced. If the machine doesn’t arrive, the illusion shatters.
This is the decoupling fallacy: that AI demand creates a shield against geopolitical risk. In reality, it amplifies the risk. The more the world relies on a single node—Samsung’s HBM, TSMC’s 3nm, NVIDIA’s GPUs—the more vulnerable the entire system becomes to a single point of failure. During my DeFi Summer disillusionment in 2021, I saw how “sovereign” protocols collapsed when the underlying oracle (Chainlink) failed to deliver accurate price feeds. The oracle was the single point of failure. In semiconductors, ASML’s High-NA EUV is the oracle. If it fails, the entire AI narrative fails.
Takeaway: Settlement Before Liquidity
As a CBDC researcher, I’ve spent years studying how central banks approach systemic risk. The principle is simple: liquidity is a tool, but only settlement—final, irreversible, trust-minimized settlement—creates stability. Samsung’s record profits are a liquidity event. They are real in the moment, but they are built on a foundation of concentration, subsidy, and geopolitical fragility.
What happens when the AI hype cycle peaks? When NVIDIA’s next generation of GPUs no longer requires as much HBM? When SK Hynix’s 12-layer HBM4 captures market share? When the cost of Samsung’s foundry subsidies becomes unsustainable? The answer is a liquidity crisis disguised as a profit warning. The same thing happened to DeFi in 2022: Terra crashed, Three Arrows collapsed, and the “decentralized” financial system revealed itself as a house of cards. The semiconductor industry is no different.
The regulators in Manila, Basel, and Frankfurt are watching. They are designing CBDCs not just as digital cash, but as settlement layers that can withstand liquidity dry-ups. The lesson from Samsung is that settlement—the ability to produce and deliver a physical chip in a geopolitically fractured world—is the only real moat. Everything else is speculative liquidity.
I’ll leave you with this: the next time you see a headline about “record AI-chip profits,” ask yourself: where is the settlement? Is the company actually building a defensible supply chain, or is it just riding the liquidity wave? For Samsung, the answer is uncomfortably close to the latter.
