The Physical Layer of AI-Crypto: What Korean Substrate Makers Just Told Us

Daeduck Electronics just reported a 3,599% year-over-year surge in operating profit. That is not a typo. A mid-tier Korean PCB manufacturer, hardly a household name, produced financials that would make any DeFi protocol yearn for its 2021 glory days. Meanwhile, crypto remains stuck in a sideways grind, awaiting a macro trigger that never seems to arrive. Markets lie, but liquidity tells the truth. This quarter, the most consequential liquidity event did not occur on-chain. It happened on the factory floor, in the form of FC-BGA substrates, high-layer-count PCB boards, and a supply chain that is tightening faster than any algorithmic stablecoin yield. The companies in question are Daeduck Electronics, Simmtech, and TLB. They are not chip designers or foundries. They sit in the unglamorous middle of the semiconductor value chain, manufacturing package substrates and multi-layer PCBs — the intricately layered boards that sit beneath a GPU die and connect it to memory, power, and the rest of the server. When NVIDIA ships a Blackwell GPU, that chip requires a large, warpage-controlled package substrate. When Samsung or SK Hynix ships HBM memory, it is mounted on a high-speed interposer board. None of this works without the quiet, capital-intensive work of Korean substrate makers. For the past decade, these companies competed on thin margins in a mature industry. AI has flipped the script. Daeduck’s second-quarter net income exploded, and operating margins at Simmtech hit 12.2%, TLB 14.5%. These numbers exceed the global PCB industry average of 8-12% — a clear statistical anomaly that points to a structural shift, not a one-time blip. The only way to explain such compression in earnings announcements is that demand for AI-server-grade substrates has overwhelmed supply. We are looking at a seller’s market. Let me unpack the technical layer, because the details matter. In semiconductor packaging, the critical metric for a substrate is line width and spacing (L/S). The leading Japanese and Taiwanese players — Ibiden, Shinko Electric, Unimicron — have pushed FC-BGA substrates down to 5/5 micrometers or better, supporting die sizes above 80×80 mm. Korean manufacturers, with the exception of Samsung Electro-Mechanics, lag by roughly one to one and a half nodes. Daeduck’s FC-CSP technology is world-class, but its FC-BGA is still catching up. This is measured in terms of build-up layers (typically 12-20) and substrate area (moving from 50×50 mm toward 70×70 mm and beyond). In multi-layer PCB for AI servers, the construction involves 16-24 layers of high-speed material at M6 or M7 grade. Here is the insight others miss. That technology gap is exactly why the earnings surge is so telling. In a normal market, a second-tier supplier with lower yield and weaker design rules would struggle to raise prices. But when demand outpaces the top-tier capacity, the overflow flows downstream. Mid-tier Korean manufacturers absorb that overflow, and their margins jump from 5-8% to double digits. From my experience observing the AI hardware supply chain, I started calling this the Physical Liquidity Index: when second-tier suppliers start reporting margins above industry average, it means the AI buildout is moving from the engineering phase into the mass-manufacturing phase. On-chain metrics might be flat, but this index is screaming expansion. Alpha is found where others see only noise. Most crypto analysts focus on GPU prices, cloud computing rates, or the token emissions of decentralized compute networks. They ignore the substrate suppliers because these companies do not mention AI or crypto in their earnings calls. But the substrate is the physical bottleneck. An AI chip cannot function without it, and the chip cannot reach the end-user without a server board that costs more to produce than a comparable traditional PCB. The supply chain tells you where the real demand is flowing. If Substrates are sold out for the next three quarters, every GPU sold is a GPU that needs one of these boards. The hidden lock is material supply. The most critical raw material for ABF substrates is Ajinomoto Build-up Film, an interlayer insulation film produced by — you guessed it — the Japanese seasoning giant which holds over 90% global market share. There is no synthetic replacement. This is not a standard commodity. The film requires specific dielectric properties for high-frequency signal integrity, and it must be laminated layer by layer, with laser-drilled microvias aligning each stratum precisely. Korean firms are completely dependent on this import. A geopolitical disruption similar to the 2019 Japan-Korea export restrictions would freeze their production lines. The same applies to high-speed copper-clad laminates (CCL), where Panasonic’s M6/M7 grades remain the industry benchmark, though domestic Doosan has made credible strides. The customer concentration compounds the risk. Simmtech depends heavily on Samsung and SK Hynix for memory module substrates. Daeduck’s AI-related revenue is heavily weighted toward a small number of design houses. In theory, large customers hold pricing power over component suppliers. Yet the Q2 results prove the opposite — for now. Supply is so constrained that even mid-tier suppliers can demand premium prices. This is a temporary power shift, but it is also a signal about the AI-crypto convergence trade. When I directed our fund’s allocation to decentralized GPU rendering protocols, I did not base the thesis on abstract ideas about AI agents. I based it on this type of data: the physical capacity of the AI server supply chain and the pricing power of its less glamorous suppliers. There is a contrarian angle that most analysts will miss. The market narrative across crypto is that AI-crypto convergence will democratize compute. The belief is that decentralized networks will allow anyone to monetize idle GPUs and bypass the hyperscale giants. But the Korean data suggests the opposite: the AI buildout is accelerating centralization in the physical layer. NVIDIA controls the design. TSMC controls the advanced packaging. Ibiden and Ajinomoto control the substrate and material ecosystem. Korean suppliers are becoming more entrenched in this centralized hierarchy, not less. Their dependence on NVIDIA, Samsung, and SK Hynix is deepening, not diversifying. Even as we invest in decentralized compute protocols, we must recognize that the underlying hardware is still minted in a tightly controlled oligopoly. Structure emerges from the chaos of contraction. The 2022 bear market forced out speculative capital and left a hardened core of physical infrastructure builders. The result is not a decentralized Utopia. It is a reassertion of industrial hierarchy. The Korean substrate makers benefited from this contraction by buying cheap equipment, improving yields, and waiting for the demand surge. Their fat margins are the reward for survival. This is the pattern that repeats across every technological paradigm: the sharpest alpha is found in the companies that no one is watching while the hype cycle focuses on the flashy application layer. The takeaway for strategic positioning is direct. Track Daeduck, Simmtech, TLB, and the ABF film supply chain as leading indicators for the AI-crypto cycle. When their margins begin to compress and capacity expands, that will be your sell signal for AI-crypto narratives. When margins are expanding as they are now, the physical buildout is accelerating, and the digital asset plays attached to that buildout have underlying value support. We do not predict; we position. And right now, the physical layer is telling us that the AI-crypto trade is far from exhausted. It is only beginning to source its raw materials.