Daeduck Electronics just posted a quarterly operating profit figure that looks like it was printed by a meme coin on a good day: up 3,599% year over year.
Stop there with me for a second. This is not a DeFi protocol with a rebase mechanism and a friendly whitepaper. This is a Korean manufacturer whose daily work involves laminating copper layers and drilling microscopic vias through build-up film. When a company's operating profit rises more than thirty-five times in a single year, organic demand growth cannot carry the explanation on its own. Something structural moved under the surface.
I spent 2024 building a transparent copy-trading dashboard and watching institutional order flow tear through the NVIDIA-adjacent complex. Every cycle moves the same way: retail chases the visible asset — the GPU, the token, the AI coin — while durable money sits one layer deeper, inside companies most traders cannot name. Korea's PCB and package-substrate makers live in that layer. Their Q2 numbers just leaked what the AI trade actually runs on. It is not the silicon. It is the board the silicon mounts to.
Quick translation for readers who think in blocks rather than chips. An IC substrate is the physical bridge between a raw chip die and the server motherboard. For AI workloads, the family that matters is FC-BGA — Flip Chip Ball Grid Array: multilayer structures of typically 12 to 20 layers, with line-and-space features in the 8/8 to 15/15 micron range. These are the landing pads for CPUs, GPUs, and custom ASICs. Below them sits FC-CSP, the smaller cousin used for RF, power management, and baseband silicon. At the system level, an AI server motherboard runs 16 to 24 layers with high-speed materials rated M6 or M7.
This is not obscure trivia. It is the plumbing of the entire AI economy — and by extension, the plumbing behind every crypto narrative promising decentralized compute. Render, Akash, Bittensor: all claims on physical GPUs. A GPU does nothing until it is mounted on a large-format FC-BGA substrate, itself integrated inside a CoWoS-style package once TSMC interconnects the GPU die with HBM memory stacks. The substrate sits right underneath the visible bottleneck. Package substrates account for an estimated 30% to 50% of total advanced packaging cost. That is real value locked in an invisible layer.
Korea's role in this chain is specific. Daeduck Electronics holds genuine traditional strength in FC-CSP and is chasing the FC-BGA frontier from behind. Simmtech is wired into the memory ecosystem, shipping substrates tied to Samsung and SK Hynix. TLB builds DDR5 modules and enterprise SSD boards — the storage layer that scales with every AI cluster. None of these are front-page names. They are exactly the kind of hands that matter when hype fades, which is why I read their earnings reports the way other people read liquidation data. In my copy-trading work, I obsess over traders whose edge shows up in a delayed metric — not the leaderboard P&L, but the execution quality underneath. Substrates are that delayed metric for the AI trade. They tell you whether demand is real months before the software revenue lands.
What the 3,599% actually is.
Operating margin tells the story better than the headline. Daeduck printed 17.5% in Q2. Simmtech delivered 12.2%. TLB came in at 14.5%. The global PCB industry average sits between 8% and 12%. These Korean players cleared the high end of that band — and Daeduck blew past it entirely.
I have audited enough tokenomics to recognize this signature. When a manufacturing business suddenly produces margins like this, one of two things happened: the product mix shifted violently toward premium, or the company is pricing with genuine scarcity. In the substrate world, both happened at the same time.
Watch the order flow. AI demand has kept TSMC's CoWoS capacity near its limits for well over a year. Every high-end GPU that gets packaged needs a large-area FC-BGA substrate under it. The Japanese leaders — Ibiden, Shinko Electric — and Taiwan's Unimicron operate at the frontier: line-and-space below 5 microns, substrate sizes climbing past 80 by 80 millimeters. Korean makers trail by roughly one to one-and-a-half product generations. In a normal market, being a generation behind in substrate technology means collecting scraps. In this market, even the scraps carry fat margins.
But here is the hidden insight I keep circling. Revenue growth alone cannot produce a 3,599% profit expansion. The mathematics require both volume and price — and for a second-tier player like Daeduck, that implies one more component: share gains inside the NVIDIA supply chain. The most plausible reading of these numbers is that Daeduck's FC-BGA and FC-CSP parts stepped up in allocation as frontier capacity saturated. That is the classic second-source surge. It gets rewarded handsomely. It is also a rented position, re-evaluated every quarter and re-written at every design change.
There is an opacity problem here that we in crypto understand all too well. In 2025, I led a coalition of copy-traders demanding transparency standards from AI trading bots, and we pushed developers to build open-source audit tools for decision logs. That fight taught me something: when the logic is hidden, the risk is hidden. The allocation decisions inside the AI hardware supply chain are exactly that kind of black box. We see the output — record profits — but not the input logic. Who got the allocation? At what price? For how many quarters? The earnings report is the decision log, but it is a heavily redacted one.
The ABF problem nobody wants to name.
Here is where this supply chain starts to look like a badly designed token. Ajinomoto Build-up Film — ABF — is the core insulating dielectric for high-end substrate manufacturing. Ajinomoto, a Japanese company, controls more than 90% of global supply. Every serious substrate maker on earth buys the most critical material in its process from one supplier.
I have spent years warning my community about token distributions where a single wallet holds the float. This is the hardware equivalent. If Japan's export-control machinery were ever aimed at this material — and I remember 2019, when Japan restricted semiconductor materials to South Korea amid the wartime labor dispute — Korean substrate factories would suffocate first. That memory is not lost on the supply chain, even if it is invisible in the glossy earnings slides.
The dependence does not end with film. The laser drilling machines, the exposure tools, the plating chemistry — all come from a small set of suppliers, mostly Japanese with a few German names in the mix. Korean makers have a domestic option for high-speed copper-clad laminate through Doosan Electronics, but the highest-end M7/M8 materials the newest server boards demand still skew heavily toward imports, with Panasonic as the benchmark. The upstream picture is blunt. Korean substrate makers are excellent manufacturers but renters in the technology supply chain. They own the drills. They do not own the formula. In copy-trading terms: they are the strategy followers with good execution, not the strategy creators with the edge. And we all know what happens to followers when the regime shifts.
Warpage and the unforgiving physics of large substrates.
Let me get technical for a moment, because the physics matter when forecasting who wins. The core challenge in large-format FC-BGA is warpage control. A substrate at 50 by 50 millimeters already fights internal stress across copper, glass, and dielectric layers. Moving to 70 by 70 millimeters or larger multiplies the difficulty: coefficient of thermal expansion mismatch grows, layer alignment tolerances tighten, and the substrate can physically bow enough to break the connection between the package and the board. Warpage kills yield at the assembly step, long after the substrate has been manufactured. That is why precision laser drilling and layer-to-layer alignment are the structural moats of this industry.
Korean firms have crossed the fundamental skill threshold — Daeduck's margins prove they moved past the yield-ramp phase into stable production. But they still run five to ten percentage points behind the global leaders on high-end ABF yield, where the frontier sits at 80% to 90%. That yield gap is the mathematical expression of weaker pricing power. It is also time-sensitive: with AI demand this hot, my honest estimate is that the gap narrows by three to five percentage points over the next 12 to 18 months — but only if Korean firms keep spending into the catch-up cycle, which requires the demand to stay hot.
The APY trap of the hardware world.
There is a pattern here that every DeFi veteran will recognize. For years, we watched protocols buy TVL with liquidity mining incentives. The yields looked glorious while the subsidies flowed. Stop the incentives, and the real users vanish, leaving only mercenary capital. The lesson I keep repeating is simple: yield fades, loyalty compounds. Activity subsidized by a single sponsor is not demand. It is rent.
The AI capex cycle is running the same playbook at the hardware level. Substrate capacity expansion is being subsidized by extraordinary customer urgency. The capital expenditure Korean makers are now committing to FC-BGA lines is a bet that NVIDIA-scale demand is durable, not cyclical. I am not saying the bet is wrong. I am saying it is a bet — and the tokenomics comparison is uncomfortable. When the reward emission schedule slows down, the yield on that expansion collapses faster than the optimists project.
That is why I separate the profit line from the pricing power line. Profits are a point-in-time measurement. Pricing power is the ability to survive a bad quarter. Korean substrate makers have profits. The pricing power still belongs to the film monopoly, the design authority, and the frontier yield leaders.
The hand-me-down signal.
One more signal, subtle but telling. Taiwanese substrate leaders have been quietly trimming BT substrate capacity — an older, lower-margin material class — to redeploy capital into ABF and high-end FC-BGA. On the surface, that looks like portfolio management. Take the deeper read, and it is a disclosure.
BT substrates are not the AI frontier. If the industry leaders happily cede that territory, they are telling you the moat there is shallow and the return on ABF engineering is higher. That is short-term good news for Korean manufacturers who inherit the volume. It is a long-term warning, because the center of gravity is moving up the technology curve, into the exact segment where Korea trails. Korean firms are being handed a commoditizing market while the strategic core migrates elsewhere. That is not a repositioning. That is a hand-me-down.
We see the same pattern inside crypto too. Dozens of Layer 2s launch, all competing for the same small user base. That is not scaling; that is slicing already-scarce liquidity into fragments. Multiple Korean substrate makers chasing the same allocation from the same two chip designers is not a diversified supply chain. It is the same pie, cut thinner.
Here is the uncomfortable part. Read the Q2 numbers one way, and Korea is winning. Read them another way, and Korea is thriving inside a window someone else built and someone else can close.
Retail will see a 3,599% profit jump and conclude the play is to chase these companies or their tokenized proxies. I read it differently. The margins are real today because global substrate supply is tight and NVIDIA's appetite is voracious. But that pricing power was not earned at the technology frontier. It was allocated by scarcity. The moment AI capex growth breathes, mid-tier yield and Japanese material dependency become the dominant variables again. The genuine frontier holders — Ibiden, Shinko, Unimicron — carry better cards. Daeduck, Simmtech, and TLB hold the second-best hand. And the dealer controls the deck.
The counter-intuitive angle is that the very concentration which produced Daeduck's staggering profit growth is the same force that will produce the pain cycle. NVIDIA's Vera Rubin platform changes substrate specifications — larger packages, more layers — and every design shift resets the map. Today's second-source winner becomes tomorrow's excess capacity if the design moves elsewhere.
This mirrors the governance problem in our own systems. Everyone says delegation decentralizes control, but users hand their votes to the same five KOLs and we call it governance. Same structure here: a nominally diversified global supply chain whose real control concentrates in one design house, one film monopoly, and a handful of machine-tool vendors. The Korean substrate makers are executing a delegated mandate. They are delegates, not delegators — replaceable if yield lags, or if a new design drops them out of allocation.
Which is why my answer to the community when they ask about AI hardware bets is always the same. Follow the people, follow the profit. Follow whoever controls the ABF, the design rules, and the machine tools — not just whoever runs the drills. The drills are replaceable. The formula is not.
So here is what I will be tracking in the coming quarters, and what I would track if I were you.
One: Ajinomoto's capital expenditure plans. If the film monopoly starts expanding capacity aggressively, that is the bottleneck itself telling you the demand is durable. Two: any Korean announcement of large-format FC-BGA lines targeting 70 by 70 millimeters and above. That is the marker of a real catch-up attempt rather than a cyclical windfall. Three: the Vera Rubin substrate specifications. Every GPU packaging generation rewrites the allocation map for the entire substrate industry.
The AI trade is not a story about tokens, and it is not really a story about chips. It is a story about the physical layer where the chip meets the board — and about who controls the materials that layer depends on. Community first, coins second. Always. This quarter, the community that matters is a group of Korean factories betting their future on a Japanese film they cannot replace.
Trust the hands, not just the charts. Right now, the hands are stacking layers, fighting warpage, and praying their supplier keeps shipping.