실리콘 커패시터 on Korea Invest Insights

AI Token Futures and Cost Per Token: It's Now a Cost Race, Not a Performance Race

Sat, 30 May 2026 09:00:00 +0900

📚 Context for follow-up posts This is a follow-up to Dell’s Earnings Surprise and Korean Semiconductors and Marvell Q1 FY2027 and Korean Semiconductors. Where the first two posts argued that “the AI bottleneck migrates from the GPU down to memory, interconnect, substrate and power — and who earns the margin at that bottleneck,” this post goes one step further. Once AI usage itself gets a market price, the battleground shifts from “a faster chip” to “a cheaper token.” Related hubs: AI HBM Hub, Korea Semiconductor Value Chain Hub, AI PCB & Substrate Hub.

TL;DR

Until now, AI investing has been a fight over “who builds the more powerful chip.” But a quiet signal has just switched on. AI usage itself is starting to get a market price.

According to Reuters, China’s Shanghai Futures Exchange (SHFE) is early-designing futures linked to AI token prices. In the U.S., CME Group (with Silicon Data) and ICE (with Ornn) announced plans for GPU compute futures. None of these are listed, confirmed products yet. But the direction is clear.

Once AI compute and tokens have a “market price,” the industry’s axis shifts from a performance race to a cost race.

It is the same as when oil got a futures price: refining, shipping and power generation all lined up by “cost per barrel.” Once AI’s price per token becomes visible, every participant asks the same question: “What does it cost me to produce one token?"

The answer is not generic “AI stocks.” It is the bottleneck parts, chips and platforms that actually lower the customer’s cost per token. The priority order:

Priority	What to Buy	Key Names	View
1	Custom ASIC / AI networking	Broadcom, Marvell, TSMC	Most direct to the token-cost stack. But it’s a consensus long — valuation discipline needed, customer-concentration risk
2	HBM + eSSD + KV-cache storage	Samsung Electronics, SK Hynix, Micron	Memory bandwidth and cache management are the inference bottleneck. eSSD/KV-cache cut repeated compute and ease HBM load
3	Advanced packaging / FC-BGA / MLCC / silicon capacitor	Samsung Electro-Mechanics, Daeduck Electronics, Murata, Ibiden	Power & signal integrity → effective token throughput. “Right direction, wrong price” risk
4	Compute pricing rail	CME, ICE	Financialization itself. Option value, pre-launch, liquidity uncertain. Least-crowded 2nd-order idea
5	Nvidia / AMD full-stack	Nvidia, AMD	Still core. But the thesis becomes “highest tokens-per-watt & AI-factory economics,” not “GPU shortage”
6	Power equipment	HD Hyundai Electric, LS Electric	Not an outright short. Power demand keeps rising (IEA: datacenter power ~945TWh by 2030). Use only as a relative-value hedge against an over-stretched power-scarcity premium

The Korea bottom line first: Samsung Electronics is the most balanced practical pick. HBM4E catch-up + DDR5 + SOCAMM2 + PCIe Gen6 eSSD + KV-cache storage give it the broadest exposure across the token-cost stack. SK Hynix is the best-quality name but crowded. Samsung Electro-Mechanics has high thesis purity but price matters. Daeduck Electronics is a second-order FC-BGA beta candidate.

1. What Is Happening: The Financialization of AI Usage

First, separate the facts.

[Fact] Per Reuters, SHFE is early-designing futures linked to AI token prices. The contract specification, the definition of the underlying (which token), the settlement method, and the approval/launch timeline are all unconfirmed.
[Fact] CME Group announced plans to launch compute futures with Silicon Data; ICE announced plans for GPU compute futures contracts with Ornn.

What matters here is not “futures are already listed.” They are not. What matters is that the industry has started trying to put a price on AI usage.

An analogy: imagine using electricity without any standard price of “X per kilowatt-hour.” The moment a power exchange opens and prices get quoted, every factory starts computing “how much electricity do we use per unit of product?” AI is now standing at exactly that threshold.

When the price becomes visible, the cost becomes visible. When the cost becomes visible, whoever lowers it wins.

2. Why This Changes the Rules: Performance Race → Cost Race

For years the AI race was about “benchmark scores” — who runs a bigger model on a faster chip. Investing followed the same logic: “buy the company that makes the most powerful GPU.”

Once usage gets a price, the question changes.

Cost per token, in the simplest form:

Cost per token = total compute cost / tokens processed

You lower it by shrinking the numerator (compute cost) or growing the denominator (tokens processed). Whoever extracts more tokens from the same electricity and the same capital wins. So two metrics become the new battleground:

Tokens per watt = tokens/sec / power (W)
Tokens per dollar = tokens / total cost ($)

Per Reuters, TSMC sees chip-design focus shifting from raw performance to power efficiency because of AI power demand — meaning tokens per watt becomes a key metric.

In investment language, this is clear:

Not “buy AI stocks,” but “buy the companies that lower the customer’s cost per token."

3. The Kitchen Analogy: The Expensive Chef Is Sitting Idle

Let’s make it more concrete.

A GPU is a very expensive chef. Their hands are fast, but they can only cook if the ingredients arrive on time. The conveyor belt that brings the ingredients is the memory (HBM).

What happens if the belt is slow? The expensive chef stands around waiting for ingredients. The wage clock keeps running. No tokens come out. Cost per token spikes.

So the real bottleneck is often not “hire another chef” (= a more powerful GPU). It is more effective to speed up the belt and pre-stack the frequently used ingredients next to the chef (= memory bandwidth + cache).

KV-cache storage is exactly the act of “pre-stacking frequently used ingredients nearby.” It avoids redoing the same calculation, saving the expensive chef’s time. eSSD makes that pantry cheap and large. This also reduces the load placed on HBM.

This analogy is the investment map for this post. Money flows to wherever the cost per token goes down.

4. Priority 1: Custom ASIC / AI Networking — Broadcom, Marvell, TSMC

The most direct global stack for lowering token cost is custom ASIC and AI networking.

When a hyperscaler uses a chip tailored to its own workload (a custom ASIC), it does the same job more cheaply and with less power than a generic GPU. AI networking is what ties those chips into a cluster. Both lift tokens per watt directly.

Company	Confirmed Facts	View
Broadcom	Q1 FY2026 AI revenue $8.4B (+106% YoY); Q2 AI semiconductor revenue guided ~$10.7B	The core of custom ASIC / networking. Consensus long — valuation discipline needed
Marvell	FY2027 Q1 revenue $2.418B; per Reuters, targets custom-chip revenue >$10B by FY2029	Custom-silicon growth path. Customer concentration is the key variable
TSMC	Per Reuters, AI power demand is shifting chip-design focus from performance to power efficiency → “tokens per watt” a key metric	The foundry every custom ASIC / GPU ultimately passes through. The floor of the token-cost stack

A caution, though. This is already the consensus long the market loves most. A great company and a great price are not the same thing. And custom ASIC revenue is concentrated in a few large customers, so a single customer’s order swings can move the numbers sharply.

5. Priority 2: HBM + eSSD + KV-cache Storage — Samsung Electronics, SK Hynix, Micron

In inference, the real bottleneck is not raw compute but memory bandwidth and cache management. It is exactly the chef-belt analogy from Section 3.

HBM is the ultra-fast belt right next to the chef.
eSSD and KV-cache storage are the pantry where frequently used ingredients are pre-stacked. They avoid repeating the same computation, save expensive GPU time, and ease the load on HBM.

Samsung Electronics’ 1Q26 results materials cited demand for DDR5, SOCAMM2, PCIe Gen6 eSSD, and KV-cache storage. That lineup maps precisely onto the token-cost thesis.

Company	Token-Cost Exposure	View
Samsung Electronics	HBM4E catch-up + DDR5 + SOCAMM2 + PCIe Gen6 eSSD + KV-cache storage	Broadest exposure. Spans multiple layers of the token-cost stack
SK Hynix	Best-quality pure HBM exposure	The best company itself, but the most crowded position in the market
Micron	One of the three global HBM/memory leaders	The U.S. comparable. HBM/DRAM cycle exposure

For a deeper discussion of Samsung Electronics, see Samsung Electronics Stock-Bonus Buyback Analysis and the Samsung Electronics Deep Dive; for SK Hynix, see the SK Hynix Deep Dive.

If forced to pick one, the most balanced Korean exposure to the token-cost thesis is Samsung Electronics. SK Hynix is cleaner as pure HBM beta, but across KV-cache, eSSD and SOCAMM2 Samsung spans more layers — and the market crowding is lighter on Samsung than on SK Hynix.

6. Priority 3: Advanced Packaging / FC-BGA / MLCC / Silicon Capacitor — Samsung Electro-Mechanics, Daeduck

No matter how good a chip is, if the power wobbles and the signal breaks, effective token throughput falls. Advanced packaging, substrates and power-stabilization components are what protect power integrity and signal integrity.

FC-BGA is the high-performance substrate that connects the chip to the board.
MLCCs and silicon capacitors dampen transient voltage swings inside the GPU/HBM package.

A solid layer here lets the expensive chef run at full speed, producing more tokens. So while these parts are indirect, they clearly affect cost per token.

Company	Exposure	View
Samsung Electro-Mechanics	FC-BGA + high-end MLCC + silicon capacitor bundle	High thesis purity. But valuation is already stretched — a “right direction, wrong price” spot
Daeduck Electronics	Second-order FC-BGA beta candidate	The next seat after SEMCO. Direct qualification / volume confirmation is a precondition
Murata, Ibiden	Japanese MLCC / packaging comparables	Tracked together as global peers

For detailed SEMCO analysis, see AI Server Passive-Component Bottleneck: A SEMCO Technical Explainer.

The direction is right. But the price is rich.

The token-cost thesis does point to packaging and substrates, but Samsung Electro-Mechanics has already priced much of that in. So in Korea, a second-order FC-BGA beta like Daeduck Electronics can be the “same direction, less stretched price” alternative — provided direct qualification and volume are confirmed first.

7. Priority 4: Compute Pricing Rail — CME, ICE

The starting point of this post, and its least-crowded second-order idea, is the compute pricing rail.

If the financialization of AI usage truly takes hold, the exchanges where that trading happens benefit structurally — just as exchanges enjoyed steady fee flows once oil futures took hold.

Company	Exposure	View
CME	Plans to launch compute futures with Silicon Data	An option on financialization itself. Pre-launch, liquidity uncertain
ICE	Plans for GPU compute futures contracts with Ornn	Similarly a pre-launch option value

But this is option value, not confirmed revenue. Whether the contracts actually list, and whether volume builds, is unknown. So treat it as a “side-branch of AI financialization” — the smallest-weight, second-order idea.

8. Priority 5: Nvidia / AMD Full-Stack — The Thesis Changes

Nvidia and AMD remain core. They are not dropping out. But the thesis changes.

The old thesis was “GPUs are scarce, so whoever makes them wins.” The token-cost-era thesis is different.

Not “GPU shortage,” but “who provides the full stack with the highest tokens-per-watt and the best AI-factory economics."

Nvidia is strong not simply because its chips are fast, but because it bundles GPUs, networking and software to lower the cost per token of the entire AI factory. In this frame, the key question is not “how fast is the next chip” but “how cheaply does the next system produce tokens.”

9. Priority 6: Power Equipment — A Relative-Value Hedge, Not a Short

Here is a common misconception worth addressing. “If token-cost efficiency matters, power demand falls, so just short the power-equipment names, right?”

No. Power demand itself keeps rising.

The IEA forecasts global datacenter electricity consumption reaching ~945TWh by 2030 (roughly 2x versus 2024).

That is no basis for an outright short on power equipment. Simply shorting names like HD Hyundai Electric or LS Electric is dangerous.

There is only one way to use power equipment here: a relative-value hedge. Use it only, and sparingly, when all three of the following hold at once:

Semiconductor EPS estimates are being revised up, while
Power-equipment EPS estimates stall, and
Power-equipment valuations sit at historic highs.

In other words, it is meaningful only as a relative-value hedge against an “over-stretched power-scarcity premium.” Not an outright short.

10. Discipline: “Financialization = Buy Everything” Is Greed

This thesis is attractive. But its size and the investment decision are separate. Three things must be kept apart.

[Fact]: CME-Silicon Data and ICE-Ornn compute-futures plans were announced. Broadcom’s, Marvell’s and TSMC’s numbers were reported. The IEA’s 945TWh forecast is public.
[Inference]: Once usage gets a price, the industry shifts from a performance race to a cost-per-token race. Then the bottlenecks that lower token cost (custom ASIC, memory, packaging) are structurally favored. This is a reasonable inference from the facts.
[Speculation]: The SHFE futures’ contract spec, underlying-token definition, settlement method and approval timeline are all unconfirmed. Which Korean component maker supplies which customer, and at what revenue share, is also undisclosed.

In particular, the SHFE AI token futures should be treated not as “an already-listed, confirmed product” but as an early signal and option value. And specific customer/share claims — “Samsung supplies this customer’s KV-cache storage,” “Daeduck holds X% of this FC-BGA” — are not verifiable from public sources.

Even a sound thesis ends as a theme unless it is validated in the numbers.

The conclusion is not “AI is being financialized → buy all the related stocks.” It is to selectively pick the bottlenecks that actually lower the customer’s cost per token — at a price you can defend.

11. Next Checkpoints

Checkpoint	Strong Signal	Weak Signal
AI usage financialization	SHFE / CME / ICE actually lists and builds volume on at least one	Stalls in design phase, liquidity falls short
Custom ASIC	Broadcom / Marvell custom-revenue visibility rises	Single-customer order delays / concentration risk surfaces
Memory / KV-cache	Samsung eSSD / SOCAMM2 / KV-cache volume and pricing confirmed	HBM competition intensifies, pricing weakens
Packaging / substrate	SEMCO package margin confirmed; Daeduck qualification / volume	Growth decelerates while valuation stays high
Tokens per watt	TSMC / Nvidia foreground power-efficiency metrics	Only performance headlines repeated
Power equipment	Semis EPS↑ + power EPS stall + power valuation at historic high all met	Any one of the three breaks → hold off on the hedge

Final Interpretation

AI token futures and GPU compute futures are still early. Nothing is listed; no volume has built. But the direction is clear.

The moment AI usage gets a market price, the battleground shifts from “a faster chip” to “a cheaper token.”

From this vantage point, what to buy is not generic AI stocks but the bottlenecks that actually lower the customer’s cost per token. Globally, custom ASIC / networking (Broadcom, Marvell, TSMC) is most direct, and memory / KV-cache storage solves the inference bottleneck. In Korea, the most balanced practical pick is Samsung Electronics — its HBM4E catch-up, DDR5, SOCAMM2, PCIe Gen6 eSSD and KV-cache storage span the token-cost stack most broadly. SK Hynix is the best quality but crowded; Samsung Electro-Mechanics has high thesis purity but a rich price; Daeduck Electronics is a second-order FC-BGA beta candidate. Power equipment is not a short — only a relative-value hedge used when the conditions line up.

The conclusion is discipline. The more AI is financialized, the more “at what price” matters as much as “what to buy.”

Fact / Inference / Speculation / Blocked

[Fact]

Per Reuters, SHFE is early-designing AI-token-price-linked futures. The contract spec, underlying-token definition, settlement and approval timeline are unconfirmed.
CME Group announced plans to launch compute futures with Silicon Data; ICE announced plans for GPU compute futures contracts with Ornn.
Broadcom Q1 FY2026 AI revenue $8.4B (+106% YoY); Q2 AI semiconductor revenue guided ~$10.7B.
Marvell FY2027 Q1 revenue $2.418B. Per Reuters, Marvell targets custom-chip revenue >$10B by FY2029.
Per Reuters, TSMC sees chip-design focus shifting from performance to power efficiency due to AI power demand, making “tokens per watt” a key metric.
The IEA forecasts global datacenter electricity reaching ~945TWh by 2030 (~2x versus 2024).
Samsung Electronics’ 1Q26 results materials cited DDR5, SOCAMM2, PCIe Gen6 eSSD, and KV-cache storage demand.

[Inference]

Once AI usage gets a price, the industry shifts from a performance race to a cost-per-token race.
The bottlenecks that lower cost per token (custom ASIC / AI networking → memory / KV-cache → packaging / substrate) are structurally favored.
In Korea, Samsung Electronics has the most balanced exposure across the token-cost stack; SK Hynix is the best quality but heavily crowded.
Because power demand keeps rising, an outright short of power equipment is inappropriate; use it only as a relative-value hedge when the conditions line up.

[Speculation]

Whether the SHFE AI token futures actually list and trade, and how the underlying token would be defined, is uncertain.
The launch timing and liquidity of the CME / ICE compute futures are not confirmed.
Which global customers specific Korean component makers (Samsung Electronics, Samsung Electro-Mechanics, Daeduck Electronics) supply, and at what revenue share, is undisclosed.

[Blocked]

The SHFE futures’ contract spec, settlement method and approval timeline.
Real-time consensus EPS revision rates and NTM valuations for the Korean names.
Korean suppliers’ customer-level share and revenue mix in the custom ASIC / memory / packaging supply chains.

Disclaimer: For research and information purposes only. Not investment advice. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.

Dell's Earnings Surprise and Korean Semiconductors: Pricing Power Sits With Components, Not Server Assembly

Fri, 29 May 2026 22:00:00 +0900

📚 Context for follow-up posts This is a follow-up to Marvell Q1 FY2027 and Korean Semiconductors. Where the Marvell post argued that “the AI bottleneck migrates from the GPU down to interconnect, substrate, and power,” this Dell post re-checks “who actually earns the money at that bottleneck” — through margin numbers. Related hubs: AI HBM Hub, Korea Semiconductor Value Chain Hub.

TL;DR

Dell reported FY2027 Q1 results on May 28, 2026 (U.S. time), and by any measure the numbers were a large earnings surprise. Revenue grew 88% in a single year, and adjusted EPS jumped 214%. The next day (May 29) the stock surged roughly 33%.

But buried in the same release was one uncomfortable number. No matter how many AI servers Dell sold, gross margin actually fell from 21.6% a year ago to 18.1%. Management’s own stated AI-server operating-margin target is mid-single-digit. In other words, Dell sells an enormous volume of product, but the margin on that product is thin.

That single line is the entire point of this post.

A company that assembles AI servers can post big revenue but thin margins. The real pricing power and the fat margin sit upstream — in memory and high-value components.

Dell itself said demand exceeds supply and that the primary constraint is memory. The moment the company building the servers says “what’s holding me back is memory,” it becomes clear where investors should look.

The Korea-translated priority order:

Priority	Korea Read-Through	Key Names	View
1	Memory pricing power (HBM, server DRAM, eSSD)	Samsung Electronics, SK Hynix	Most direct. SK Hynix is the best company but richly priced; Samsung has HBM4E / eSSD / foundry base-die catch-up optionality
2	FC-BGA, high-end MLCC, silicon capacitor	Samsung Electro-Mechanics	Direct second-order beneficiary, but P/E already in the 200s — “right direction, wrong price” risk
3	Datacenter power (second-order)	HD Hyundai Electric, LS Electric	Not a direct beneficiary of Dell’s print itself; a side-branch of the larger AI-power-demand trend
4	Avoid	Low-margin server assembly / set makers	The very spot Dell just illustrated: big revenue, thin margin

Dell itself had a strong quarter, but the source’s house view was HOLD, with a $475 12-month target. The discipline is not “AI servers are good → buy everything,” but to selectively look upstream, where margin and pricing power flow.

1. Dell’s Print: Revenue Explodes, Margin Compresses

First, the numbers as reported. (Dell)

Item	FY2027 Q1	Commentary
Revenue	$43.8B	+88% YoY; beat consensus (~$35.4–35.7B) by over $8B
GAAP diluted EPS	$5.24	+282% YoY
Non-GAAP diluted EPS	$4.86	+214% YoY; ~+$1.9 vs consensus (~$2.94–2.96)
ISG (servers, networking, storage)	$29.0B	+181% YoY
AI-optimized server revenue	$16.1B	+757% YoY
Traditional server/networking	$8.5B	+92%
Storage	$4.3B	+8%
CSG (PC)	$14.6B	+17% (Commercial $13.0B, +18%)
Operating cash flow	$4.1B	Adjusted FCF $3.165B, shareholder return $2.1B

Orders and guidance were strong too.

AI orders $24.4B; AI backlog $51.3B; over 5,000 AI customers.
Q2 revenue guidance $44–45B; Non-GAAP EPS midpoint $4.80.
FY2027 revenue $165–169B; Non-GAAP EPS midpoint $17.90.
FY2027 AI-server outlook ~$60B (+144%).

Up to this point it looks flawless. But the same release contains one decisive line.

Non-GAAP gross margin fell from 21.6% to 18.1%.

That means AI-server revenue grew more than sevenfold while the company-wide margin rate actually declined. Management put the AI-server operating-margin target at mid-single-digit. In other words, AI servers carry enormous revenue volume but a lower profitability profile than PCs or traditional servers.

2. Why the Margin Is Thin: The Value Sits in the Components

The reason is simple. The most expensive part of an AI server’s cost is not something Dell makes. It is the components: GPUs, HBM, server DRAM, eSSD, CPUs. Dell buys these, puts them in a chassis, connects the cabling, validates the system, and ships it to the customer.

As components get more expensive, revenue grows accordingly. But the margin on those expensive components is captured by the component makers. The assembler sees inflated revenue and thin margin. Dell’s print this quarter laid that structure bare.

The most important clue in management’s call was not the margin figure but the statement that demand exceeds supply, and the explanation that the primary constraint is memory. DRAM, NAND, CPU, HDD, and leading-node allocation were all cited as bottlenecks in the Q&A.

Translated into investor language:

It’s not that they can’t build the boxes — it’s that they can’t get enough of the key components. Whoever holds the scarce part sets the price. And the scarce part is memory.

If demand exceeds supply and the side that controls supply is memory, then pricing power clearly sits with memory. That is the starting point for the Korea read-through.

3. Korea Priority 1: Memory — Samsung Electronics and SK Hynix

The companies that sell the most of the very memory Dell called constrained are Samsung Electronics and SK Hynix.

Their Q1 results prove the structure directly.

Company	1Q26 Revenue	1Q26 Operating Profit	Note
Samsung Electronics	KRW 133.9T	KRW 57.2T	DS (semiconductor) division revenue KRW 81.7T, OP KRW 53.7T
SK Hynix	KRW 52.58T	KRW 37.61T	72% operating margin. HBM + high-capacity server DRAM + eSSD

SK Hynix’s 72% operating margin is no ordinary number. Set it next to Dell’s mid-single-digit AI-server operating-margin target, and you can see at a glance who captures the margin inside the same AI server. The side assembling the server is in the single digits; the side making the memory that goes inside it is in the 70s.

By name, the call differs.

SK Hynix is the strongest company in itself. It holds the core of AI memory: clean HBM exposure, high-capacity server DRAM, and eSSD. But as of May 29 its price is ~KRW 2,333,000, P/E ~22.1x, near its 52-week high. A great company, but the price is already elevated.
Samsung Electronics, as of May 29, is ~KRW 317,000, P/E ~25.4x, near its 52-week high (KRW 323,000). It is not as clean an HBM beta as SK Hynix, but it carries more catch-up optionality. It shipped the industry’s first HBM4E 12-high samples (in Q2), and eSSD plus a foundry-based base die also sit inside the same AI-memory stack.

For deeper discussion of Samsung Electronics, see Samsung Electronics Stock-Bonus Buyback Analysis and the Samsung Electronics Deep Dive.

The point: the moment Dell admitted that memory is the constraint, the pricing power of the two Korean companies holding that constraint structurally strengthens.

4. Korea Priority 2: High-Value Components — Samsung Electro-Mechanics

After memory, the next place pricing power flows is package substrates and power-stabilization components. In Korea, that seat is Samsung Electro-Mechanics.

Its Q1 results:

Item	1Q26	Note
Revenue	KRW 3.2091T
Operating profit	KRW 280.6B
Package-division revenue	KRW 725.0B	+45% YoY

The company directly cited strong demand for high-value MLCCs for AI servers and datacenters, and FC-BGA for AI/server applications. This dovetails precisely with the structure Dell revealed: as AI servers grow, demand for power-stabilization components and high-performance substrates around GPU/HBM packages grows alongside them.

For detailed analysis of Samsung Electro-Mechanics, see:

But here, price must be flagged. As of May 29, Samsung Electro-Mechanics is ~KRW 2,127,000, P/E ~227x. That is a very high number. It means the market has already priced in big expectations.

The direction is right. But the price is expensive.

Dell’s print did confirm that high-value components are positive, but Samsung Electro-Mechanics has already reflected much of that “good” in its price. So between memory (Priority 1) and Samsung Electro-Mechanics (Priority 2), there is a key difference: even the same beneficiary has a very different entry price.

5. Korea Priority 3: Datacenter Power (a Side Branch)

A lower-priority area is datacenter power. HD Hyundai Electric, as of May 29, is ~KRW 1,053,000, P/E ~48x.

The logic that more AI servers means more datacenter power demand, benefiting transformer and power-distribution makers, is sound. But this is not a direct result of Dell’s Q1 print — it is a side branch of the broader AI-power-demand trend. The constraint Dell explicitly named was not power; it was memory.

So power belongs in a second-order bucket: “related, but not the core read-through of this print.”

6. The Spot to Avoid: Low-Margin Assembly

The most important lesson from Dell’s print may lie less in what to buy than in what to avoid.

The fact that Dell sold more than seven times the AI servers and still saw its margin rate fall lays bare the structural limit of low-margin server assembly and set businesses. The revenue looks dazzling, but the money goes to the component makers.

So buying assembly/set-stage names indiscriminately just because they carry the “AI server beneficiary” label is dangerous. Even within the same AI-server revenue growth, the side with pricing power (memory, high-value components) and the side without it (assembly) are completely different investments.

7. Discipline: “AI Servers Are Good = Buy Everything” Is an Overreach

This print is clearly a strong positive. But the size of the positive and the investment call are separate things.

Three layers must be kept distinct.

Fact: Dell revenue +88%, EPS +214%, AI servers +757%, yet gross margin 21.6% → 18.1%. Memory is the primary constraint. These are numbers the company reported.
Inference: Margin and pricing power flow toward memory and high-value components rather than the assembler. In Korea the order is Samsung Electronics / SK Hynix → Samsung Electro-Mechanics → power. This is a judgment reasonably drawn from the facts.
Speculation: How much HBM4E a specific company supplies to Dell, and what share each maker holds in Dell’s servers, is not disclosed. This is not confirmed fact.

In particular, specific customer/share claims such as “Samsung Electronics supplies HBM4E to Dell” or “Samsung Electro-Mechanics holds X% of Dell server FC-BGA” are not confirmed in public disclosures. Such content should be classified as speculation or unverifiable, not fact.

Even a sound thesis ends as a theme unless it is validated in the numbers.

This post’s conclusion is not “AI servers are good, so buy all the related names.” It is: look upstream, where margin and pricing power flow — and select with the entry price in mind.

8. Next Checkpoints

Checkpoint	Strong Signal	Weak Signal
Dell AI-server margin	Mid-single-digit target achieved or raised	Margin falls further or misses target
Dell memory constraint	Memory shortage persists, sustaining pricing strength	Memory supply eases, prices soften
Samsung Electronics	HBM4E mass production / customer qualification, eSSD volume / pricing	Intensifying HBM competition, price declines
SK Hynix	HBM / server DRAM margins hold	2027 supply-overhang signals
Samsung Electro-Mechanics	Sustained package-revenue growth, confirmed margin	Growth decelerates while P/E stays high

Final Interpretation

Dell FY2027 Q1 is not a signal to “buy all AI servers” for Korean semiconductors. The more precise message is:

As AI-server revenue grows, margin and pricing power move up, away from the assembler and toward memory and high-value components.

From this vantage point, the most direct Korean names are memory — Samsung Electronics and SK Hynix. SK Hynix is the strongest company in itself but already priced high; Samsung Electronics carries HBM4E / eSSD / foundry base-die catch-up optionality. Samsung Electro-Mechanics is a direct second-order beneficiary through FC-BGA, high-end MLCC, and silicon capacitor, but at a P/E in the 200s it sits in a “right direction, wrong price” spot. Datacenter power is a side branch; low-margin assembly is the spot to avoid.

The conclusion is discipline. The stronger the positive, the more it matters not just what you buy, but at what price.

Fact / Inference / Speculation / Blocked

[Fact]

Dell FY2027 Q1 revenue was $43.8B (+88% YoY); Non-GAAP diluted EPS was $4.86 (+214%). (Dell)
ISG revenue $29.0B (+181%); AI-optimized server revenue $16.1B (+757%). (Dell)
Non-GAAP gross margin fell from 21.6% to 18.1%; AI-server operating-margin target is mid-single-digit. Memory (DRAM/NAND) was cited as the primary constraint. (Dell)
AI orders $24.4B; AI backlog $51.3B; FY2027 AI-server outlook ~$60B (+144%). (Dell)
Samsung Electronics 1Q26 revenue KRW 133.9T, OP KRW 57.2T; DS-division revenue KRW 81.7T, OP KRW 53.7T.
SK Hynix 1Q26 revenue KRW 52.58T, OP KRW 37.61T, 72% operating margin.
Samsung Electro-Mechanics 1Q26 revenue KRW 3.2091T, OP KRW 280.6B, package-division revenue KRW 725.0B (+45% YoY).

[Inference]

As AI-server revenue grows, margin and pricing power flow toward memory and high-value components rather than the assembler (Dell).
A reasonable Korea read-through priority is Samsung Electronics / SK Hynix (memory) → Samsung Electro-Mechanics (FC-BGA / MLCC / SiCap) → power (second-order).
SK Hynix is the strongest company in itself but carries a high valuation; Samsung Electronics has more catch-up optionality.

[Speculation]

Whether Samsung Electronics directly supplies HBM4E into Dell AI servers, and at what share, is not disclosed.
Samsung Electro-Mechanics’ share in Dell server FC-BGA / MLCC cannot be specified from public disclosures.

[Blocked]

Supplier identity and share by component within Dell’s AI-server BOM.
Samsung Electronics’ and Samsung Electro-Mechanics’ direct revenue exposure to Dell.
Real-time consensus EPS revision rates and NTM valuations for the Korean names.

Sources

Disclaimer: For research and information purposes only. Not personalized investment advice; this is public-information-based analysis. Names cited are for analytical illustration; readers should perform their own due diligence and consult licensed advisors before any investment decision.

Marvell Q1 FY2027 and Korean Semiconductors: The Bottleneck Is Interconnect, Substrate, and Power — Not Just HBM

Thu, 28 May 2026 10:20:00 +0900

📚 Context for follow-up posts This is a follow-up to Marvell & Broadcom Pre-Earnings: Checking Korea AI Semiconductor Bottlenecks. The preview asked whether the single-bet on HBM was broadening into AI ASICs, networking, and power stabilization — this post revisits that question against Marvell’s Q1 FY2027 results. Related hubs: AI HBM Hub, AI Substrate & PCB Hub, Korea Semiconductor Value Chain Hub.

TL;DR

The key takeaway from Marvell Q1 FY2027 is not the EPS beat. It is that the company raised its AI data center growth trajectory for FY2027–FY2028, attributing the upside to custom XPUs, optical interconnects, Ethernet switching, DCI, scale-up/scale-across networking, and XPU attach.

Translating this into Korean semiconductors, the answer is not simply “buy more HBM.” HBM remains the primary thesis, but the incremental alpha confirmed by these results is in the interconnect, substrate, power integrity, and test bottlenecks surrounding the GPU.

Priority order:

Priority	Korea Read-Through	Key Names/Group	View
1	FC-BGA + AI server MLCC + silicon capacitors	Samsung Electro-Mechanics (SEMCO)	Most direct. Already re-rated, so SiCap/FC-BGA margin verification is the next step
2	HBM4, SOCAMM2, eSSD, advanced packaging	Samsung Electronics, SK Hynix	HBM beta intact. New alpha is in memory attach and packaging beyond core HBM
3	Custom ASIC/XPU test sockets & interfaces	ISC, Leeno Industrial, TSE	Structural tailwind, but watchlist until direct revenue is confirmed
4	High-speed PCB/MLB for AI networking	ISU Petasys, Daeduck Electronics, Simtech, etc.	Selective. Not all “AI PCB” exposure carries equal benefit

For Marvell itself, the call is HOLD / Buy watch. Reference price $198.70, 12-month target $225, implying roughly +13% upside. The growth path is strong, but the stock already reflects elevated expectations. For Korean investors, what matters more than Marvell’s own valuation is where the bottlenecks Marvell identified are flowing next.

1. What Actually Matters in the Marvell Print

Marvell reported Q1 FY2027 results after the U.S. close on May 27, 2026. The official figures are as follows. (Marvell)

Item	Q1 FY2027	Commentary
Revenue	$2.418B	+28% YoY, +$18M above guidance midpoint
GAAP EPS	$0.04	Depressed by Celestial AI / XConn acquisition accounting
Non-GAAP EPS	$0.80	Near consensus
GAAP gross margin	52.1%	Improved YoY
Non-GAAP gross margin	58.9%	Held in the high-58% range despite expanding AI mix
Operating cash flow	$638.8M	Record operating cash flow
Q2 revenue guidance	$2.70B ±5%	Range: $2.565B–$2.835B
Q2 non-GAAP EPS guidance	$0.93 ±$0.05	Next-quarter profitability baseline

The print itself was roughly a “modest beat / EPS in-line.” What matters for the stock and for the Korea semiconductor read-through is not a few cents in Q1 — it is the FY2027–FY2028 revenue trajectory management described on the call.

Based on third-party call transcripts, management indicated FY2027 revenue of approximately $11.5B and FY2028 revenue of approximately $16.5B, and raised the FY2027 interconnect growth outlook from roughly 50% to above 70%. These figures come from unofficial transcripts rather than the formal IR filing, but the direction is consistent with the language in the company’s official press release. Marvell’s official release explicitly named the sources of growth as 800G/1.6T optics, 51.2T Ethernet switches, scale-up optical for NPO/CPO, DCI modules, custom XPUs, and XPU-attach. (Marvell, StockAnalysis transcript)

The one-sentence summary:

Marvell confirmed in numbers that AI data center capex is shifting from buying GPUs to building cluster interconnect infrastructure.

2. The Core Story Is Interconnect Architecture, Not SOCAMM

The most dangerous misreading of this earnings call for Korean investors is to reduce it to a “SOCAMM theme.” SOCAMM matters, but it is not the center of gravity here.

The actual hierarchy is as follows:

Marvell Growth Pillar	Why It Matters	Korea Semiconductor Translation
Custom XPU / custom ASIC	Signal that hyperscalers are expanding proprietary AI silicon beyond NVIDIA GPUs	HBM customer diversification, package substrates, test sockets
Optical interconnect	800G/1.6T, DCI, scale-up optics are the binding constraint for AI cluster expansion	High-speed PCB/optical is selective; SEMCO FC-BGA and power stabilization are structural
Ethernet switching	51.2T, 100T, 200T roadmap means rising dollar content in AI networking silicon	FC-BGA for network ASICs, high-speed boards, inspection/test
XPU attach	CXL, NIC, memory attach, inference KV cache exposure	Samsung Electronics SOCAMM2, eSSD, server DRAM; memory IP options such as OpenEdges
NVLink Fusion	Custom silicon coexisting inside the NVIDIA ecosystem	Supply chain expansion rather than an NVIDIA-vs-ASIC binary

The NVIDIA-Marvell announcement reinforces the same direction. NVIDIA invested $2 billion in Marvell; Marvell supplies NVLink Fusion-compatible custom XPUs and scale-up networking. The two companies also collaborate on silicon photonics and AI-RAN. (Marvell NVIDIA)

The implication is straightforward:

An AI data center is not a GPU box. It is a system in which GPUs, custom XPUs, HBM, optics, switches, retimers, CXL, NICs, FC-BGA substrates, and MLCCs move as a single organism.

Korean semiconductor investing must therefore shift the frame from “should we hold just the large-cap memory names” to “which bottlenecks below memory are translating into numbers.”

3. Korea Read-Through Priority 1: Samsung Electro-Mechanics (SEMCO)

The cleanest Korean translation of the Marvell print is Samsung Electro-Mechanics. Three reasons:

First, every growth pillar Marvell highlighted — custom XPUs, Ethernet switches, optical interconnects, XPU attach — requires high-speed, high-density packaging and substrates. That maps directly to FC-BGA demand.

Second, AI servers draw large instantaneous currents at low voltages. Suppressing voltage fluctuations around GPU, HBM, and XPU packages requires power stabilization components: MLCCs and silicon capacitors.

Third, Samsung Electro-Mechanics already holds both layers. In its Q1 2026 earnings, the company reported Package Solution revenue of KRW 725.0 billion, up 45% YoY and 12% QoQ, citing expanding supply of high-value substrates for AI accelerators, server CPUs, and networking applications. (Samsung Electro-Mechanics Q1)

Additionally, Samsung Electro-Mechanics signed a silicon capacitor supply contract worth KRW 1.5570 trillion with a large global company. The contract runs from January 1, 2027 through December 31, 2028. The component is described as improving power supply stability inside AI server GPU and HBM packages. (Samsung Electro-Mechanics SiCap)

The investment thesis for Samsung Electro-Mechanics has now shifted:

Old Frame	New Frame
Smartphone MLCC & camera module cyclical	AI package power integrity + FC-BGA platform
Mobile demand recovery	AI accelerator, server CPU, network ASIC demand
Commodity MLCC cycle	High-capacitance, low-ESR, ultra-thin, high-reliability MLCC/SiCap mix
Comparison vs. Ibiden or Murata individually	Hybrid comparison: MLCC + FC-BGA + SiCap

This conclusion is not the same as “buy at any price.” Samsung Electro-Mechanics has already seen meaningful re-rating on the silicon capacitor contract and the AI infrastructure narrative. The next confirmation variables are therefore not price momentum but gross margin, Package Solution OPM, SiCap production yield, and further customer diversification.

4. Samsung Electronics & SK Hynix: HBM Beta Intact, New Alpha Is at the Periphery

The Marvell print is not negative for HBM — if anything, the opposite is true. Even as custom XPUs and scale-up networking grow, memory dollar content per cluster does not shrink. As AI models evolve toward agentic AI, reasoning, and mixture-of-experts architectures, data movement and memory requirements grow larger, not smaller.

From an investment standpoint, however, “HBM is positive” is already the consensus thesis. What the Marvell call added incrementally:

The HBM customer base is broadening from a single NVIDIA GPU structure toward hyperscaler custom XPUs.
XPU attach connects CXL, NIC, memory attach, and KV cache, extending influence to server DRAM, SOCAMM, and eSSD.
As AI clusters scale, the value of the packaging, substrates, and power integrity that link memory and compute rises alongside them.

Samsung Electronics carries optionality across multiple dimensions here. HBM4, HBM4E, SOCAMM2, PM1763 SSD, foundry, and advanced packaging all sit within the same AI infrastructure stack. At NVIDIA GTC 2026, Samsung presented HBM4E, SOCAMM2, and the PM1763 SSD as part of a collaborative AI infrastructure product family with NVIDIA. (Samsung Semiconductor)

SK Hynix remains the purest HBM beta. Looking solely at the Marvell print, however, the new incremental alpha is larger in Samsung Electro-Mechanics, test sockets, and high-speed substrates growing alongside HBM than in SK Hynix itself. SK Hynix is the primary beneficiary, but it is also where the most market attention is already concentrated.

5. Test Sockets: The Quiet Beneficiary of Custom ASIC Proliferation

The importance of custom revenue in the Marvell call is not about chip volume alone. The key is SKU count and qualification cycles.

In a world where AI accelerators are standardized around a single NVIDIA GPU, test component demand is relatively predictable. Conversely, as hyperscaler-specific custom XPUs, XPU attach, CXL, NICs, switch ASICs, and DPUs proliferate, test conditions and socket designs fragment further.

In this scenario, test sockets and interface components can see three simultaneous tailwinds:

Variable	Direction	Reason
Volume	Increasing	More custom ASIC, network ASIC, and memory attach SKUs
ASP	Rising	Higher pin counts, faster signal speeds, higher power test complexity
Replacement cycle	Structural	Generational transitions and per-customer qualification repeats

The Korean names on the watchlist are ISC, Leeno Industrial, and TSE. Conviction should be held lower here, however. Whether Korean test socket companies are directly embedded in Marvell’s or a specific hyperscaler’s custom XPU chain cannot be determined from public disclosures alone. The current assessment is therefore “possible beneficiary” — not “confirmed customer mapping."

The metrics to track in quarterly results: AI/HPC logic revenue, new customer count, high-value socket mix, and OPM defense.

6. Generic PCB Is Not a Blanket Buy

The Marvell print is positive for AI networking and optical interconnects. But the inference that “AI networking is good → all PCBs are good” is dangerous.

The benefit concentrates in companies that satisfy the following conditions:

Ability to process low-loss materials for high-speed signal integrity.
Exposure to high-layer-count MLB or high-value package substrates.
Qualified supplier status with AI server or network equipment customers.
Confirmed ASP, layer count, and area expansion — not just volume growth.

More GPUs and XPUs do not translate into proportionally more substrate units. In an architecture where multiple chips integrate more densely onto a single high-performance package and board, what matters is substrate area, layer count, material difficulty, and yield — not unit shipment volume.

Grouping ISU Petasys, Daeduck Electronics, Simtech, TLB, and Korea Circuit into one undifferentiated basket is therefore imprecise. The actual read-through from Marvell’s results is closer to: “be selective — identify names with high-layer-count substrates for networking and materials qualified for high-speed signal environments.”

7. MRVL Valuation: A Great Company and a Great Entry Price Are Not the Same Thing

The standalone view on Marvell is HOLD / Buy watch.

Item	Detail
Reference price	$198.70, regular-session close 2026-05-27
12-month price target	$225
Implied upside	~+13.2%
Valuation framework	FY2028E EV/Sales
Core view	Growth path raised, but valuation already elevated

The base-case target price calculation:

Target price = (FY2028E Revenue × Target EV/Sales − Net Debt) ÷ Diluted shares

Assumptions: FY2028E revenue $16.5B, target EV/Sales 12.5×, net debt ~$1.117B, diluted shares 915M. This produces a target of approximately $224, rounded to $225.

For Marvell to re-rate toward Broadcom-level multiples, three things are required:

Custom silicon revenue must be confirmed as a recurring program — not a single-customer event.
Non-GAAP gross margin must hold in the 58–59% range even as interconnect and switching grow.
Supply-chain prepayments must convert into actual revenue ramp and free cash flow.

In short: Marvell has become a great company. Whether it is a great entry price is a separate question.

8. Next Checkpoints

Checkpoint	Strong Signal	Weak Signal
Q2 FY2027 revenue	At or above $2.835B (guidance upper end)	Below $2.70B midpoint
Data Center revenue	High-teens or better sequential growth	Sequential growth deceleration
Non-GAAP GM	59.25%+ or upper-end defense	Below 58.25%
Interconnect	FY2027 +70%+ outlook maintained or raised	800G/1.6T ramp slowing
Custom XPU	FY2028 2×+ growth, FY2029 $10B+ visibility	Per-customer ramp delays
Scale-up switching	Tier-1 customer volume production confirmation	Many engagements but no revenue
Korea read-through	SEMCO Package Solution, SiCap, FC-BGA numbers confirmed	Theme strong but margins/orders absent

Per-name Korea confirmation variables:

Name/Group	What to Track
Samsung Electro-Mechanics	Package Solution growth rate, AI network FC-BGA revenue, SiCap yield/margin, additional long-term contracts
Samsung Electronics	HBM4 customer qualification, SOCAMM2 actual shipments, eSSD pricing/volume, foundry/packaging customers
SK Hynix	HBM4 ramp, customer diversification, 2027 supply overhang risk
ISC, Leeno Industrial, TSE	AI logic/test socket revenue, new customers, high-value mix, OPM
PCB/MLB	AI network customer qualification, ASP expansion, low-loss materials and higher layer counts

9. Invalidation Conditions

The conditions under which this thesis weakens are clear:

Marvell Q2 revenue comes in below the $2.70B midpoint and Data Center growth decelerates.
Non-GAAP gross margin falls below 58.25%, confirming that custom/interconnect mix is diluting margins.
The FY2028 $16.5B revenue outlook is revised lower.
Custom XPU and XPU attach are tied up in specific customer schedule delays.
Samsung Electro-Mechanics SiCap is confirmed as low-margin revenue, or FC-BGA growth decelerates.
AI/HPC logic revenue growth fails to appear in test socket company results.
HBM lead times shorten and 2027 supply overhang signals emerge.

Final Interpretation

Marvell Q1 FY2027 is not a signal to “just buy HBM” for Korean semiconductors. The more precise message is:

As AI clusters scale, the bottleneck migrates down from the GPU toward interconnects, substrates, power stabilization, and test.

From this vantage point, the most direct Korean name is Samsung Electro-Mechanics. SEMCO’s MLCC, FC-BGA, and silicon capacitor businesses all sit inside the same AI package bottleneck. Samsung Electronics and SK Hynix remain central as the core HBM plays, but the incremental alpha that Marvell’s results newly surfaced is larger at the HBM periphery. ISC, Leeno Industrial, and TSE are second-order beneficiaries of the custom ASIC buildout — but watchlist is the right posture until direct revenue is confirmed.

The conclusion is not to buy indiscriminately. Even a sound thesis ends as a theme unless it is validated in the numbers. After this quarter, the variables to watch in Korean semiconductors are not price but Package Solution revenue, SiCap margin, AI logic test socket revenue, and high-speed substrate ASP.

Fact / Inference / Speculation / Blocked

[Fact]

Marvell Q1 FY2027 revenue was $2.418B, +28% YoY; Q2 revenue guidance is $2.70B ±5%. (Marvell)
Marvell Q1 non-GAAP gross margin was 58.9%; Q2 non-GAAP gross margin guidance is 58.25–59.25%. (Marvell)
Marvell named 800G/1.6T scale-out optics, 51.2T Ethernet switches, scale-up optical, DCI modules, custom XPUs, and XPU-attach as growth drivers. (Marvell)
NVIDIA invested $2 billion in Marvell; Marvell announced it will supply NVLink Fusion-compatible custom XPUs and scale-up networking. (Marvell NVIDIA)
Samsung Electro-Mechanics Q1 2026 Package Solution revenue was KRW 725.0 billion, up 45% YoY and 12% QoQ. (Samsung Electro-Mechanics Q1)
Samsung Electro-Mechanics signed a silicon capacitor supply contract worth KRW 1.5570 trillion; contract period is January 1, 2027 through December 31, 2028. (Samsung Electro-Mechanics SiCap)

[Inference]

It is reasonable to map Marvell’s growth axes to Korean names in the following order: SEMCO FC-BGA/MLCC/SiCap, Samsung Electronics and SK Hynix memory attach, test sockets, and high-speed substrates.
HBM remains the primary thesis, but the incremental alpha newly surfaced by these results is larger in the interconnect, packaging, power integrity, and test layers than in the large-cap memory names.
Samsung Electro-Mechanics’ re-rating should be interpreted as a transition to an AI infrastructure passive/substrate platform — not a recovery in conventional MLCC demand.

[Speculation]

Whether Samsung Electro-Mechanics’ SiCap customer is linked to a specific North American hyperscaler or AI accelerator supply chain is speculated by the market, but the counterparty has not been disclosed publicly.
Whether domestic Korean test socket companies are directly embedded in Marvell’s or any hyperscaler’s custom XPU program cannot be determined from public disclosures.
AI-RAN could create long-term opportunities in Korean RF and network semiconductors, but it is premature to treat it as a near-term earnings catalyst in 2026.

[Blocked]

Direct Korean supplier status in Marvell’s custom XPU, optical, and switch programs by company.
Customer identity, per-product margin, and monthly ramp pace for Samsung Electro-Mechanics’ SiCap contract.
AI logic revenue breakdown by customer for ISC, Leeno Industrial, and TSE.
Real-time NTM PER, EV/EBITDA, and consensus EPS revision rates for Korean PCB and substrate names as of current 2026 data.