An Era Where Even Models Cannot Cross Borders

SOVEREIGN AI · 2026.06.14

Even models cannot

cross borders

How Sovereign AI is Rewriting the Value of Five Companies

NVIDIA NVDA · Broadcom AVGO · Marvell MRVL · LG CNS 064400 · LG Electronics 066570

Where Claude 5 Stopped — The Shift to Inference and the Competition for Compute Sovereignty

June 14, 2026 · Gogooma Team 3rd Round Discussion · 8-Topic Comprehensive In-depth Investigation on Business/National Investment
Intelligence with Borders · Map of Money · Arms Dealers and Kings · Bull/Super Bull Take-Profit Prices

Onion (Macro/Theme) · Acorn (Industry/Reality) · Peanut (Risk/Value) · Carrot (Chart/Supply & Demand) · Gogooma (Summary)

Prologue

The Day Claude 5 Went Dark, Intelligence Gained Borders

On June 9, 2026, one of the world's most powerful AI models was unveiled, and three days later, on June 12, it disappeared for all users outside the United States. This was because the U.S. government ordered that "foreigners cannot access this model." It was the first moment that software, not chips, became a target of export controls like a weapon.

The flagship models 'Fable 5' and 'Mythos 5' released by Anthropic on June 9 were deactivated globally on June 13 following the U.S. Department of Commerce's export control directive (blocking foreign access) issued on June 12. Since there was no way to verify nationality in real-time, it effectively became unusable for everyone, including Americans. While the stated reason was the potential for bypassing cybersecurity safety guardrails, the essence was greater — Frontier AI models were classified as national security assets, and at that moment, borders were drawn on 'intelligence'.

This incident imprinted one fact on decision-makers in every country: Riding on another country's cutting-edge AI means accepting the risk that the switch could be flipped off at any time. Therefore, domestic models, domestic data centers, and domestic chips — so-called 'Sovereign AI' — are no longer a choice but a security necessity. This report tracks where that massive transition is flowing money and how the value of the five companies standing at the crossroads — NVIDIA, Broadcom, Marvell, LG CNS, and LG Electronics — is being re-evaluated. Based on three rounds of intense debate between bullish and cautious teams, we provide answers solely through numbers like national investment amounts and corporate performance.

Chapter 1

Intelligence with Borders — Why Sovereign AI Has Become a Security Necessity

Export controls were originally in the realm of physical goods: advanced semiconductor equipment, weapons, and encryption technology. However, in June 2026, that boundary crossed over into 'intelligence woven by software' itself. This is the inflection point that elevated Sovereign AI from a theme to a security agenda.

The gist of the incident is as follows: On June 12, the U.S. Department of Commerce ordered Anthropic to block foreign access to Fable 5 and Mythos 5, and the company completely deactivated both models the next day before rolling them back to a lower-tier version. It was the first precedent in history where export controls were applied to 'AI model software' rather than chips. Anthropic promised a swift recovery, citing a "narrow technical misunderstanding," but the message received by the market was clear — relying on U.S. models means embracing policy risk.

This realization poured fuel on a trend already in motion. Market research firms expect the Sovereign AI market to grow from approximately $19 billion–$79 billion in 2026 to $177 billion–$727 billion by 2035(28% annual growth). Governments worldwide have begun pushing for their own foundation models (LLMs), local data centers, and domestic AI semiconductors simultaneously. Once intelligence gained borders, every country started wanting to keep intelligence within its own territory. Money flows where that desire is directed.

Key takeaway of the first chapter. The core is the precedent of 'model control' itself. While a switch turned off once can be turned back on, countries that have seen the risk of dependency will not go back. Sovereign AI is now a hard-to-cut demand tied to national security budgets, not business cycles.

Chapter 2

Map of Money — Who Is Pouring In How Much?

To gauge the scale of Sovereign AI, one only needs to look at the investment commitments by country. The total reaches trillions of dollars, and the primary destination for that money is data centers and AI semiconductors.

The U.S. is aiming for $500 billion and 10GW by 2029 through the 'Stargate' project, with a cumulative $400 billion and 7GW already committed. The Middle East is even more aggressive — Saudi Arabia, with the Public Investment Fund (PIF)'s HUMAIN at the center, has seen the Crown Prince raise investment promises to $1 trillion, while the UAE is building a $100 billion AI infrastructure fund called MGX and a 5GW campus. The European Union has pledged a €200 billion 'InvestAI' initiative, and Japan has announced ¥10 trillion in support for semiconductors and AI. South Korea is deploying a 150 trillion won National Growth Fund over five years.

Country/Region	Key Program	Scale
USA	Stargate (OpenAI, Oracle, etc.)	Goal: $500B, 10GW (Cumulative $400B, 7GW)
Saudi Arabia	HUMAIN (PIF)	Crown Prince promise $1T, NVIDIA initial $10B, 500MW
UAE	MGX/Stargate UAE (G42)	Fund $100B, Campus 5GW
EU	InvestAI/AI Gigafactory	€200B (4–5 Gigafactories)
Japan	Semiconductor/AI support + Foundation model	¥10T + ¥1T
South Korea	National Growth Fund/Government AI budget	150T KRW over 5 years (2026 AI 6T), AI budget 10.1T KRW

This national funding explodes as it joins the capital expenditures of Big Tech. The combined 2026 capex of the 4 major hyperscalers (Amazon, MS, Google, Meta) is surging 60–77% year-over-year to $660 billion–$725 billion(Goldman Sachs forecasts a cumulative $1.15 trillion from 2025–2027). The evidence that this money is actually flowing into chips is recorded on NVIDIA's books — 'Sovereign AI' revenue from national clients tripled in one year to approximately $30 billion (14% of the total), with those clients spanning 40 countries.

$660B–$725B

2026 total capex of 4 major hyperscalers (+60–77% YoY) — National capex is layered on top of this

Chapter 3

The Shift in Center of Gravity — From Learning to Inference, From GPU to ASIC

The center of gravity for AI computing is shifting from 'training' (teaching the model) to 'inference' (using the model). This shift is changing the power landscape of the chip market — moving from a single narrative of general-purpose GPUs to the emergence of a second protagonist: custom ASICs.

Numbers testify to the transition. The share of inference in total AI computing is soaring from about one-third in 2023 to half in 2025, and about two-thirds (67%) in 2026. Because inference tasks are repetitive and predictable, power efficiency (performance per watt) and Total Cost of Ownership (TCO) are more decisive than general-purpose capability. This is where custom semiconductors (ASICs) hold a 40–65% TCO advantage over general-purpose GPUs (Image generation firm Midjourney lowered its monthly costs by 65%, from $2.1 million to $700,000, by switching to Google TPUs).

As a result, the 2026 growth rate for custom ASIC shipments is 44.6%, about three times that of general-purpose GPUs (16.1%). Within AI servers, the share of GPU systems will drop to 69.7% for the first time, while ASICs rise to 27.8% (projected to be about 40% by 2030). And who holds the design and production of these custom chips? Broadcom and Marvell dominate with a combined share of about 95%.

Key hypothesis, team verification. "If countries and Big Tech challenge themselves with proprietary inference chips, Broadcom's opportunity to help with those designs grows; until then, the competition to secure NVIDIA chips as much as possible keeps chip prices elevated." The team concluded that this hypothesis holds true simultaneously. In the inference era, ASICs do not 'replace' GPUs but 'delineate segments' — large-scale repetitive inference goes to ASICs, while training, general-purpose, initial inference, and software ecosystems remain with NVIDIA GPUs. During the transition, demand for both is strong; while competition for NVIDIA's volume (Blackwell sold out) supports prices, Broadcom's ASIC revenue skyrockets.

However, the pressure over time is clear. Forecasts suggest NVIDIA's inference market share could drop from over 90% currently to 20–30% by 2028. This is not a 'collapse' but a 'ceiling for re-evaluation' — absolute revenue continues to grow, but the monopoly premium in inference is trimmed. This tug-of-war is the core axis determining the value of NVIDIA and Broadcom.

Chapter 4

NVIDIA — The Most Honest King

One thing the four team members agreed on most quickly was that NVIDIA is the company among the five where 'value and performance are most honestly aligned.' Growth is explosive, yet the forward multiple is closer to the market average.

The scale is overwhelming. FY2026 revenue was $215.9 billion (+65%), with data centers alone accounting for $193.7 billion, and in the following quarter, it accelerated to $81.6 billion in quarterly revenue (+85%) and $75.2 billion in data centers (+92%). Non-GAAP gross margin recovered to 75%, and the company announced $80 billion in stock buybacks and a 25-fold increase in quarterly dividends. Yet, the forward P/E ratio is about 20.7x, which is not expensive considering the growth rate.

Indicator	Value	Meaning
FY2026 Revenue / Data Center	$215.9B / $193.7B	+65% / +68%
Recent Quarter Revenue / DC	$81.6B / $75.2B	+85% / +92% acceleration
Sovereign AI Revenue	Approx. $30B (14% of total)	3x YoY, 40 countries
FY2027 Consensus Revenue	Approx. $367B–$391B	Linked to national/Big Tech capex
Value / Consensus Target Price	Fwd P/E 20.7x / $299	+46% vs current price $205

NVIDIA is the 'King of Sovereign AI' because it sits at the apex of the path where national funds flow into chips. Saudi HUMAIN is initially introducing 18,000 GB300 units and will add hundreds of thousands more. The next-generation platform, 'Vera Rubin,' will have its first shipments in the summer of 2026, and the next-generation high-bandwidth memory (HBM4) for these chips is estimated to be supplied 60–70% by SK Hynix and 25–30% by Samsung Electronics, leading to a direct benefit for the Korean memory industry. Blackwell is sold out through mid-2026 (backlog of 3.6 million units), so supply cannot keep up with demand — this supply constraint supports prices and margins.

The King's Weakness — Which is Why It's Honest

NVIDIA has two shadows. First, ASIC encroachment in inference (market share 90%+ → projected 20–30% by 2028) suppresses the ceiling for long-term multiples. Second, FY2027 consensus revenue relies on Big Tech capex nearly doubling again, but the Return on Investment (ROI) is not yet proven (OpenAI spends $60 billion annually on compute, but revenue is in the $13 billion range). Nevertheless, the team ranked NVIDIA first because, among the five companies, it is the one where 'revenue most clearly follows value'.

Chapter 5

Broadcom and Marvell — Arms Dealers of the Inference Era

There is an old saying that those who sell pickaxes earn more reliably than those who mine for gold. In the era of inference, Broadcom is the arms dealer selling those pickaxes — custom AI chips. It is also the company most directly pointed to by the hypothesis of the inference transition established earlier.

Broadcom's AI semiconductor revenue curve is steep. It recorded $10.8 billion (+143%) in the previous quarter, and the company projected FY2026 AI revenue at approximately $56 billion (+180%) and over $100 billion for 2027. AI backlog stands at $73 billion. The client list is the moat — Google (TPU), Anthropic (1GW in 2026, +5GW in 2027), OpenAI (10GW by 2029, single $10B order), Meta (3GW by 2028), and 2 additional clients. Broadcom holds about 70% of custom ASIC co-design, and 95% when combined with Marvell.

Indicator	Value	Meaning
Previous Quarter Revenue / AI Semiconductor	$22.19B / $10.8B	+48% / +143%
AI Revenue Trajectory	FY25 ~$20B → FY26 ~$56B → FY27 $100B+	Directly linked to national/Big Tech inference transition
AI Backlog / Free Cash Flow	$73B / $10.26B (46% of revenue)	Visibility + Cash Engine
Custom ASIC Market Share	Approx. 70% (95% combined with Marvell)	Oligopoly barrier to entry
Value / Consensus Target Price	Fwd P/E 24.3x (TTM 91x) / $522	+37% vs. current price of $382

However, even arms dealers have their shadows. Immediately after the earnings announcement on June 3, the stock price plummeted by -12.59%. While revenue and profit were strong, the AI guidance for the next quarter ($16 billion) fell short of the consensus ($17.2 billion), and with a trailing P/E ratio of 91x, it was 'priced for perfection.' However, our team's chart analyst interpreted this not as a trend reversal, but as a dip—because it held the 200-day moving average (around $350) and rebounded. Another weakness is customer concentration. If the capital expenditure schedules of a few key large clients are delayed, Broadcom's numbers wobble. Therefore, Broadcom is a company that is 'strong in business but expensive in price' — a stock to buy on pullbacks.

The second arms dealer — Marvell (MRVL)

There isn't just one company selling pickaxes. Alongside Broadcom, Marvell is the No. 2 player, splitting about 95% of the custom ASIC market. While its market share is 20~25%, trailing the leader, Marvell possesses another weapon — 'optical connectivity' that links AI data centers. As data centers grow, the pathways between chips and servers become bottlenecks; Marvell is strong in signal processing (DSP) for optical transceivers and high-speed switches.

Metric	Value	Significance
Stock Price / Market Cap	$279.70 / Approx. $245 billion	No. 2 in custom ASIC (20~25% share)
Custom Silicon Roadmap	FY28 2x → FY29 $10 billion+	Targeting 20% share of the $94 billion 2028 data center semiconductor market
Key Clients / Connectivity Weapons	AWS Trainium, MS Maia, Google / 800G, 1.6T optics, 102.4Tbps switches	Joined NVIDIA NVLink Fusion + $2 billion investment (2026.3)
Valuation / Consensus Target Price	TTM P/E 95.8x / Average $225 (High $345)	Current price $279.70, significant expectations for ASIC and optics reflected

Marvell's sovereign AI opportunity shares the same vein as the leader but is more focused on 'connectivity.' Countries, telecom carriers, and public data centers that want their own inference chips may have the models and operational capabilities, but they struggle to complete the 3nm chip design, high-bandwidth memory (HBM) interface, and optical connectivity on their own. Marvell is a co-design partner that helps with those designs and a company that resolves 'connectivity,' the true bottleneck of inference clusters.

However, Marvell is 'more expensive and narrower' than Broadcom. Its trailing P/E ratio of 95.8x is higher than the leader's (91x), and because the weight of one client, Amazon, is significant, earnings volatility is high if that client's chip schedule (Trainium) fluctuates (one brokerage firm downgraded its investment opinion for this reason). Therefore, our team classified Marvell as _'A complementary bet to Broadcom — a more aggressive but higher-volatility No. 2'_The strategy is to buy the ASIC theme, leaning heavily on the leader while adding the No. 2 player.

Chapter 6

LG CNS — Korea's Sovereign Full-Stack

Among the two companies considered key positions for Korea's sovereign AI, the one our team viewed as having 'clearer substance of direct benefits' was LG CNS. It possesses a full-stack that connects models, infrastructure, and chips, and this is already reflected in its performance.

First, the quality of its performance has changed. It achieved record-breaking results in 2025 with 6.1295 trillion KRW in revenue and 555.8 billion KRW in operating profit (9.1% margin), and in Q1 2026, while revenue grew by 8.6%, operating profit increased by 19.4% and net profit by 41.2%, confirming margin leverage. The engine of growth is AI and cloud, which account for 58~60% of total revenue.

Business Axis	Details	Figures
Performance Leverage	2026 Q1 Revenue/Operating Profit/Net Profit	+8.6% / +19.4% / +41.2%
Data Center (DBO)	No. 1 in domestic design, construction, and operation	60% design market share, long-term order for Samsung DC 780 billion KRW (~2035)
Sovereign Model/Chip	EXAONE + Cohere LLM, Furiosa domestic NPU	Anti-NVIDIA public AI ecosystem
Big Tech Alliance	OpenAI, MS, Google, AWS, Palantir + Claude	Implementation partner regardless of model winner
Valuation / Consensus Target Price	Fwd P/E 18x, P/B 3.08 / 95,333 (High 150,000)	Current price 92,800

The meaning of full-stack is as follows: LG CNS ① directly designs and operates data centers (60% domestic design market share, 780 billion KRW in confirmed revenue over 10 years from the Samsung data center), ② layers LG AI Research's EXAONE model and its own agent platform 'AgenticWorks,' and ③ combines it with NPU from the domestic AI semiconductor company FuriosaAI to aim for a 'public AI infrastructure that runs without NVIDIA'. It is the company entrusted with implementing the funds the government pours into sovereign AI (2026 AI budget of 10.1 trillion KRW, National Growth Fund of 150 trillion KRW). The 'M.A.P. (Mobility, AI infrastructure, Physical AI)' alliance, made official by the meeting between Chairman Koo Kwang-mo and NVIDIA's Jensen Huang on June 8, added a global link to this.

If you look at LG CNS only through its core business multiples, you get trapped in the short-term frame that 'P/B 3.08x is already expensive.' However, the three options of Data Center (780 billion KRW confirmed Samsung revenue), Physical AI (AgenticWorks), and CBDC (national payment network) are barely captured in its core business profit, so if you add them separately as business value (SOTP), the target price opens up above the core business multiple. What to watch out for is volatility — the 52-week range of 54,000 to 153,000 KRW, and foreign ownership of 5.88% means there are fewer hands to support it if the theme cools down. Performance leverage (operating profit +19.4%) is real, and the options remain separately alive on top of that.

Chapter 7

LG Electronics — The pickaxe of cooling, and the weight of options

As AI data centers increase, there is one thing that inevitably increases with them — cooling to lower heat. LG Electronics holds this 'second pickaxe.' By the standards of home appliance stocks, it looks expensive, but when you add future options like cooling, actuators, and automotive electronics as business value (SOTP), the story changes.

LG Electronics' AI exposure lies in the data center cooling of its HVAC division. It involves chillers that catch the heat of AI servers, as well as liquid cooling (direct-to-chip) and immersion cooling solutions. Chiller orders tripled in 2025 compared to the previous year, and the Serviceable Addressable Market (SAM) for data center cooling will grow from $1.6 billion in 2026 to $12.7 billion in 2030. It is also in the final stages of NVIDIA's Coolant Distribution Unit (CDU) certification. The company targets 20 trillion KRW in HVAC revenue by 2030.

Metric	Value	Significance
Core AI Exposure	Data center cooling (chiller, immersion)	Chiller orders 3x (2025), final stages of NVIDIA CDU certification
Cooling SAM	$1.6B → $12.7B (2026→2030)	Targeting 20 trillion KRW in 2030 HVAC revenue
2025 Performance	Revenue 89.2 trillion / Operating Profit 2.25 trillion	Operating profit -27.5% (core business slowdown)
Valuation / Target Price	Estimated P/E 21x, P/B 1.59	Domestic consensus 166,750 vs. foreign maximum 400,000 — large gap in target price

The problem is the price. With operating profit down 27.5% in 2025, the stock price has exceeded the consensus target price (166,750 KRW) by 35% on expectations for AI cooling and robots. Revenue for the cooling business takes 3~4 quarters from order to recognition, and the volatility of estimates is also high. This reveals why target prices differ significantly. The domestic consensus (average 166,750 KRW), which looks only at home appliance multiples, is a conservative figure that effectively sets future options like actuators, cooling, and the NVIDIA alliance to zero. On the other hand, if you add those options as business value (SOTP), the core business value alone exceeds the current price, and the 400,000 KRW range target price suggested by foreign firms is a product of that method. In other words, 'expensive' is just the yardstick of home appliances; by the yardstick of adding options, there is room for upside. However, the slowdown in core operating profit (-27.5%) and volatility are clear burdens.

Chapter 8

Four perspectives — Where bullishness and caution collide

Even regarding the same facts, perspectives differ. In three discussions, our team clashed directly over whether 'this revaluation is justified despite core business slowdown and high valuations.' That conflict made the conclusion more solid.

Onion · Macro/Sovereign Theme (Bullish)

"Overheating comes when earnings (E) stop, not from price. But trillions of dollars in national capex support that E. While reliance on corporate bonds for Big Tech may be risky, sovereign capex comes from national balance sheets — and NVIDIA is a creditor lending money, not the one taking on debt. Focus on the mainstream (the two US companies)."

Peanut · Risk/Valuation (Cautious)

"The trap is that those earnings (E) stand on the 'assumption of return on investment for capex bought with debt.' 95% of enterprise generative AI has zero measurable profit, and OpenAI spends $60 billion on computing while revenue is $13 billion. I acknowledge that NVIDIA is the most honest. But honesty is not risk-free. A mechanical stop-loss of -15% per stock is non-negotiable."

Acorn · Industry/Substance (Balanced)

"The bubble theory is only half right. Big Tech capex is based more on operating cash flow than debt, and Broadcom's backlog and Blackwell sell-outs are not fake demand. The encroachment of ASICs is not NVIDIA's 'collapse' but a 'valuation limitation.' In Korea, LG CNS is more direct than LG Electronics through execution revenue like Furiosa NPU and DBO — I rank it first for Korean sovereign beta."

Carrot · Chart/Supply & Demand (Practical)

"Capital flows are overwhelmingly concentrated in the US Big 2. Broadcom's -12.59% plunge is a dip that held the 200-day line, not a trend reversal. LG CNS has low foreign ownership of 5.88%, so it has the greatest elasticity when buying turns. If you have new capital, lean on the two US companies, but absorb volatility for the two Korean companies through fractional buying."

After three rounds, what was agreed upon and the boundaries that remained until the end became clear. Consensus — ① Sovereign AI is structural demand tied to security budgets, ② The ranking is NVIDIA > Broadcom > LG CNS > LG Electronics, ③ NVIDIA is the most honest valuation, LG Electronics is the most forward-looking price. Remaining boundaries — The ROI of capex has not been proven, and the speed of ASIC encroachment in inference determines the long-term multiples of NVIDIA and Broadcom. The bullish side looks at the 'scale of money,' while the cautious side looks at the 'return on money.' Both are right, and that's why discipline is needed.

Chapter 9

Valuation — Bull/Super-Bull profit-taking prices and allocation

The target prices below are not fiction but three paths drawn based on multiples agreed upon in three discussions, brokerage consensus, and country-specific investment facts. Core business and performance set the base, and the concretization of sovereign AI and mega-deals open up Bull and Super-Bull scenarios. The criteria for evaluation were tailored to the nature of the company — the three US companies with clear business performance were evaluated based on consensus and growth multiples, while the two LG companies, where future options (businesses with small or no revenue yet) are the core of value, were evaluated using sum-of-the-parts (SOTP: core business + sum of individual option business values). This is because if you look at companies with live options only through a single earnings multiple, future value disappears to zero.

NVIDIA (NVDA) — Currently approx. $205

Scenario	Profit-taking Price	Rationale
Base	$290 +41%	Justifies Fwd P/E 20.7x, consensus target $299. FY27 revenue consensus $390 billion.
Bull	$340 +66%	Sovereign revenue ($30B) additional acceleration + Blackwell/Verarubin sell-out continues.
Super-Bull	$380 +85%	Quarterly realization of HUMAIN, UAE, EU mega-deals, upward revision of FY27 data center revenue.

Safe fractional buying $185~205 (200-day line), stop-loss $178~188. No. 1 in valuation and visibility among the five stocks.

Broadcom (AVGO) — Currently approx. $382

Scenario	Profit-taking Price	Rationale
Base	$500 +31%	Consensus target $522, FY26 AI revenue $56 billion, backlog $73 billion.
Bull	$580 +52%	Inference transition, ASIC share expansion, FY27 AI $100 billion visibility strengthened.
Super-Bull	$680 +78%	Full-scale expansion of outsourcing for national/Big Tech custom inference chips, additional new clients.

Buy on pullback — fractional buying $349~382 (check 200-day line), stop-loss $333~348. Note TTM 91x high valuation and customer concentration.

Marvell (MRVL) — Currently approx. $279.70

Scenario	Profit-taking Price	Rationale
Base	$300 +7%	Consensus upward trend (Stifel $321, Barclays $275), data center revenue weight 76%.
Bull	$345 +23%	Custom silicon FY29 $10 billion visibility + NVLink Fusion/optical connectivity full-scale (highest target price).
Super-Bull	$420 +50%	Achieving 20% share in the $94 billion 2028 market + expansion of sovereign inference chip outsourcing.

High valuation compared to Broadcom (TTM 95.8x), Amazon customer concentration. A complementary bet to the leader, high volatility — fractionalize on adjustments.

LG CNS (064400) — Currently approx. 92,800 KRW

Scenario	Profit-taking Price	Rationale
Base	120,000 KRW +29%	Core business (SI + AI/Cloud 60%) 2027E profit × forward P/E 18x ≈ 12 trillion KRW business value.
Bull	175,000 KRW +89%	Sum of core business + Data Center DBO (Samsung 780 billion KRW confirmed) + Physical AI (AgenticWorks) + CBDC options (SOTP).
Super Bull	210,000 KRW +126%	Furiosa NPU full-stack + CBDC payment network SaaS/STO option full value. Option value increased due to strengthening sovereign AI.

Due to the sharp stock price drop, the upside potential to the SOTP target price has actually expanded (the target price is the sum of business values and is unrelated to the stock price drop). However, the Super Bull target is based on full option valuation (assuming simultaneous realization), which carries high volatility. Split buy: 88,000–93,000; stop-loss: 78,000.

LG Electronics (066570) — Currently approx. 225,500 KRW

Scenario	Level	Rationale
Base	250,000 KRW +11%	Core business (Appliances/HVAC/Automotive) SOTP forward PER 13–14x + LG Innotek stake (40.8%) value. Core business alone exceeds current price.
Bull	380,000 KRW +69%	Core business + DC cooling (2027 revenue 1 trillion KRW) + Humanoid actuators (AXIUM) + Nvidia alliance options. Consistent with foreign brokerage targets in the 400k range.
Super Bull	520,000 KRW +131%	Full value of 6 options + Nvidia M.A.P. formalization + global actuator market share visibility (SOTP).

The appliance multiple (consensus 166,750) is a conservative figure that ignores future options — SOTP shows the core business alone is worth more than the current price. Note the 27.5% slowdown in core operating profit and be mindful of volatility. Split buy on dips (200k range); stop-loss: 200,000.

New capital allocation (agreed). Nvidia 35 · Broadcom 22 · Marvell 13 · LG CNS 20 · LG Electronics 10 (US 70 : Korea 30). Focus on the core (US 3 GPU/ASIC firms), add Korean sovereign direct beneficiaries (LG CNS) through split buying, and minimize weight in LG Electronics, where revaluation has already run ahead. Entry order: Nvidia (confirm 200-day moving average support) → LG CNS (split buy in low 90k range) → Broadcom/Marvell (wait for dips) → LG Electronics (dips only).

The Tenth Chapter

Signals of a Broken Narrative

Every narrative can be broken. If the path above was the plan, the downside is the signal that the path is blocked. When this signal turns on, the story must be rewritten.

Signal	Meaning	Action
Poor Capex ROI	Delayed generative AI monetization / Big Tech capex slowdown → Multiple compression starting from the core.	Split take-profit on Nvidia/Broadcom. National capex (security budgets) is relatively insensitive — defend with Korean sovereign plays.
Accelerated ASIC encroachment	If Nvidia’s inference market share (90% → 20-30% by 2028) drops faster than expected, multiple de-rating will occur.	Hedge by reducing Nvidia weight and increasing weight in the 2 ASIC firms (Broadcom/Marvell) (caution: do not over-bet by buying all three).
AI Debt/Credit Risk	Rollover of surging Big Tech corporate bonds / simultaneous volatility in chips and credit if interest rates shock.	Sovereign (national balance sheet) and Nvidia (creditor) are relatively safe. Be wary of firms heavily reliant on corporate bonds.
LG Dual-Company Volatility	LG CNS theme overheating / foreign capital exit, LG Electronics price reversal (exceeding target price by +35%).	Stop-loss at 78,000 for LG CNS and 205,000 for LG Electronics. Re-evaluate at every tournament/earnings event.

Conclusion. The story that began where Claude 5 left off concludes as follows: As intelligence gains borders, every country seeks to house it within its own territory, and trillions of dollars of desire flow into chips and data centers. At the peak of that path stands the most honest king (Nvidia), the arms dealers (Broadcom/Marvell) stand at the crossroads of the inference era, and within Korean territory, there are the full-stack implementer (LG CNS) and the pickaxe of cooling (LG Electronics). The final question of an era where narrative becomes value is always the same: is the story backed by earnings? Sovereign AI finds that answer in a funding source that rarely runs dry: 'national budget.' However, until the returns at the end of that path are proven, one must view the upside with a long-term perspective and protect the downside with discipline.

This report is a narrative analysis summarizing the 3rd round of discussions by the Goguma Team (Onion, Acorn, Peanut, Carrot, Sweet Potato) and in-depth research on 8 thematic businesses and national investments.
The target prices provided are scenarios based on brokerage consensus and the latest industry trends, not investment advice. Investment decisions and responsibilities rest solely with the investor.
As of June 14, 2026 · Nvidia (NVDA) · Broadcom (AVGO) · Marvell (MRVL) · LG CNS (064400) · LG Electronics (066570)