Saltlux (304100): Korea's Sovereign AI Pioneer — Gangbangcheon C × Geochajesi 8/20, No Buy Until Profitable
25-year NLP heritage + Luxia LLM 3.0 + GOOVER 1.5M users. Has all the ingredients of a sovereign AI pioneer, but 5 consecutive loss years and bearish downtrend are the blockers. Gangbangcheon C × Geochajesi 8/20 — no entry until RSI divergence or profitability is confirmed.
Core Position
Korea's sovereign AI pioneer with 25-year NLP heritage and Luxia LLM — no investment until profitability is proven
Investment Thesis
Saltlux has built 25 years of Korean NLP expertise since its founding in 2000. The 2023 launch of its proprietary large language model Luxia (LUXIA) marked a clear pivot from AI solutions to LLM foundation model company. Global Hugging Face Open LLM Leaderboard #1 (March 2024) and the claimed 1/20 inference cost advantage of Luxia 3.0 vs GPT-4o provide tangible evidence of domestic technological competitiveness. However, from an investment perspective, 5 consecutive years of operating losses (-₩8B FY2025), accumulated deficit of ₩40.1B, and a sustained bearish downtrend (-65% from 52-week high) are the present reality. Gangbangcheon C × Geochajesi 8/20 — currently in a no-buy zone. Reassess when one of two conditions is confirmed: (1) RSI divergence + long bullish candle (technical reversal signal), or (2) AI national flagship tech designation + two consecutive quarters of operating profit (fundamental reversal signal).
① Non-Financial — 25-Year NLP Heritage and Sovereign AI Positioning
The moat's core is 25 years of Korean NLP expertise and the proprietary Luxia foundation model. ~2,000 public sector references, three World-Class Product certifications (Luxia, Luxia-ON, GOOVER), and the Hugging Face #1 ranking form a brand moat in domestic public and enterprise markets. Luxia-ON (on-premises inference) directly targets sovereign AI demand from government, financial, and healthcare clients that avoid cloud for security reasons. CEO Lee Gyeong-il has led the company for 25+ years post-founding with a 10.45% direct stake — positive long-term alignment. Key risks: founding shareholder at only 12.65% stake (weak governance defense), and 5-year accumulated loss structure creates persistent dilutive equity raise risk. → Full moat ratings, competitive landscape, and risk analysis in the Non-Financial tab.
② Validator — Gangbangcheon C × Geochajesi 8/20 = No Buy
Only 1 of 5 Gangbangcheon steps passed (AI agent TAM CAGR 47.3%, sovereign AI trend tailwind). Steps 2–5 failed: market position grade C (volatile revenue, B2G dependency), business model ⚠️ (SaaS transition incomplete), financial quality ❌ (5 consecutive loss years), K-PER ❌ (operating income negative). PSR basis: ~₩25.7B market cap / ₩41.6B revenue = 0.62x PSR. Conservative scenario (revenue 400B, PSR 0.5x) implies -22% downside. Geochajesi 8/20 — Volume 2 (institutional net selling, order intensity unavailable) + Chart 1 (bearish MA alignment, -65% from 52-week high) + Catalyst 3 (1Q26 +55% YoY rebound, Luxia 3.0) + Market 2 (AI sector tailwind, but weak direct Saltlux theme). No veto triggered. → Full 5-step breakdown, PSR scenarios, and Geochajesi item scores in the Validator tab.
③ Technical — Bearish MA Alignment, RSI 38 Approaching Oversold
Full bearish MA alignment across all timeframes (5/20/60/120/200-day). Current price (₩20,450) is -65% from 52-week high (₩58,900). Falling wedge converging since Nov 28, 2025 high (₩28,500) — the breakout direction determines the next major move. RSI(14) ~38 approaching oversold (30 line); Bollinger Band at lower band — volatility expansion imminent. MACD below 0 with histogram negative bars weakening (declining momentum fading). Primary support ₩21,500 (Jan 17, 2026 support), critical support ₩18,680 (52-week low = absolute support). First resistance ₩28,500 (short-term high), second resistance ₩35,000 (Fibonacci 61.8%). Conservative strategy (enter after RSI divergence + support confirmation) recommended. Upside targets: 1st ₩22,450, 2nd ₩24,750. → Full 3 scenarios, support/resistance, RSI chart, and bull/bear signals in the Technical tab.
Key Metrics
Current Price (May 14)
20,450원
52주 고점 대비 -65%
Market Cap
~257억원
PSR 0.62배
FY2025 Revenue
416억원
-9.3% YoY
1Q26 Revenue
89억원
+55.1% YoY
Accumulated Deficit
401억원
5년 연속 영업적자
Gangbangcheon×Geochajesi
C × 8/20
매수 금지
Bull Case
- Luxia 3.0 — claimed inference cost 1/20 vs GPT-4o; directly positioned for sovereign AI (on-premises) tailwind
- 1Q26 revenue +55.1% YoY (₩8.9B) — first concrete rebound signal after a 5-year declining trend
- GOOVER reached 1.5M users in 4 months — first visible path toward B2C monetization
- AI national flagship designation would bring preferential public project allocation + state GPU/data support
- NHN strategic partnership (₩15.6B investment, 3rd shareholder) — potential Dynamics acquisition synergies and cloud integration
Bear Case
- 5 consecutive years of operating losses (2021–2025) — accumulated deficit ₩40.1B, no clear BEP timeline
- Bearish MA alignment — -65% from 52-week high, sustained downtrend channel for 6+ months
- Government revenue dependency (B2G 55%) — structural revenue contraction risk from administration changes or budget cuts
- Global LLM competition intensifying — Korean language advances by OpenAI and DeepSeek are rapidly narrowing Luxia's differentiation window
- Founding shareholder stake only 12.65% — weak governance defense, repeated equity raises possible, shareholder dilution risk
Technical Summary
Full bearish MA alignment (5>20>60>120>200-day inverted). Current price ₩20,450 is -65% from 52-week high (₩58,900). Falling wedge converging. RSI(14) ~38 approaching oversold territory; MACD histogram negative bars weakening signals fading downside momentum.
Saltlux 304100 Price & RSI Chart (Nov 2025–May 2026)
Support
21,500원 · 18,680원
Resistance
28,500원 · 35,000원
Trend Analysis
Short-term (20-day): Bearish. Current price ₩20,450 is below 20-day MA (~₩24,100). Medium-term (60-day): Bearish. Price is -22% below 60-day MA (~₩26,300). MA order: Full bearish alignment (price < 20d < 60d < 120d < 200d). Death cross 30+ trading days elapsed — downtrend entrenched. 6+ month downtrend channel since Nov 28, 2025 high. Currently converging inside a falling wedge.
Momentum & Indicators
RSI(14) ~38 — approaching oversold (30). Previous low (Jan 17, 2026, RSI 32) showed divergence followed by a short-term bounce. Currently re-testing the second low. Bollinger Band squeeze at lower band — volatility expansion imminent. MACD: below 0 line, histogram negative bars gradually narrowing (declining momentum easing). Potential bullish divergence formation zone.
Key Technical Points
₩21,500 — Jan 17, 2026 trough support. Key reference for short-term bounce. Break below opens path to 52-week low (₩18,680).
₩18,680 — 52-week low. Break below accelerates technical decline. Holding this level is a prerequisite for bottoming.
₩28,500 — Nov 28, 2025 short-term high. Near Fibonacci 38.2% retracement (high 58,900 → low 18,680). Breakout here signals medium-term trend reversal.
₩35,000 — Fibonacci 61.8% retracement. Aligns with K-PER base-scenario target.
Since Nov 2025, both highs and lows declining in a wedge. Typically: downside break = sharp decline; upside break = strong reversal. Currently near the wedge bottom.
Basis: 52-week high ₩58,900 → current low ₩18,680. 38.2%=₩22,450, 50%=₩23,590, 61.8%=₩24,750. Current price (₩20,450) is below the 23.6% retracement (₩28,180) — below all Fibonacci levels.
Trading Scenarios
Entry
Enter 1/3 position after RSI divergence confirmation + bounce candle at ₩21,000–21,500 support
Stop
₩18,200 (-11% from avg ₩20,800)
Target
1st ₩22,450 (Fib 38.2%), 2nd ₩24,750 (Fib 61.8%)
52-week low (₩18,680) break invalidates entry. R:R 2.0 based on 2nd target. Full position not recommended while MA alignment remains bearish.
Entry
1/3 at ₩21,000 / 1/3 at ₩19,000 / 1/3 after ₩22,500 + 60-day MA breakout
Stop
₩18,200 (-12% from avg entry ~₩20,800)
Target
₩24,750 (R:R 1.5 from avg entry)
3rd tranche (₩22,500) requires 60-day MA breakout + volume confirmation. Without confirmation, maintain only 1st and 2nd tranches.
Entry
Immediate entry after AI national flagship designation announcement or 2Q26 profit turnaround confirmation
Stop
-5–7% from the announcement-day intraday low
Target
₩28,500 (first resistance) and above → R:R 2.0+
Pure event trade. No entry without a confirmed announcement. Factor in potential short squeeze if short interest is elevated.
Bullish Signals
RSI(14) ~38 — re-entering oversold zone after Jan 17, 2026 trough (RSI 32); potential divergence formation zone
Falling wedge converging — typically produces strong upward reversal on breakout
MACD histogram negative bars gradually narrowing — fading downside momentum signal
1Q26 revenue +55.1% YoY — first meaningful fundamental rebound
AI national flagship tech candidate — potential public project allocation catalyst
Bearish Risks
Full bearish MA alignment — structurally bearish; short-term bounces are selling opportunities
-65% from 52-week high (₩58,900) — structural time required before institutional/foreign confidence recovers
Death cross 30+ days elapsed — downtrend entrenched, momentum freshness exhausted
Repeated touches of Bollinger lower band — persistent downside pressure signal
If ₩18,680 (52-week low) breaks, no technical support below — potential sharp decline
Editor Note
Saltlux is currently converging in a falling wedge within a structurally bearish downtrend. RSI 38 is approaching oversold, but entering on RSI alone — without a trend reversal — is the classic 'early contrarian' trap. The prerequisites for a technical reversal are: ① RSI divergence, ② long bullish candle, ③ volume confirmation. No entry without all three. If a fundamental reversal is confirmed (AI national designation or two consecutive profitable quarters), event-driven entry can be considered even without a technical signal first.
* Technical analysis is based on historical data and does not guarantee future returns. Final investment decisions are your own responsibility.
Saltlux 3-Year Financials (Revenue, Op. Income, ROA, ROE)
Switching Cost & Moat
Moat Strength by Type
Technology / Patents
25-year cumulative NLP technology, Luxia 1.0–3.0 in-house development. Hugging Face Open LLM #1 (March 2024). However, rapidly narrowing technology gap due to global open-source LLM surge
Brand
25-year references in domestic public/AI market. Three World-Class Product certifications. Low general consumer awareness
Switching Costs
Switching costs exist once public/enterprise clients build customized AI solutions. However, project-based contracts limit recurring revenue transition, reducing switching cost effectiveness
Network Effects
GOOVER B2C is accumulating early users (1.5M) but network effect creation remains at a very early stage
The moat's core is 25 years of Korean NLP expertise and the proprietary Luxia foundation model. ~2,000 public sector references and three World-Class Product certifications form high entry barriers in B2G/B2B markets. Luxia-ON (on-premises inference) differentiates from cloud LLMs in security-sensitive government, financial, and healthcare verticals. However, global open-source LLMs (DeepSeek, etc.) improving Korean language performance are rapidly pressuring the technology moat, and B2C network effects remain at an early stage. Core vulnerability: incomplete monetization, public dependency structure, 5 years of accumulated losses.
Management & Governance
CEO Lee Gyeong-il (born 1972) has been sole CEO since founding the company (then Softmeta) in 2000 — a 25+ year founder-CEO tenure. He graduated from Hankuk University of Foreign Studies with an English degree. His public sector network is strong through policy roles (Youth Industry 4.0 Committee member, Korea AI Industry Association vice chairman, AI Industry Convergence Division chairman, MSIT adjunct professor). Direct stake of 10.45% (~1.31M shares) provides positive long-term alignment. However, total affiliated shareholder stake is only 12.65%, leaving governance defenses structurally weak. At the 2026 AGM, management shortened executive terms (6 months → 4 years) and raised director removal thresholds — interpreted as attempts to strengthen governance protection.
Competitive Landscape
네이버 (HyperCLOVA X)
#1 domestic search platform + large-scale LLM. Direct competition in Saltlux's public B2B market. Overwhelming infrastructure, data, and capital advantages
LG AI연구원 (EXAONE)
LG Group AI research lab. Strong enterprise B2B. Same public/enterprise client targets as Saltlux. Massive GPU and infrastructure investment underway
OpenAI / DeepSeek
Rapid Korean language performance improvement by global LLMs. GPT-4o and DeepSeek cost reductions could dilute Luxia cost advantage. Indirect competition but largest structural threat
업스테이지 (Solar)
Domestic independent LLM startup. Focused on enterprise B2B and overseas markets. Similar positioning to Saltlux but different VC funding scale
Domestic LLM competitive landscape: Naver (HyperCLOVA X), LG AI Research (EXAONE), Kakao (Kanana), Samsung (Samsung Gauss) vs. Saltlux (Luxia). Saltlux holds a specialized B2G/on-premises position, but faces absolute disadvantages vs. large conglomerates in capital, talent, and compute resources. Direct competitors: Upstage (Solar), legal-tech (government AI), and Wrtn Technologies (consumer AI). Indirect competition: OpenAI (GPT-4o), Meta (LLaMA), Mistral, DeepSeek (already high Korean performance). Hugging Face Open LLM Leaderboard #1 (March 2024) is meaningful recognition, but sustaining that ranking is the key question. Rapid global open-source LLM improvement is the biggest structural threat.
ESG & Summary
As a software company without manufacturing facilities, direct carbon emissions are minimal, but GPU computing energy usage is a growing indirect environmental burden. Employee count is in the hundreds — no standalone ESG report published. Governance concerns: low founding shareholder stake (12.65%), limited disclosure on outside director independence, unclear performance linkage for executive compensation. Silicon Valley subsidiary (planned ₩40B investment) and Vietnam operations represent early-stage globalization. No shareholder return history (no dividends). AI ethics and bias management policy disclosure is insufficient.
Key Risks
Monetization Delay Risk
No clear BEP timeline. Sustaining operations without additional capital raises (equity issuance) is difficult given 5-year accumulated losses; any equity raise will dilute shareholders. Planned ₩40B Silicon Valley investment adds further capital pressure.
Global LLM Erosion Risk
Korean language performance of global LLMs (OpenAI, Meta, Mistral, DeepSeek) is improving rapidly. Even if cost advantage (claimed 1/20 vs GPT-4o) holds, if 'using global LLMs directly is cheaper' perception spreads, the differentiation rationale weakens.
Government Revenue Dependency Risk
~55% of revenue from government (B2G). Project-based revenue still dominant — transition to recurring revenue incomplete. Government administration changes, AI policy shifts, or budget cuts can structurally compress revenue.
Governance / Control Risk
12.65% founding shareholder stake is vulnerable to hostile M&A or management change pressure. With 71% retail shareholder concentration, shareholder meeting resolutions may not align with management. Key Man risk from 25-year founder-only leadership.
R&D Investment Decline Risk
1Q2025 R&D ratio fell sharply to 21.6% (from 35.5% YoY). If R&D is being sacrificed under monetization pressure, sustaining Luxia's technical competitiveness becomes increasingly difficult.
Valuation and Price Risk
Current PSR of 0.62x may appear cheap, but PSR for a persistently loss-making company can compress further depending on loss magnitude and equity erosion pace. Break below 52-week low (₩18,680) removes technical support and opens sharp decline risk.
Gangbangcheon 1/5 passed
1 of 5 Gangbangcheon steps passed. Only industry/infrastructure (high-growth AI agent TAM, sovereign AI trend) cleared. Market position (grade C), business model (incomplete), financial quality (5 consecutive loss years), and K-PER (negative operating income) all failed. PSR-based conservative scenario upside –22%. Geochajesi 8/20 — no buy. Investment prohibited until both profitability and MA alignment reversal are simultaneously confirmed.
Saltlux 3-Year Financials (Revenue, Op. Income, ROA, ROE)
Gangbangcheon 5-Step Checklist
Step 1
Industry & Infrastructure — AI Agent TAM CAGR 47.3%, Sovereign AI Trend
AI agent market CAGR 47.3% (to $61.8B by 2030). Sovereign AI (domestic model, on-premises) policy tailwind. Korean government initiative to designate 5 AI national flagship tech companies — Saltlux among candidate pool. However, rapid global LLM commoditization is quickly eroding the core rationale for sovereign AI.
Step 2
Market Position Grade C — Hugging Face #1, but Volatile Revenue
Hugging Face Open LLM Leaderboard #1 (March 2024) and ~2,000 public sector references are positive. However, standalone revenue declined from ₩23.3B (2022) to ₩17.1B (2025), and B2G dependency (55%) limits revenue expansion headroom. GOOVER B2C with 1.5M MAU is early-stage with negligible revenue contribution.
Step 3
Business Model ⚠️ — SaaS Transition Incomplete, Project-Based Revenue Still Dominant
Efforts to transition to recurring revenue (AWS Marketplace listing, overseas expansion) are underway. However, revenue remains predominantly public SI project-based with low SaaS/subscription contribution. Risk of repeated equity raises (failing non-dilution requirement) persists.
Step 4
Financial Quality ❌ — 5 Consecutive Years Operating Loss, Negative FCF, ROA –11%
5 consecutive years of operating losses 2021–2025. FY2025 operating loss –₩8.0B, net loss –₩11.9B. FCF negative for 3 consecutive years. PQC worst combination (P↓ Q↓ C↑). ROA –11%, ROE –28% worsening. Accumulated deficit ₩40.1B accelerating equity erosion.
Step 5
K-PER (PSR Proxy) ❌ — Negative Operating Income, PSR 0.62x
K-PER not calculable due to negative operating income. Using PSR (Price-to-Sales) proxy: ~₩25.7B market cap ÷ ₩41.6B revenue = PSR 0.62x. Conservative scenario (revenue –1% CAGR to ₩40B, PSR 0.5x) → target cap ₩20B, upside –22%. Reassessment needed after profitability is achieved.
K-PER Scenario Analysis (3-Year Target)
K-PER not applicable due to negative operating income — using PSR (Price-to-Sales) proxy. Current market cap ~₩25.7B (at ₩20,450) ÷ FY2025 revenue ₩41.6B = PSR 0.62x. 3-year forward revenue scenarios × PSR multiples applied. PSR can be re-rated once profitability is demonstrated — proving profitability is the prerequisite for any valuation re-rating.
| Scenario | Annual Growth | Non-GAAP Profit | Applied PER | Target Cap | Upside |
|---|---|---|---|---|---|
| Optimistic | +8% | 매출 520억원 | 1.2x | ~624억원 | +143% |
| Base | +3% | 매출 460억원 | 0.8x | ~368억원 | +43% |
| Conservative | –1% | 매출 400억원 | 0.5x | ~200억원 | -22% |
Geochajesi Score (8/20)
Volume at average levels but sustained institutional net selling. Large single-day net sell event (Mar 13, 2026) followed by volume decline (one-off signal). Order flow intensity data unavailable. Retail buying vs. institutional selling structure persisting.
Full bearish MA alignment. –65% from 52-week high ₩58,900. Repeated death crosses, 30+ trading days elapsed. RSI ~38 approaching oversold (sole positive element). Repeated Bollinger lower band touches. Investing.com daily "Strong Sell".
1Q26 revenue +55.1% YoY (strongest positive catalyst). Luxia 3.0 launch (pivot to AI agents). AI national flagship tech candidate. However, catalyst consumed immediately after earnings announcement. Potential catalyst vacuum until 2Q26 earnings release.
Broad AI sector tailwind is real. KOSDAQ AI retail investor interest persisting. However, weak direct Saltlux theme participation; no 6-month outperformance vs. KOSDAQ index. Small/mid-cap AI lagging behind AI mega-cap strength.
Entry Strategy (3 Tranches)
No entry on RSI bounce alone within a bearish downtrend. Buying without support confirmation is an "early contrarian" trap. Existing holders should re-assess risk.
Enter 1/3 ONLY when all three conditions are simultaneously met: RSI divergence + long bullish candle + volume confirmation. Stop: ₩18,200.
Additional 1/3 on bounce after 52-week low (₩18,680) support is re-confirmed. Break below this zone = immediate full exit.
Final 1/3 added after 60-day MA breakout with volume confirmation. First signal of bearish MA alignment unwinding.
Exit Triggers
52-week low ₩18,680 break → immediate full exit
2Q26 operating loss worse than –₩8.0B → full exit
RSI repeatedly failing to break 60 while MA alignment remains bearish → reduce position 50%
Official exclusion from AI national flagship tech designation → exit event-driven holdings
₩28,500 (first resistance) breakout + MA alignment turns bullish → raise target to ₩35,000, apply trailing stop
Portfolio Weight Recommendation
New entry at current price: 0% (no buy). Enter maximum 1/3 of target weight only after all three conditions are confirmed (RSI divergence + long bullish candle + volume). Short-term swing: small position after technical reversal signal only. Long-term: full position not permitted until BEP is achieved AND MA alignment turns bullish simultaneously.
Editor Note
Saltlux is Korea's oldest sovereign AI pioneer. The 25-year NLP heritage and Luxia LLM are real assets, and the 1Q26 +55.1% revenue rebound and 1.5M GOOVER users are the first signals that something is stirring at the bottom. However, 5 consecutive loss years and a bearish downtrend mean the investment prerequisites — BEP achievement and chart reversal — are not yet met. Buying at current prices risks averaging down into a downtrend rather than true contrarian investing. Wait for RSI divergence confirmation or an AI national flagship designation announcement before considering even a small position.
Financial Data
Saltlux fiscal year: Calendar = fiscal year (Jan 1–Dec 31). FY2025 = Jan–Dec 2025. Currently in FY2026 Q2. 1Q2026 earnings reported.
| Period | Revenue | Growth | Op. Income | Op. Margin |
|---|---|---|---|---|
| FY2023Net loss approx. –₩6.6B. ROA –7.5%, ROE –18% | 308억원 | +18.3% | –66억원 | –21.4% |
| FY2024Revenue surge due to Dynamics acquisition (Nov 2023) M&A effect; organic growth was low. ROA –9.0%, ROE –22% | 459억원 | +49.0% | –66억원 | –14.4% |
| FY2025Net loss –₩11.9B. Accumulated deficit ₩40.1B. ROA –11.0%, ROE –28% | 416억원 | –9.3% | –80억원 | –19.2% |
| 2026 1QLoss narrowed from –₩5.9B YoY. Early rebound signal but still loss-making structure | 89억원 | +55.1% | –49억원 | –55.1% |
GAAP vs Non-GAAP Note
FY2024 revenue surge reflects Dynamics acquisition (Nov 2023) consolidation. Organic standalone revenue declined from ₩23.3B (2022) to ₩17.1B (2025). Silicon Valley subsidiary (planned ₩40B investment) and Vietnam operations have no meaningful revenue contribution yet. Next earnings: 2Q2026 (expected August 2026).
Key Valuation Metrics
PSR (Market Cap / Revenue)
0.62배
Market cap ~₩25.7B / Revenue ₩41.6B (as of May 14, 2026)
ROA (FY2025)
–11.0%
5 consecutive years negative, worsening trend
ROE (FY2025)
–28.0%
Equity rapidly declining due to ₩40.1B accumulated deficit
Accumulated Deficit
401억원
Total cumulative losses since listing
* GAAP basis. All figures are estimates based on public information and are not investment advice.
Same Exchange
- 194480Devsisters (194480): Gangbangcheon C × Geochajesi 6/20 Do Not Enter — Cookie Run IP 3B Users · ₩1.5T Lifetime Revenue · 50M TCG Cards · FY25 Op. Income ₩6.4B · FY26Q1 ₩-17.4B Loss Return · K-PER Optimistic -17% / Base -53% / Conservative -77% — Do Not Enter Until: Crumble (3Q26) + Quarterly Profit + Geochajesi ≥ 14
- 083450GST Global Standard Technology (083450): Gangbangcheon B × Geochajesi 16/20 Split-Buy — Korea's Only Direct Bigtech (TSMC/Micron/YMTC) Scrubber & Chiller Supplier · S-OIL AI Datacenter Immersion Cooling Demonstration Agreement · K-PER Conservative -0.9% / Base +13% / Optimistic +29% — 3-Tranche Avg ₩58,700 · Stop ₩50,672 (Fib 61.8%) · Target ₩70,800 · R:R 1.51