AI Infrastructure 12-Company Comparative Analysis Series
Part 1: Peer Groups & Multiples | Part 2: Valuation Deep-Dive (Current) | Part 3: Sector Trends & Investment
4. Multiple Premium/Discount Analysis
Premium/Discount vs. Peer Median
| Company | PER | vs. Peer | Assessment | Key Driver |
|---|---|---|---|---|
| AMD | 77.8x | +87.8% | Overvalued | MI400 expectations, ROE 7.1% |
| NVIDIA | 45.2x | +15.9% | Fair+ | GPU 90%, ROE 107% |
| Hanmi Semiconductor | 52.5x | +56.4% | Overvalued | TC bonder monopoly risk |
| SK hynix | 6.4x | -21.5% | Undervalued | HBM4 #1 |
| Meta | 27.2x | -22.4% | Undervalued | OPM 41% |
| Microsoft | 24.9x | -9.4% | Slight discount | Azure AI, FCF $71.6B |
4.1 AI Semiconductor Group
| Company | PER | vs. Peer Median | Premium/Discount Drivers |
|---|---|---|---|
| AMD | 77.0x | +87.8% (vs. 41.0x) | MI400 expectations, data center revenue acceleration. However, ROE of 7.1% signals weak profitability |
| NVIDIA | 47.5x | +15.9% (vs. 41.0x) | 90% GPU market share, ROE 107.4%, revenue +114.2% |
| TSMC | 34.4x | -16.1% (vs. 41.0x) | Dominant foundry position, stable margins. Limited cyclical concerns |
| SK hynix | 32.2x | -21.5% (vs. 41.0x) | #1 HBM market share, FY2025 revenue KRW 97.1T. Korea discount potential |
Analysis:
AI infrastructure valuation analysis Part 2: Deep-dive — AMD’s PER of 77.0x trades at an 87.8% premium to the semiconductor peer median of 41.0x. This reflects market expectations for next-generation AI accelerators such as MI400, but given ROE of 7.1% and an operating margin of 10.7%, the current premium appears excessive. The fact that AMD’s PER exceeds NVIDIA’s despite a stark profitability gap (operating margins: NVIDIA 55.9% vs. AMD 10.7%) suggests that optimistic growth expectations are already substantially priced into AMD shares.
Conversely, SK hynix (PER 32.2x) trades at a 21.5% discount to the semiconductor peer median. Despite strong fundamentals — 60% of NVIDIA’s HBM4 supply, FY2025 revenue of KRW 97.1 trillion (+46.7% YoY), and operating profit of KRW 47.2 trillion — the Korea market discount and memory cycle concerns continue to cap valuation (SK hynix Newsroom, Merl Blog).
TSMC (PER 34.4x) trades at approximately 16% below the median, but this appears to be a reasonable level reflecting the stable, predictable cash flow characteristics of the foundry business. Given its solid fundamentals (45.7% operating margin, 35.2% ROE) and strategic position as the critical bottleneck in AI chip production, this is closer to fair value than a genuine discount.
Premium/Discount Key Figures
+152%
NVIDIA PER Premium
-22%
Meta PER Discount
+63%
Oracle EV/EBITDA Premium
4.2 Big Tech/Cloud Group
| Company | PER | vs. Peer Median | EV/EBITDA | vs. Peer Median | Premium/Discount Drivers |
|---|---|---|---|---|---|
| Oracle | 28.8x | +4.0% (vs. 27.7x) | 25.7x | +61.6% (vs. 15.9x) | Cloud transition bet, CapEx 37%/revenue |
| Alphabet | 28.3x | +2.2% (vs. 27.7x) | Data unavailable | — | AI search transition, Cloud growth acceleration |
| Amazon | 27.7x | 0.0% (vs. 27.7x) | 15.0x | -5.7% (vs. 15.9x) | AWS profitability, retail margin improvement |
| Microsoft | 25.1x | -9.4% (vs. 27.7x) | 17.0x | +6.9% (vs. 15.9x) | Azure AI + stable enterprise SW revenue |
| Meta | 21.5x | -22.4% (vs. 27.7x) | 15.8x | -0.6% (vs. 15.9x) | High margins, but metaverse risk perception |
Analysis:
Meta’s position as the lowest-multiple stock (PER 21.5x) within the Big Tech group is noteworthy. Despite top-tier fundamentals within the peer group — 41.4% operating margin, 30.2% ROE, and 22.2% revenue growth — the market’s skeptical view of metaverse/Reality Labs investment acts as a discount factor (Meta IR, Feb 2026). However, with FY2025 FCF of $46.1B, Meta has ample financial capacity even with projected AI CapEx of $115–135B.
Oracle is near the peer median on a PER basis, but its EV/EBITDA of 25.7x stands out at a 62% premium to the median of 15.9x. This is driven by aggressive investment at a CapEx/revenue ratio of 37%, which elevates Enterprise Value relative to EBITDA, while the leverage structure (debt $104.1B, cash $10.8B) pushes EV to $555.0B (Oracle IR). Successful cloud transition could normalize multiples, but the negative FCF trend is a near-term risk factor.
4.3 AI Infrastructure Value Chain Group
| Company | PER | vs. Peer Median | Key Premium Factor | Key Risk Factor |
|---|---|---|---|---|
| Hanmi Semiconductor | 115.7x | +56.4% (vs. 74.0x) | TC bonder monopoly, revenue +251% | Hanwha Semitec entry threatens monopoly |
| Doosan Enerbility | 74.0x | 0.0% (vs. 74.0x) | SMR/large nuclear, 2026E orders KRW 14T | 4.2% operating margin, uncertain realization timeline |
| Western Digital | 43.9x | -40.7% (vs. 74.0x) | AI data center HDDs sold out | Storage cycle, SSD competition |
Analysis:
Hanmi Semiconductor (PER 115.7x) trades at an extreme premium even within the value chain group. FY2024 revenue growth of +251% and operating profit growth of +638% provide the basis, but SK hynix’s shift to dual-sourcing with Hanwha Semitec poses a risk to the monopoly premium (Merl Blog, April 2025). While Hanmi has responded with a 25% price increase and after-sales service personnel withdrawal, it remains uncertain whether the TC bonder monopoly structure can be sustained long-term.
Doosan Enerbility (PER 74.0x, 12M Forward basis) has not yet escaped its low-profitability structure at 4.2% operating margin, but the turnaround narrative rests on FY2026E operating profit of KRW 583.9B (+84% YoY) and order projections of up to KRW 14 trillion in 2026. U.S. large nuclear (AP1000) supply expansion and dedicated SMR production facility investment are the growth catalysts, but time-to-realization remains the critical valuation variable (DartPoint AI, FnGuide).
5. Implied Valuation
This section applies peer group median multiples to each company’s financial metrics to calculate implied valuations. However, since EBITDA and Revenue multiple data are unavailable for several companies, PER-based analysis takes precedence.
5.1 PER-Based Implied Valuation
AI Semiconductor Group (Peer Median PER: 41.0x)
| Company | Current PER | Net Income (Latest FY) | Current Market Cap | Implied Market Cap (Peer Median PER) | Gap |
|---|---|---|---|---|---|
| NVIDIA | 47.5x | $72.9B | $4.59T | ~$2.99T | -34.9% |
| AMD | 77.0x | $4.3B (TTM) | ~$329B | ~$176B | -46.5% |
| TSMC | 34.4x | $49.5B (TTM) | ~$1.6T | ~$2.03T | +26.9% |
| SK hynix | 32.2x | KRW 42.9T | KRW 66.2T | ~KRW 1,759T | Interpret with caution |
Caution: SK hynix’s implied market cap — calculated by mechanically applying the global semiconductor peer median PER of 41.0x to FY2025 net income of KRW 42.9 trillion — yields approximately KRW 1,759 trillion. However, this figure requires adjustment for the Korea market discount, currency fluctuations, and corporate governance differences. The gap from the current KRW 66.2 trillion market cap stems more from structural differences in market cap/net income ratios than from PER differentials alone, highlighting the limitations of simple implied valuation.
Big Tech/Cloud Group (Peer Median PER: 27.7x)
| Company | Current PER | Net Income (Latest FY) | Current Market Cap | Implied Market Cap | Gap |
|---|---|---|---|---|---|
| Amazon | 27.7x | ~$59.2B | $2.52T | ~$1.64T | -34.9% |
| Alphabet | 28.3x | $132.2B | $3.65T | ~$3.66T | +0.3% |
| Meta | 21.5x | $60.5B | $1.62T | ~$1.68T | +3.7% |
| Microsoft | 25.1x | $101.8B | ~$3.2T | ~$2.82T | -11.9% |
| Oracle | 28.8x | Data unavailable | $459.5B | Not calculable | — |
Key Implications:
- Meta’s implied market cap at the peer median PER is ~$1.68T, representing approximately 3–4% upside from the current $1.62T. This suggests the current market cap is close to peer-based fair value. While Meta has the lowest multiple in the Big Tech group on a PER basis, its high profitability (41.4% operating margin, 30.2% ROE) suggests the discount may be excessive. Multiple re-rating becomes possible once short-term FCF pressure from CapEx expansion subsides.
- Amazon’s PER exactly matches the peer median, but the gap between implied market cap (~$1.64T) and current market cap ($2.52T) indicates the market is pricing in AWS’s future growth premium. Amazon’s valuation is significantly influenced by the AWS business unit value, which PER alone fails to capture.
- Applying the peer median PER of 41.0x to NVIDIA yields ~$2.99T, suggesting approximately 35% downside from the current $4.59T, but given its differentiated fundamentals (ROE 107.4%, revenue growth 114.2%), a simple median application may be inappropriate. A growth premium is justifiably warranted for NVIDIA, and a fair PER should be established above the peer median (41.0x).
5.2 EV/EBITDA-Based Implied Valuation
With EBITDA data obtained for all U.S. companies via Alpha Vantage MCP, we can perform comprehensive EV/EBITDA-based implied valuations.
AI Semiconductor Group (Peer Median EV/EBITDA: 31.0x)
| Company | Current EV/EBITDA | EBITDA | Current EV | Implied EV (Peer Median) | Gap |
|---|---|---|---|---|---|
| NVIDIA | 37.3x | $112.7B | ~$4,208B | ~$3,494B | -17.0% |
| AMD | 45.5x | $6.7B | ~$307B | ~$209B | -32.0% |
| TSMC | 24.7x | — (TWD basis) | ~$1,889B | — | Limited by currency difference |
Big Tech/Cloud Group (Peer Median EV/EBITDA: 15.7x)
| Company | Current EV/EBITDA | EBITDA | Current EV | Implied EV (Peer Median) | Gap |
|---|---|---|---|---|---|
| Amazon | 13.1x | $145.7B | ~$1,908B | ~$2,288B | +19.9% |
| Meta | 15.3x | $101.9B | ~$1,562B | ~$1,600B | +2.4% |
| Microsoft | 15.7x | $175.3B | ~$2,752B | ~$2,752B | 0.0% |
| Oracle | 19.9x | $26.3B | ~$524B | ~$413B | -21.2% |
| Alphabet | 20.1x | $150.2B | ~$3,019B | ~$2,358B | -21.9% |
AI Infrastructure Value Chain Group (Peer Median EV/EBITDA: 22.0x)
| Company | Current EV/EBITDA | EBITDA | Notes |
|---|---|---|---|
| Hanmi Semiconductor | 30.1x | — (FnGuide) | +36.8% premium vs. peer, reflecting monopoly premium |
| Western Digital | 22.0x | $3.4B | Matches peer median, near fair value |
| Doosan Enerbility | 12.3x | — (FnGuide) | -44.1% discount vs. peer, reflecting low-profitability structure |
EV/EBITDA Key Implications:
- Amazon (EV/EBITDA 13.1x) trades at a 16.6% discount to the Big Tech peer median (15.7x), making it the most undervalued company in the Big Tech group on an EV/EBITDA basis. Applying the peer median yields an implied EV of ~$2,288B, representing approximately 20% upside from the current EV of ~$1,908B.
The Core of Implied Valuation
Implied valuation using peer median multiples quantifies over/undervaluation versus current market cap. However, Trading Comps are a snapshot of how the market values comparable companies — differentiated growth narratives require separate evaluation.
- Meta (EV/EBITDA 15.3x) is near the peer median, consistent with the PER-based analysis suggesting proximity to fair value. EBITDA of $101.9B underpins the valuation.
- Alphabet (EV/EBITDA 20.1x) and Oracle (19.9x) trade at 28% and 27% premiums to the peer median, respectively. Alphabet’s $150.2B EBITDA and Cloud growth acceleration, and Oracle’s cloud transition narrative, are the premium drivers.
6. Historical Multiple Band Analysis
6.1 Analytical Limitations
This brief does not include detailed 3–5 year multiple time-series data, preventing a precise +/-1 standard deviation band analysis. The following section uses available historical revenue/earnings trends and current multiples to qualitatively assess where each company’s multiples stand relative to history.
6.2 AI Semiconductor Group Multiple Trend Analysis
| Company | PER (Current) | Historical Earnings Trajectory | Estimated Position Within Band | Trigger Event |
|---|---|---|---|---|
| NVIDIA | 47.5x | FY2023 $27.0B -> FY2024 $60.9B -> FY2025 $130.5B (revenue) | Upper band | AI accelerator demand surge (2023~), post-ChatGPT multiple re-rating |
| AMD | 77.0x | FY2023 $22.7B -> FY2024 $25.8B -> TTM $34.6B | Above upper band | MI300/MI400 roadmap expectations, data center revenue acceleration |
| TSMC | 34.4x | FY2023 $69.3B -> FY2024 $90.1B -> TTM $114.0B | Upper-mid band | AI chip foundry demand, overseas fab construction (U.S., Japan) |
| SK hynix | 32.2x | FY2023 KRW 32.8T -> FY2024 KRW 66.2T -> FY2025 KRW 97.1T | Mid band | HBM3E/HBM4 shipment expansion, memory cycle recovery |
Analysis:
NVIDIA’s current PER of 47.5x is actually in the lower range compared to its pre-AI boom (2021–2022) average PER (roughly 50–80x range, per MacroTrends). This is a result of explosive earnings growth — FY2025 net income of $72.9B — causing multiple compression, entering a phase where “earnings growth outpaces share price appreciation.” However, de-rating risk exists when revenue growth decelerates (FY2025 +114% to likely lower FY2026E estimates).
AMD’s PER of 77.0x appears to exceed the upper bound of its historical band. AMD traded in the PER 30–50x range during the 2019–2021 Ryzen/EPYC cycle, and the current 77.0x reflects additional premium from AI accelerator expectations.
SK hynix’s PER of 32.2x is estimated at mid-band level considering typical memory semiconductor cycle valuation patterns. Memory PER historically compresses at cycle peaks (rising earnings compress PER), and the current structural growth driver of HBM partially offsets the cycle premium.
6.3 Big Tech/Cloud Group Multiple Trend Analysis
| Company | PER (Current) | Estimated Position vs. History | Multiple Change Drivers |
|---|---|---|---|
| Amazon | 27.7x | Lower-mid | Massive profitability improvement (operating income FY2023 $36.9B -> FY2025 $80.0B) compressing PER |
| Alphabet | 28.3x | Mid band | AI search transition uncertainty vs. Cloud growth acceleration |
| Meta | 21.5x | Near lower band | 2022–2023 metaverse investment skepticism, recovering on AI pivot but discount persists |
| Microsoft | 25.1x | Lower-mid | Azure AI growth offset by large-cap premium compression |
| Oracle | 28.8x | Near upper band | Cloud transition narrative driving re-rating |
Research References
- NYU Stern (Damodaran) — PE Ratios by Sector
- Goldman Sachs — AI Investment: Still Focused on Infrastructure
- Stock Analysis — Real-Time Valuation Data
Related Analysis
- Palantir Deep-Dive: 70% Revenue Growth and the Reality Behind a $313B Valuation
- AI Agents Update: From MCP Security Crisis to $3 Trillion Data Center Investment
- The Real Bottleneck in the AI Infrastructure War: Power and Semiconductors in the $1 Trillion CapEx Era
Analysis:
The Big Tech group overall is in a compressed multiple state compared to 2021–2022 highs. This is structural multiple compression driven by earnings growth outpacing share price appreciation, with FCF pressure from AI CapEx expansion as an additional factor. Meta (PER 21.5x) in particular dropped below 10x during the 2022 metaverse investment controversy before recovering, but the peer discount persists — suggesting room for further re-rating.
Oracle is the exception, with the cloud transition narrative pushing multiples to the upper band. Oracle traditionally traded at PER 15–20x; the current 28.8x re-rating reflects market acceptance of AI infrastructure investment benefits. However, de-rating risk exists if negative FCF persists.
6.4 Value Chain Group — Re-Rating/De-Rating Triggers
| Company | Re-Rating Trigger | De-Rating Trigger |
|---|---|---|
| Hanmi Semiconductor | HBM4/HBM5 TC bonder order expansion, overseas customer diversification | Hanwha Semitec competition intensifies, SK hynix in-sourcing |
| Doosan Enerbility | U.S. nuclear order confirmation (AP1000), SMR market opens | Order delays, tighter nuclear regulations, insufficient margin improvement |
| Western Digital | Sustained AI data center HDD/SSD demand exceeding supply | Storage cycle downturn, price competition intensifies |
AI Infrastructure 12-Company Comparative Analysis Series
Part 1: Peer Groups & Multiples | Part 2: Valuation Deep-Dive (Current) | Part 3: Sector Trends & Investment
Part 2 Summary
The premium/discount analysis reveals that high premiums in the AI semiconductor group are partially justified by growth rates but carry persistent cycle risk. Within the Big Tech group, Meta appears relatively undervalued while Oracle sits in overvalued territory.
Frequently Asked Questions (FAQ)
Q1. What are the key implied valuation findings?
Using EBITDA data obtained via Alpha Vantage MCP for all U.S. companies, comprehensive EV/EBITDA-based implied valuations show Amazon as the most undervalued Big Tech name at 13.1x vs. 15.7x peer median, with approximately 20% upside to implied EV of ~$2,288B.
Q2. What does the EV/EBITDA analysis reveal?
Amazon (EV/EBITDA 13.1x) trades at a 16.6% discount to the Big Tech peer median (15.7x), making it the most undervalued on this basis. Applying the peer median implies EV of ~$2,288B versus current EV of ~$1,908B — approximately 20% upside.
Q3. What is the historical multiple band analysis?
The analysis uses available historical revenue/earnings trends and current multiples to qualitatively assess where each company’s multiples stand relative to history, given the absence of detailed 3–5 year time-series data required for a formal +/-1 standard deviation band analysis.
—
Disclaimer: This analysis is for educational and informational purposes only and does not constitute a recommendation to buy or sell any securities. U.S. company data was sourced from Alpha Vantage MCP API (S&P Capital IQ-based), and Korean company data from FnGuide. Accuracy and completeness of data are not guaranteed, and figures may differ due to timing discrepancies. Investment decisions should be made based on individual judgment and risk tolerance, and additional analysis tailored to personal investment objectives is recommended.