AI Infrastructure
Platforms, networking, cloud, power, and frontier compute for the AI buildout.
Platforms, networking, cloud, power, and frontier compute for the AI buildout.
Public market coverage
Sector: Conglomerate / AI Exposure (Upstage minority stake)
Date: June 2026
Verdict: Avoid, Fair Value ~KRW 9,000
Abstract: SK Networks closed June 9 at KRW 14,290 after a five-week, 117% advance with two limit-up closes in ten sessions, both tied to its 12.9% stake in Upstage rather than to the operating businesses. Against a KRW 2.87T market value, the operating company at its last pre-re-rating price (KRW 8,400 per share) leaves a KRW 1.2T residual on a stake marked near KRW 258B, capitalizing Upstage at KRW 9.2T: 4.6 times the last round and nearly double the KRW 5T bull-case listing scenario, before any discount. Fair value of ~KRW 9,000 (operating midpoint KRW 7,400 plus a probability-weighted stake value of KRW 1,670) implies about 37% downside; even the undiscounted bull case reaches only ~KRW 11,600, 19% below the market, and the price stands 53% above the street average target. AVOID: no entry above KRW 9,500, with the Upstage IPO filing as the repricing event and August 18 second-quarter results as the margin test.
Sector: AI Networking / Communications Equipment
Date: May 2026
Verdict: Conditional GO - Tranche entry within $135–$185 base band; do not chase above $185
Abstract: Post-earnings sell-off of 15.4% on May 6 reflects multiple compression on a misread of supply-constraint commentary, not an earnings break. Three quarters of revenue acceleration to 35.1% YoY, deferred revenue at $6.2B (+101% YoY), and raised FY26 guide of $11.5B (+27.7%) all point to demand strength rather than a growth ceiling. Probability-weighted 12-month target of $181 against $144 entry, with pre-committed thesis-break triggers at Q2 revenue below $2.65B, Spectrum-X share above 35%, or single-customer concentration above 30%.
Sector: AI / Megacap
Date: May 2026
Verdict: GOOGL: GO at 23x forward P/E | NVDA: Conditional Avoid at 32x
Abstract: Ranked 11 megacaps across the AI value chain on chip, infrastructure, model, and application exposure. GOOGL scored highest at 35/40 as the only name with anchor presence in all four layers, with probability- weighted 12-month expected value of +20.0% versus NVDA at +2.7%. Flagged NVDA as the clearest match to the Novo Nordisk pattern, with four hyperscaler custom silicon programs (TPU, Trainium, Maia, MTIA) and AMD's $60-100B Meta deal compressing chip-layer dominance over a 12-24 month horizon.
Sector: Emerging Tech / Quantum Computing
Date: April 2026
Verdict: Trigger-conditional watch — entry only after May 6 Q1 earnings, position capped at 3% of portfolio
Abstract: A pre-commercial company where DCF and forward P/E do not apply (2025 revenue $130M against $510M GAAP loss; P/S 50–120x). I built a separate rule set for milestone-driven equities and defined two entry gates: Q1 revenue above the $51M guidance ceiling and commercial customer mix holding at 60%+. Probability-weighted 2030 fair value ~$114 against current $30–35, with 30% downside scenario priced in through hard 30% stop-loss and strict size limit.
Sector: Power & Utilities
Date: April 2026
Verdict: Conditional GO — staged entry within $260–$305 base band; weighted average target ~$280; chase rule above $325
Abstract: Recommended $260–$305 entry after a 28% selloff from $402.95 driven by FERC and PJM regulatory repricing, not a fundamental break. Anchored the downside on 32.4 GW of operating nuclear capacity valued at ~$2,500/kW versus $12,000/kW new-build replacement cost, with the 20-year Microsoft Crane PPA and the post-Calpine 55 GW combined platform underwriting a 13%+ EPS growth path. Probability-weighted 12-month expected value of $354 (bull $465 / base $360 / bear $230) implies ~20% upside; thesis reassessment is triggered if the mid-2026 FERC ruling on PJM large-load interconnection is categorically restrictive.
Sector: AI Infrastructure / Mega-cap Tech
Date: April 2026
Verdict: Conditional GO with stage entry between $275–340; base case fair value $385 (+21% from $317)
Abstract: Three concerns drive the 18% peer discount: AI cannibalization of Search, $180B CapEx burden, and DOJ antitrust overhang. Each fails under examination. Search revenue growth accelerated from 9.8% to 17% YoY through 2025 while ChatGPT, Claude, and Perplexity scaled, contradicting the cannibalization thesis. CapEx intensity at 42.9% of revenue is in line with MSFT (43.5%) and below META (45.5%), so the discount is not relative. Chrome divestiture was rejected in September 2025, removing the structural tail risk. Probability-weighted fair value of $377 against $317, with a tighter bear case (-8%) than bull case (+47%).
Sector: AI Infrastructure
Date: February 2026
Verdict: GOOGL: GO (High conviction) | MSFT: Conditional GO | AMZN: Watch
Abstract: Multi-cloud has shifted from best practice to regulatory mandate after the October 2025 AWS US-EAST-1 (15hr) and Azure Front Door (~8hr) outages, combined with the BoE/FCA CTP regime, DORA, and EU Data Act. Microsoft's banking lock-in is eroding at the margin while Google Cloud captures disproportionate share through TPU-Gemini vertical integration, validated by $155B backlog (+82% YoY) and operating margin expansion to 23.7%. SOTP analysis identifies Google Cloud as the cleanest re-rating opportunity at 12x revenue versus the discount embedded in current consolidated multiple.