Semiconductors
Memory, foundry, and advanced-node economics.
Memory, foundry, and advanced-node economics.
Public market coverage
Sector: Technology / Semiconductors (AI Compute and Networking)
Date: June 2026
Verdict: Overweight. Fair value ~$275 / Range $175 to $350. Accumulate gradually below $260, preference below $240, no new buying above $260, trim modestly on strength above $260.
Abstract: NVIDIA closed at $205.19 on June 12, 2026, after a week in which the share price fell as much as 10.1 percent on an AI funding scare while the demand data accelerated. The report's central claim: the burden and the benefit sit at different addresses. Oracle's $40B FY2027 funding plan dilutes Oracle shareholders and funds NVIDIA's purchase orders, and on June 11 the market traded exactly this way, marking Oracle down 11.9 percent while semiconductors rallied nearly 8 percent on the same dataset. The update examines four questions: cash conversion, pricing power, capital return, and demand concentration. Free cash flow runs from $96.7B toward an estimated $186.5B in FY2027, a forward yield near 3.7 percent. Rack prices rose generationally from roughly $3.2M to $6 to 7M while gross margin held at 75 percent. Blending a through-cycle case near $185 with a structural case near $345 yields fair value near $275, about 34 percent above the close and below the $298 to $306 consensus, which applies no earnings-quality haircut. The decisive signal is the first negative quarter in the unrealized investment gains line, possible as soon as August 27; until then the discipline is accumulation below $260.
Sector: Semiconductors (Memory)
Date: June 2026
Verdict: Neutral, Fair Value ~KRW 1,950,000
Abstract: SK Hynix entered June with record first-quarter results (revenue KRW 52.6T, +198% Y/Y; operating profit KRW 37.6T at a 72% margin) and a share price that had outrun its own estimate revisions. The test arrived early: Broadcom guided third-quarter AI revenue near $16B against a $17.2B consensus and buy-side expectations near $18.5B, and a +172,000 US payrolls print repriced rate risk, sending the shares down a combined 12.3% over June 4 to 5 to KRW 2,070,000 on no company-specific news. The close now sits at the 38-analyst average target (KRW 2,076,603) and within 1% of our probability-weighted scenario value (~KRW 2,056,000). Fair value is unchanged at ~KRW 1,950,000, blending a through-cycle and a structural-growth lens at the estimated 70/30 earnings mix; the implied downside is about 6%, and the market-implied structural weight fell from above 50% to roughly 37% in two sessions. The cycle triggers themselves did not move: the hyperscaler capex consensus remains +81%, the HBM book is committed, supplier capex is the one warning firing, and the ADR window now opens into a broken tape. The rating is NEUTRAL: the price sits at the top of the Fair band, where the rule is hold or accumulate gradually, with the ADR pricing, the third-quarter price peak, and the June 16 to 17 FOMC as the near-term tests.
Sector: Technology / Semiconductors (AI Compute and Networking)
Date: June 2026
Verdict: Overweight. Fair value ~$275 / Range $175 to $350. Accumulate aggressively below $150, add $150 to $180, accumulate gradually $180 to $260, no new buying $260 to $285, trim selectively $285 to $350, trim aggressively above $350
Abstract: NVIDIA at $214.50 (June 3, 2026 close) sits 9% below its May 14 record after a 5.7% Vera Rubin rally was given back within two sessions. Q1 FY27 revenue of $81.6B grew 85%, but the central data point is earnings quality: roughly $15.9B of GAAP profit, about 27%, was unrealized gains on stakes in NVIDIA's own customers, a portfolio that doubled to $73.6B in one quarter. Blending a through-cycle DCF of $185 with a structural path of $345 by the estimated earnings mix yields a fair value near $275, about 28% above the market and below the $297 consensus, which applies no earnings-quality haircut. The rally is earnings-led, not multiple-led: shares rose 59% in twelve months against estimate revisions near 75%, compressing the forward multiple from roughly 40x to near 30x. The decisive signal is a first negative quarter in the equity-gains line, which would lead any revenue deterioration; until then the discipline is accumulation below $260 with full size reserved for $150 to $180.
Sector: Semiconductors (Memory - DRAM, NAND, HBM)
Date: May 2026
Verdict: Avoid at current price — Fair value ~$600 / Range $405 to $1,000. Accumulate below $450, gradual entry $450 to $643, hold $643 to $800, trim above $800, no entry above $1,000
Abstract: Micron at $942.36 (May 28, 2026 intraday) has returned +892% in twelve months following Q2 FY26 revenue of $23.86B (+196% YoY) with gross margin guided to 81%. The central data point is earnings composition: conventional DRAM ASP rose roughly 110% YoY against 40% bit growth, making about two thirds of the profit stream cyclical, while HBM, the genuinely structural product, runs near 12% of revenue and is not disclosed separately. Blending a through-cycle DCF of $405 at two thirds weight with a structural path of $1,000 at one third yields a fair value near $600, 36% below the market, and the forward P/E of 8.5x sits above SK Hynix at 5.4x despite a smaller structural HBM share. Cycle-turn signals are already firing: NAND spot fell 30 to 40% in a month, contract price momentum is decelerating, and CXMT conventional supply is scaling, pointing to contract roll-over in 2H26 to 1H27. Upgrade triggers are confirmed HBM disclosure above 50% of revenue or a verified multi-year price lock; downgrade confirmation is two consecutive months of negative conventional contract pricing or inventory days above 130.
Sector: Semiconductors / Foundry & CPU
Date: May 2026
Verdict: Avoid — no entry above $50; existing holders trim aggressively above $76
Abstract: Twelve-month rally of 525% ($19.98 to $124.92) prices Foundry execution that has not occurred. Q1 2026 external Foundry revenue of $174M annualizes to $696M, only 14-23% of the $3-5B annual run-rate management targets for 2027 breakeven. Forward P/E of 152x is the highest in large-cap semiconductors, with all 32 sell-side analysts publishing targets below market (median $65, highest $100). Sum-of-parts produces a Bear/Base/Bull range of $19/$28/$50, with probability-weighted target of $31 against $125 market. Three pre-committed thesis-break triggers: external Foundry revenue above $750M quarterly before year-end 2026, Apple deal disclosure confirming $2B+ 2027 contribution at margins comparable to TSMC, or NVIDIA reversal on its 18A decision.
Sector: AI Infrastructure / Semiconductors
Date: April 2026
Verdict: Conditional Go at $370.50, scenario-weighted target $454 (+22%); Two-tranche entry: 50% at market, 50% reserve for pullback into $320–350
Abstract: Q1 2026 gross margin reached 66.2% while peer foundries operated at 20–30% or losses, confirming a profitability gap that capex alone cannot close. N2 wafer lead times extend into 2028 and CoWoS packaging is fully booked through mid-2026, turning the AI cycle into a pricing power event rather than a volume event. Bear case $260 (-30%) anchored on AI capex deceleration plus tariff imposition.