Marvell Technology, Inc.

▲ Bull Case

• ◆Google inference XPU win confirmed + inference TAM $80-100B (not $40-45B): Market underestimates inference workload volume (5-10x training); Google confirms MRVL as inference XPU partner; adds $3-5B annual revenue from FY2028; consensus EPS estimates rise 20-30%; stock re-rates to 40x FY2028E $7.50 EPS: $300 (+69%)
• ◆Celestial AI co-packaged optics becomes next Inphi: Silicon photonics emerges as dominant optical architecture for AI clusters beyond 2027; Marvell's Celestial AI is 2-3 years ahead of competitors; 2029-2031 revenue opportunity $5-10B from co-packaged optics alone; total addressable market re-rated upward: $350+ (+98%)
• ◆AI capex accelerates 40%+ through 2028-2029 with no digestion pause: AMZN, MSFT, GOOGL, META, xAI, Oracle all build unprecedented AI infrastructure simultaneously; MRVL captures 20-25% of custom silicon market in a $50B+ market; FY2030 revenue $20B+; true owner earnings $5B+; 35x = $175B → $200 (+13% from here; $175B company vs. $155B today = significant value creation)

▼ Bear Case

• ◆AI capex digestion in H2 2026—hyperscalers take a pause: AWS, MSFT, GOOGL simultaneously guide capex flat-to-down for 2027; Trainium 3 program delayed 4-6 quarters; MRVL FY2027 misses $11B target; EPS revision to $3.50 FY2028; P/E compresses to 28x: $98 (-45%)
• ◆Broadcom wins unexpected Google design competition + AMD enters custom AI silicon: AVGO takes Google's next-gen TPU program that MRVL was targeting; AMD introduces custom silicon offering and wins AWS pilot; MRVL market share in custom AI ASIC compresses from ~25% to 15-18%; revenue trajectory $10-12B by FY2028 vs. $14-16B target: $110-125 (-30-38%)
• ◆Inventory correction + margin compression: After Trainium 3 ramp, hyperscalers accumulate inventory; MRVL faces a 2-3 quarter revenue trough (as seen in memory/CPU cycles); operating deleverage compresses non-GAAP margins from 32% to 25-27%; EPS $3.00-3.50 during trough; multiple contracts: $85-105 (-41-52%)

“Is MRVL at $177 pricing in flawless execution of an $11B FY2027 target (and $15B FY2028), or is there still a genuine margin of safety for a 3-5 year holder? Bull view: The $11B FY2027 target requires ~34% growth from FY2026's $8.2B; Q1 FY2027 guide of $2.4B implies $9.6B annualized—requiring only modest H2 acceleration. Trainium 3 ramp (AWS) and Maia 200 (Microsoft) already in production qualification; revenue gating-limited by customer cluster deployment, not MRVL supply. FY2028 $15B achievable if Google inference XPU program added. On FY2028E true owner earnings of $3B, stock is 52x—expensive, but not absurd for #2 position in most important semiconductor design win in history. Bear view: At $155B market cap, MRVL priced for everything to go right. Stock is 5-6x FY2026 revenue—assuming no execution delay, no competitive incursion, no AI capex cycle pause. Broadcom's $14B AI revenue guidance for FY2026 vs. MRVL's $6-7B shows duopoly heavily weighted toward AVGO. MRVL's SBC ($600M/yr) means "true" FCF barely covers dividend—not generating owner earnings justifying the multiple on current-year basis. Tie-breaker: Q4 FY2027 data center exit rate (target $3B+/quarter). If MRVL hits $3B/Q by early 2027, $15B FY2028 target becomes credible and FY2030 owner earnings of $5B+ justify $175-200 stock price. If exit rate disappoints (<$2.5B/quarter), $150B+ market cap is indefensible.”

• ◆Q1 FY2027 earnings (June 2026)—first test of $11B revenue trajectory; ≥$2.4B = bull case; <$2.1B = bear case signal
• ◆Q4 FY2027 data center quarterly exit rate (Feb 2027)—single most important milestone for FY2028 $15B credibility; >$3B/Q = thesis confirmed (+20-30%); <$2.5B/Q = thesis revision (-25%+)
• ◆Google inference XPU program confirmation (2026-2027)—adds $3-5B FY2028 revenue not in consensus; 40% probability by end of FY2027; +20-30% upside if confirmed
• ◆Hyperscaler AI capex guidance (Q2-Q3 2026)—macro validation of AI investment thesis; flat/up guidance = bull; down >20% = bear signal
• ◆Celestial AI silicon photonics revenue signal (FY2028-2029)—co-packaged optics addressable market $5-10B+; 30% probability of meaningful revenue before FY2029

• ◆AI capex digestion pause (6-12 months): 30% probability; High severity; could delay Trainium 3 and reduce FY2027 revenue trajectory; simultaneous hyperscaler capex flatness would be triggering
• ◆AWS delays Trainium 3 program: 20% probability; High severity; economically irrational given >$5B ecosystem investment, but mid-cycle delay costs are real
• ◆Broadcom captures unexpected design win from MRVL's customer: 20% probability; Moderate severity; would signal competitive incursion and duopoly erosion; different primary customers mitigate (AVGO=GOOGL/Meta; MRVL=AWS/MSFT)
• ◆Q1 FY2027 revenue below $2.1B: 25% probability; High sentiment severity; implies program delay and $11B target already at risk; stock correction 20-30%
• ◆SBC dilution ($600M+/yr) continues long-term: Near-certainty; Moderate severity; creates wedge between non-GAAP EPS and true owner earnings; must be modeled honestly