Semiconductor Weekly: March SIA and Earnings Previews Signal Broad Non-AI Recovery, but AI Leadership Faces Growing Uncertainty

Core Thesis

The March SIA data and earnings previews across GFS, AMD, SWKS, ALAB, NVTS, MCHP point to a two‑speed semiconductor market. Non‑AI demand in analog, MCU, discrete, and trailing‑edge foundry is inflecting positively—supported by March broad markets beats, distributor inventory behavior, and company‑specific guidance. Meanwhile, AI‑related names (AMD GPU, NVTS GaN) face rising skepticism: AMD’s GPU ramp is clouded by technical and system integration risks, while NVTS trades at a premium that already prices in future share gains that may not materialize for years. The key outlier is memory: DRAM sequential growth was below consensus, but absolute levels remain extreme. Investors should position toward names with clear non‑AI cyclical exposure (MCHP, GFS) and AI networking beneficiaries with conservative expectations (ALAB), while underweighting high‑valuation AI‑themed names (NVTS) and exercising caution on AMD GPU consensus.

What the Market May Be Underpricing

The breadth and speed of the non‑AI recovery are under‑appreciated. March SIA broad markets (discrete +19.0% m/m, MCU +25.5% m/m, analog +17.3% m/m) all beat or matched optimistic estimates. Three‑month average y/y growth for analog improved to 14.9% (from 13.3% in Feb) and for MCU to 16.9% (from 13.5%)—the latter approaching a two‑year high. Distributor inventories are still above historical median but declining, while customer days‑of‑inventory dropped 6 days q/q, indicating genuine end‑demand pull rather than restocking. GFS’s largest customer QCOM built $842mn of inventory in the quarter, suggesting foundry orders are needed to replenish internal stocks. MCHP’s JunQ revenue estimate of +9.8% q/q is 3.4pp above Street consensus, supported by expanding industrial and A&D business. These data points collectively argue for a cyclical upswing that is not yet fully discounted in valuations of analog/MCU and foundry names.

Simultaneously, AI networking demand through ALAB is extremely strong (1.6T transition, copper resilience, Taurus/Scorpio diversification), but the stock suffers from Amazon concentration fears that have made consensus cautious—creating a setup for beats and raises. The risk is that investors underweight ALAB due to that single‑customer overhang, missing the multi‑year upgrade cycle.

Evidence Chain

Non‑AI cyclical recovery is real and broadening.

• March SIA: discrete (+19.0% m/m), analog (+17.3%), MCU (+25.5%) all exceeded Morgan Stanley estimates. Three‑month y/y growth rates for analog (14.9%) and MCU (16.9%) are accelerating.
• GFS: trailing‑edge foundry utilization is still depressed but pricing is not under pressure. Disti survey shows inventory build intent at highs not seen since 2021. SiPho revenue at least doubling in 2026 reiterated. Largest customer QCOM inventory build supports near‑term revenue.
• MCHP: JunQ estimate +9.8% q/q vs Street +6.4%. Inventory reserves ($58mn) and underutilization charges ($51mn in DecQ) are rolling off, targeting 62.6% gross margin in JunQ. Industrial (30% mix) and A&D (18%) are key growth verticals.

AMD: server CPU strong, but GPU is a wildcard.

• Server CPU growth accelerating materially above ~35% y/y in 2025 is well‑supported. Intel’s server beat was price‑driven (units -5% y/y), limiting Intel’s competitive read‑through.
• GPU uncertainty: MI355 is behind state‑of‑art; Helios racks (MI400) are key for 2H26 but qualification and customer acceptance are uncertain. Below‑consensus GPU estimates for 2026 leave room for downside if GPU numbers surprise up, but the base case is cautious. Gross margin base is difficult after Q1 MI308 inventory benefit and server ASP mix headwinds.

NVTS: GaN story is being priced as if already won.

• Stock up 140%+ YTD, trading at 33x NTM sales (a 7‑turn premium to Innoscience). Competition from STMicro, Infineon, Renesas is intensifying. Mgmt reiterated immaterial data center revenue in 2026, 800V only in 2027. Bear case assumes no share capture (PT $3.60). The current valuation relies on 60x CY27 sales for a company with $48mn revenue in that year—a steep multiple on an uncertain future.

Memory: DRAM is strong but leading indicators soften.

• March DRAM +10.2% m/m (below MS est 31.7%, 5‑yr avg 22.7%). ASP +4.9% m/m, bits +5.1%. Three‑month y/y sales growth is a record 245.1%, but m/m deceleration hints at maturation. NAND remains robust (+33.7% m/m, in‑line). CY26 industry forecast -91% y/y (vs prior -99%) reflects an expected massive correction. The memory cycle peak may be approaching.

Short interest concentration signals skepticism on thematic names.

• Average short interest in semi sector rose from 5.6% (Jan) to 6.0% (May). IONQ (22.9%), NVTS (21.0%), SWKS (14.8%), AEVA (13.4%) are heavily shorted. Meanwhile, AMD (2.2%), AVGO (1.2%), NVDA (1.2%) have low short interest. This divergence shows that institutional skepticism is focused on “story stocks” with binary outcomes.

Key Disagreements & Risks

1. Macro risk: The whole recovery thesis depends on AI capex sustaining its current trajectory and non‑AI demand genuinely re‑accelerating. A slowdown in cloud spending or a sudden drop in memory pricing could derail both themes.
2. AMD GPU downside: Consensus still expects meaningful GPU revenue ramp in 2026. If MI355/Helios slip or fail to win key accounts, AMD could face earnings revisions and multiple compression.
3. NVTS competitive pressure: GaN market is fragmenting; larger players with captive supply chains may dominate data center and auto sockets. High valuation leaves little room for execution slip.
4. Concentration risk: ALAB (Amazon), SWKS (Apple ~60% of revenue), GFS (QCOM) are all highly dependent on single customers. Any share loss or procurement shifts would materially impact financials.
5. Memory correction: DRAM already decelerating sequentially. If HBM demand disappoints, the 2026 revenue collapse (-91% y/y forecast) may be even worse, dragging down entire semi index.

Valuation & Trading Implications

• Overweight ALAB: cheap relative to growth (0.65x EV/sales/growth) with multi‑year AI networking tailwinds. Consensus is too cautious on Amazon concentration; earnings beats are likely.
• Equal‑weight GFS and MCHP: good cyclical setup, but valuations already reflect some recovery (GFS 26x CY27 EPS; MCHP 28x — top of +1 st dev). Upside limited unless guidance materially beats.
• Equal‑weight AMD: server CPU strength is priced in; GPU uncertainty keeps us from being more constructive. Upside only if GPU execution surprises materially.
• Underweight NVTS: valuation disconnected from fundamentals. The GaN opportunity is real but years away, and competition is fierce. Short interest (21%) reflects the risk of a squeeze, but the fundamental case remains weak.
• Avoid SWKS: mobile end‑market is challenging; QRVO acquisition is distant (CY27 close). Apple content flat y/y but high concentration keeps risk‑reward balanced at best.

The market is correctly identifying that non‑AI demand recovery is underway, but it is also over‑pricing some AI‑themed names. Focus on companies with tangible cyclical exposure and credible guidance, while avoiding those where the narrative exceeds the numbers.