Penguin Solutions (PENG) Earnings Transcript

Penguin Solutions reported Q2 net sales of $343M (-6% YoY) with non-GAAP gross margin 31.2% and non-GAAP EPS $0.52 (flat YoY). Integrated memory drove results at $172M (up 63% YoY) and management raised FY26 midpoint guidance to +12% net sales growth and $2.15 non-GAAP EPS (from $2.00), while advanced computing was weak at $116M (-42% YoY) and full-year advanced computing guidance of -25% to -15%. Company highlights include launches of MemoryAI/CXL-based solutions, $32M of share repurchases (~1.7M shares) with $64.5M remaining, quarter-end inventory $322M (vs $200M a year ago), cash $489M and $450M debt (net cash position); risks remain from rising memory input costs and supply-chain lead times.

Analysis

Penguin’s deliberate shift toward a memory-heavy AI factory strategy is creating a bifurcated value chain: memory OEMs and interconnect/IP owners capture outsized margin and working-capital upside while systems integrators absorb compressed product gross margins but gain recurring services and software annuity optionality. That subtle swap — margin from boxes to memory and software — will mechanically re-price multiplier effects across the supply chain over 6–18 months, favoring firms that own CXL/IP, KV-cache software and photonic interconnect stacks.

CXL and MemoryAI KV-cache are demand amplifiers for GPU-anchored inference but also efficiency multipliers: increased memory pooling raises effective tokens/sec per GPU, which can boost near-term revenue per deployed GPU while lengthening upgrade cycles. That creates a two-speed outcome for GPU vendors and integrators: higher attach value now (more modules, cards, interconnect), but a risk of slower unit replacement in 2–4 years as utilization per box rises — a nuanced positive for NVDA in the next 12 months but a moderating structural effect over the medium term.

The fastest reversals would come from a sharp memory-price deflation or an abrupt easing of supply-chain lead times that re-rates the mix back to hyperscaler hardware (weeks–months). Conversely, accelerated photonics commercialization (12–24 months) is an asymmetric upside for component/IP owners — it converts a niche R&D bet into recurring, higher-margin systems revenue and consolidates winners early. Watch booking-to-revenue conversion and days-inventory cadence as the earliest leading indicator that pipeline strength is translating into sustainable on-prem inference deployments.

Key Decisions for Investors

• Buy MRVL (6–12 month horizon): establish a core long (10–15% position of risk allocation). Rationale: Marvell’s acquisition/integration of photonics IP materially derisks the photonic memory pathway and positions it as a preferred CXL/optical interconnect supplier; target +30% if adoption accelerates into 2026–27, downside -30% if memory prices collapse or integration stalls. Size position to tolerate event volatility.
• Bull call spread on NVDA (6–12 months): buy a moderate-cost call spread to capture near-term inference-driven GPU demand without open-ended delta. Rationale: NVDA benefits from more on‑prem inference installs and GPU attach via MemoryAI/CXL; reward skewed to upside from increased attach while limiting premium loss if enterprise capex slips. Aim for ~3:1 upside/downside payoff (max loss = premium).
• Add DELL as a tactical systems-integration exposure (3–6 months): small overweight with a hard 10–12% trailing stop. Rationale: OEMs win incremental services/turnkey projects from enterprise AI factories even as product margins compress; position size should be conservative in case enterprise capex growth disappoints.
• Risk hedge — buy a 6–9 month put spread on MRVL or NVDA sized to ~10% of portfolio long exposure: protects against a sudden memory-price deflation or hyperscaler re-entry that collapses near-term attach economics. Use put spreads to cap hedge cost while retaining upside participation.