“AI is moving from experimentation to production with workloads increasingly shifting towards real-time inference.”（CEO）
“We are building Penguin into an AI factory platform company.”（CEO）
“Memory architecture is becoming increasingly central to AI performance, particularly as inference workloads scale.”（CEO）
“We are raising the midpoint of our full year net sales and EPS outlook.”（CEO）
“Our non-hyperscale AI/HPC net sales were down 35% year-over-year in the quarter, but up 50% for the first half of the year.”（CFO）
Q&A Batch (6-7 of 7)
Q6 — Ananda Baruah
Topic: Inference adoption, agentic AI, and CXL product outlook
Key points:
Kash believes they are "early in the adoption of inference"; customers are moving toward agentic AI, where inference powers agentic workloads.
Inference architecture is changing, making memory critical (e.g., MemoryAI portfolio for faster LLM responses).
CXL adoption is timely due to inference needs; a new order from an enterprise generative AI company working on inference workloads uses CXL cards for memory pooling.
Photonic memory appliance (with Celestial AI/Marvell) provides next-level capability, but CXL itself is an advantage; KV Cache server also addresses inference latency.
Mgmt stance: Bullish — sees inference as a growth driver with multiple product opportunities (CXL, MemoryAI, photonic memory) as adoption scales.
Q7 — Kevin Cassidy
Topic: Memory gross margin dynamics and CXL margin outlook
Key points:
Q2 memory gross margin was up due to mix (stronger flash demand, higher-margin product) and pricing increases from timing of inventory purchases vs. shipments.
If price increases slow in H2, margin favorability from timing will decrease; they use the balance sheet to secure inventory in a tight market.
CXL systems are expected to be higher margin than the module business due to software and hardware differentiation.
Mgmt stance: Neutral on near-term memory margins (dependent on pricing trends); bullish on CXL as a higher-margin solution down the road.