Mark Lipacis

Rene expressed high confidence in growing beyond 50%, citing customization advantages and the benefit of standardizing on a single CPU architecture and software stack across data centers.

Jason quantified ~$0.01/quarter FX headwind for the year (~$0.04 total). Two-thirds of OpEx in euros and pounds, with some unhedged exposure from increased R&D spending.

Jason clearly explained no direct tariff impact on services. Indirect risk limited to 10-20% of royalties from US-destined shipments, with potential couple percent royalty impact at most.

Jason confirmed ~50-50 CSS split between infrastructure and client. Rene added that chiplet architecture will be pervasive across nearly every SoC, creating a major CSS opportunity.

Murphy gave forceful defense of custom business. Not compromising interconnect investment to fund custom. Custom provides strategic advantage as bleeding-edge technology driver across nodes, packaging, IP. Customers fund NRE reducing Marvell's own investment burden. Dismissed competitive noise from past year as demonstrably wrong, analysts retracting notes, inaccurate articles. Business grew from zero to $1.5B and will continue scaling.

Murphy described triangulation process for forecasting, citing sole-source positioning and joint supply chain planning with customers. Sees strong setup for FY26 and FY27 across custom, connectivity, and storage.

Confirmed $40B custom TAM out of $75B data center. Described competitive moat as combination of IP roadmap, A0-quality execution on 100B+ transistor designs, manufacturing capacity, and field support. TCO advantage drives custom adoption alongside merchant solutions.

Baratz said gate-model revenue starts with government contracts, then likely system sales to research institutions; significant QCaaS revenue is not expected until around 2032.

Joe McCormick of Evercore asked, covering for Mark Lipacis. Baratz expects system sales mostly research/academic near term, with possible commercial sales tied to blockchain and AI.

Joe McCormick stood in for Mark Lipacis. Markovich said QCI contributes professional-services, QCaaS and government revenue this year, with a system-sales pipeline building.

Gary acknowledged WFE intensity at ~15% but pivoted to argue the old 1/3 logic, 1/3 ICAPS, 1/3 memory framework is outdated given AI-driven divergence in segment growth rates.

Gary confirmed high confidence in sustained margin growth, citing years of investment in high-value solutions for AI-enabling segments where AMAT has strong positions.

Colette confirmed hyperscalers sit around 50% of revenue while AI model makers, enterprises, supercomputing, and sovereigns are growing faster, and Jensen framed CUDA ubiquity as the enabler.

Colette confirmed enterprise grew 2x year on year, similar to CSPs. Jensen argued enterprise will be 'by far larger' long term, citing three computers per industrial company, agentic AI for employees, AI factories for products, and robotic AI inside the product itself.

Roche argued concentration was higher then; the industry today is broader and deeper with a pervasive AI gravity field, unlike prior cycles.