CJ Muse

Pointed back to the Investor Day framework and flagged upside, but declined to update the CSBG growth rate numerically.

Confirmed engagement with a new U.S. logic customer but gave no sizing or timing; Archer added only qualitative share-opportunity framing.

Doug indicated CSBG should sustain near March record levels through remaining quarters, supported by high utilization and Equipment Intelligence innovation.

Tim confirmed active 2027 conversations and planning discussions extending to 2028 fab openings, with longer visibility improving supply chain readiness.

Doug said Malaysia is their biggest location and they are expanding globally, but customer and product mix will matter more for gross margins than volume ramp in the near term.

Tim distinguished near-term equipment demands for enterprise SSDs and HBM from longer-term data center announcements, describing both as along the same technology transitions.

Tim cited foundry/logic moly, gate-all-around, backside power, advanced packaging expansion across device types, and NAND layer scaling as key drivers.

Tim explained two-thirds of bits remain at 1XX layers, creating a multi-year upgrade runway for moly, carbon gapfill, and new tools needed above 200 layers.

Tim ranked NAND upgrades as the biggest year-on-year change, pointing to two-thirds of bits still on sub-200-layer nodes and strength in gate-all-around and advanced packaging.

Management said it funded its priorities and still delivered over a point of operating margin leverage, hoping to continue leverage in 2025.

Mark said deposits are unrestricted but did not change near-term gross-cash needs; Micron will hold liquidity for operations and investment.

Sanjay confirmed SCA discussions span multiple customers across multiple markets (not just hyperscale), while declining to comment on specific CapEx or ROIC tie-ins.

Sanjay confirmed non-HBM margins now exceed HBM margins but said HBM demand continues to be strong, did not update the prior HBM TAM outlook, and emphasized Micron's portfolio approach across HBM, LP, SoC-M, DDR5, and SSDs.

Mark declined to guide costs beyond Q2 but said 1-gamma DRAM and G9 NAND ramps are a cost tailwind, while new fab startup costs starting 2026–2027 are a relatively small impact at these margin levels.

Mark acknowledged the entire industry will be short to demand, said Micron has pulled in tools and accelerated Idaho construction, and noted the company is only able to meet 50-66% of several key customers' medium-term demand.

Mark confirmed the ~$18B net CapEx framework, said the vast majority will be DRAM (node transitions plus greenfield construction), and declined to give gross vs net split for fiscal 2026.

Sanjay declined to give F2Q guidance but framed the demand broadening across AI servers, traditional servers, smartphones and PCs alongside tight supply as a healthy setup for calendar 2026.

Sanjay said Micron is already at a $6B+ HBM run rate and may reach industry-share parity earlier than year-end, while deferring specific 2026 share guidance.

Mark explained that structurally lower NAND capacity shifts more underutilization into inventory rather than period costs, and flagged rising Idaho start-up costs through 2026.

Jensen walked through the four AI lifecycle phases and the rising revenue opportunity per gigawatt, positioning Groq for high-interactivity workloads.

Jensen highlighted growing inference share via Anthropic partnership expanding from near-zero coverage, every frontier model company expected to adopt Vera Rubin from launch.

Jensen emphasized Grace Blackwell and Rubin's use of giant reticle-limited dies and architecture compatibility, and said Grok will extend NVIDIA's architecture similar to Mellanox.

Jensen framed supply planning as well underway and reiterated the three platform-shift thesis (accelerated computing, generative AI, agentic AI) driving compounding demand.

Jensen framed the growth runway as a $3-4T AI infrastructure opportunity over five years, citing doubled hyperscaler CapEx to ~$600B and agentic AI driving orders-of-magnitude compute needs.

Jensen said orders have grown since GTC, referenced 800 AI factories in planning, and framed intelligence as essential infrastructure like electricity, asserting there will be many more announcements ahead.

Jensen walked through pretrain, post-train and test-time compute as three scaling laws, said test-time compute is already 100x one-shot and could reach millions of times more, and argued Blackwell's fungible architecture is why the industry is consolidating on it.

Jensen directly addressed the scaling debate, explaining three scaling methods (pre-training, post-training, inference time) are all active, driving demand across the stack.

Dave explained upside exabytes are sold above contract price as yields improve; Gianluca said the LTA renegotiation trend is unchanged and pricing keeps pushing higher.

Mosley confirmed aerial density remains the primary focus. Acknowledged performance tier conversations happening and prior stacked actuator designs could be revisited, but highest capacity per drive is top priority.

Mosley suggested flat to slightly up pricing per exabyte is possible given strong demand. Romano confirmed they are executing better than the 50% incremental margin model from the Analyst Day.

Mosley pushed back firmly on SSD cannibalization, saying architectures are well-set for years. Romano noted the gap between supply and demand is actually getting bigger each quarter.

Mosley explained revenue growth is dictated by BTO commitments set 6-9 months prior and capacity additions come via technology transitions. Romano confirmed the low to mid-teens growth trajectory is on track.

Mosley noted BTO visibility extends into calendar 2026, and pricing adjustments are factored into forward agreements as Seagate continues product transitions.

Mosley noted the supply hole won't fully transfer to June but demand remains strong. Highlighted video as a major demand driver across cloud and edge with AI creation tools expanding content.

Asked about nearline demand visibility and March seasonality; Dave said mass-capacity visibility extends ~nine months on the build-to-order model, and Gianluca noted March seasonality in legacy and VIA.

Roger declined to quantify a target but said disciplined R&D and SG&A management should drive improving operating leverage in coming years.

Roger said customer concern has intensified with no line of sight on the 232 review, full-battle-mode tariff daily updates, and customers delaying confirmation of demand, which is why 2026 has become less certain.

Roger pushed back on the 50% framing, said the original was ~40% growth, and explained the step-down by the shift of tool-productivity revenue from system sales into upgrade/installed-base plus the 3800-heavy tool mix delivering the same capacity with fewer units.

Christophe said single-expose conversion will happen gradually as productivity and cost-of-technology gains compound, particularly visible in DRAM given its higher multi-patterning baseline.

Christophe said committed AI investments create a high bar for customers to change behavior and reaffirmed 2025 and 2026 as growth years, but said 2027 visibility is still limited.

Christophe pointed back to the Investor Day framing, AI chips require more advanced nodes, driving acceleration of Moore's Law and historically more advanced-litho exposures, but declined to share specific layer-count data by node.

Christophe said most of the EUV pushout risk is already reflected in the guide and that the high-end scenario is tied to securing additional clean-room space at customer fabs, an active conversation.

Hock reaffirmed the focused chip model, XPUs plus clustering/networking components, and the new XPU platform vehicle to fund frontier-lab compute.

Hock said workload-specific XPUs (prefill/decode/MoE) are diverging from GPU designs, with LLM customers building two parallel chip tracks.

Hock said mature customers are now designing parallel training + inference chips, with further variations for prefill vs decode, providing substantial per-customer content growth.

Kirsten confirmed XPU and wireless lower-margin mix partially offset by rising software revenue; no detail on semiconductor margin trajectory.

Hock said there's no inference vs training differentiation in custom vs merchant; value is in end-to-end co-optimization of hardware and software over multiple generations.

Hock said Broadcom provides base semiconductor technology and optimizes ~5 degrees of freedom (compute, bandwidth, memory capacity/bandwidth, latency, power) but customers own their models and algorithms.

Hock clarified the $60-90B SAM covers only three hyperscale customers and is a Serviceable Addressable Market, with AI connectivity at approximately 15-20% of dollar content.

Goeckeler tied the mid-80s margin guide to prioritizing durability and fair returns over maximizing a single quarter, with Visoso adding prudent component-cost assumptions.

Visoso pointed to the $6B buyback announcement but did not specify target cash balance, saying they will update as execution proceeds.

Goeckeler said data center growth keeps accelerating, mid-to-high-teens bit growth achievable through nodal transitions with declining CapEx as percent of revenue.

CEO highlighted data center exabyte demand forecasts have jumped from mid-20s to mid-40s to high 60% growth in successive quarters, almost all AI-driven.

CFO confirmed debt reduced from $2 billion to $600 million, with priorities to continue investing in business and building prudent cash reserves before discussing shareholder returns.

Management confirmed plans to grow in line with market at mid-to-high-teens bit growth, with upside from BiCS8 productivity gains enabling more bits without additional wafers.

Management confirmed multi-quarter deals emerging and customers providing demand visibility into 2027, signaling structural shift from quarterly negotiations.

Bill cited channel refill opportunity and ongoing cost efficiencies as near-term gross margin levers beyond utilization improvement.

Kurt confirmed all accelerated growth drivers on track to Investor Day targets. Auto revenue guided for Q3 is only 4% below the Q4 2023 peak.

Kurt framed all three acquisitions as offensive moves to enhance differentiation in compute and SDV, not just defensive against China competition.

Kurt highlighted Q2 auto flat YoY as a turning point after five quarters of decline, driven by stabilizing direct orders and slower Western Tier 1 inventory digestion.

Kurt provided soft Q2 guidance of flat to slightly up but heavily caveated the limited visibility. Acknowledged Q1 is on the lower bound of seasonal.

Bill confirmed gross margins would stay at current levels plus or minus 50bps until revenue growth returns, with utilization in low 70s providing upside lever.

Cristiano clarified QCOM doesn't buy memory directly. Qualified with every memory provider including CXMT. Multi-generation memory controller provides flexibility. Full fiscal year handset market sized by DRAM availability.

Akash confirmed premium tier mix shift is a margin benefit. Higher-value chips in constrained environment should support margin mix.

Akash confirmed no change to Apple share assumptions vs prior guidance.

Akash attributed margin decline to incremental data center investment, which is additive to the normal OpEx reallocation framework used in prior years.

Palkhiwala pointed to consumer strength from new XR/PC launches and said the industrial market is at an inflection point with rising demand for AI-ready solutions.

Palkhiwala reiterated the known planning assumption of the large customer's share ramping to 20% for the 2026 launch with the agreement ending after, and said there was no new information.

Rafael confirmed flat loadings for Q3, inventory growing at a slower rate, and reiterated 75-85% long-term fall-through framework.

Rafael detailed the tax law provisions (ITC increase, R&D expensing, permanent FDII) and noted significantly lower cash tax rates expected for several years, though not yet reflected in financials.

Rafael declined to forecast Q2 specifically but signaled inventory levels should hold at $4.5B, implying loadings would adjust with revenue rather than drain inventory.

Rafael provided the 75-85% fall-through framework plus depreciation guidance and explained factory loadings impact. Comprehensive long-term model guidance.

Vince quantified data center at $1B run rate (grew 50% in FY25) and ATE at $800M run rate (grew 40%), both expected to see double-digit growth over the next few years. Vertical power at the knee of the curve.

Vince ranked recovery by segment: industrial led by healthy inventories and AI/defense, then consumer, comms (wireline), and automotive last. Thorough end-market outlook.

Murphy agreed with NVIDIA's fungibility thesis, saying Marvell's ability to quickly execute custom and semi-custom designs amid supply, power, and architectural constraints is a key advantage, with the shift from training to inference monetization driving broad pickup across custom, optics, and memory.

Confirmed vast majority of custom silicon is AI for both this year and next. Other programs have come in but AI magnitude has outpaced. Declined to break out non-AI/AI growth rates but noted custom will grow faster due to lower base and ongoing ramp.