A：Sandisk presented forward-looking statements regarding their technology and product roadmaps, new business models and MSA customer partnerships, business plans and performance, market trends and opportunities, and their future financial results.

A：The key factors in maintaining earnings power, as per Sandisk's management, include the conviction in the franchise's earnings power, the successful restructuring of the business, and the ongoing investment in technology and business model innovation.

A：Sandisk has made strategic business moves such as focusing on the data center as a major growth pillar, enhancing the consumer business, and restructuring the business for better focus and execution.

A：Sandisk ensures its technological leadership by continuously investing in many generations of NAND technology, maintaining a strong partnership with Kioxia for joint investment in R&D, and working on multiple technology advancements such as Bix 9 and Bix 13, as well as systems capabilities including AI for the data center.

A：The significance of the multi-horizon innovation investment plan in technology investment is to ensure that Sandisk has the right technology prepared for the current time, the near future, and well into the next decade, allowing for a consistent and sustainable investment strategy across various technological horizons.

A：The extensive product portfolio and market presence of Sandisk imply a strong future position across various markets including consumer, edge, and data center. This broad optionality gives Sandisk the potential to adapt and succeed in a variety of technological and market changes, enhancing their conviction in the company's long-term success.

A：The business strategy involves investing in technology to ensure it remains relevant and drives innovation across various markets.

A：The motivation behind the high bandwidth flash strategy was the recognition that AI inference would be a major opportunity, with the focus mainly on model training at that time. The team saw the need for a different memory architecture for inference scaling and believed they had a crucial technology to offer.

A：The focus of the business 18 months ago was on developing high bandwidth flash, which was announced as part of their strategy.

A：As of the last update, high bandwidth flash had received significant attention, being named the most innovative technology at the previous FMS event, and an ecosystem was being developed for it.

A：The next focus is on investing in the business to systematically derisk it, which includes expanding their joint venture and ensuring access to production capacity and DRAM, a key component for data center operations.

A：The business plans to achieve sustained value creation by increasing profitability, reducing cycles, and consistently growing revenue across all time horizons.

A：The imperatives for sustained value creation are increasing profitability, reducing cycles, and consistently growing revenue.

A：The business strategy involves balancing various strategic priorities across all time horizons to achieve sustainable value creation, considering the impact across short-term, mid-term, and long-term perspectives.

A：The launch of Bix 13 is projected to be beyond 2030, with hundreds of engineers working on technology a decade into the future.

A：Market share makes a big difference in how much a large technology company can invest in R&D; when a company gets bigger, it should see improvements in its investment in R&D that are commensurate with its market share.

A：The company has its own fabrication facilities (fabs) with large-scale operations, performing all front-end manufacturing for semiconductors, and it also has teams working on the back-end manufacturing and packaging of chips into wafers and SSDs for various applications.

A：The speaker's comments about the high profitability of their business are considered surprising because the business practices in the industry are such that there is minimal profitability leakage across the model, suggesting that the high margins have been hidden and not widely recognized.

A：The company's investment in technology and manufacturing expertise, including a huge amount of engineering and billions of dollars in building out manufacturing capacity over 25 years, has contributed to its profitability by creating a 'coiled spring' of investment poised to produce significant results.

A：The speaker believes that the market is adjusting to these dynamics quite quickly, transitioning from a quarterly to multi-year engagement with customers. They emphasize the importance of strategic engagement with customers, which allows revealing the profitability of the franchise.

A：The speaker mentions that to unlock profitability, proactive supply management was necessary to manage the business from the supply side, rather than releasing new technology solely based on technological readiness, which could lead to poor economic outcomes.

A：The consumer business contributes to the company's overall financial health by providing global brand equity and a broad market base that acts as a shock absorber during industry down cycles, insulating the business from profitability impacts and allowing for consistent profit generation.

A：The consumer business, while important and profitable, is not considered sufficient for the company's needs because it is not large enough to insulate the whole business from market fluctuations, hence the need for growth and a brand-first approach.

A：The company is exploring new business models that involve strategic, long-term agreements with customers, renegotiating terms and conditions, and enhancing financial visibility. These strategies aim to provide long-term customer visibility, demand visibility, and pricing stability.

A：There has been a significant change in customer behavior, with customers proposing and entering into multi-year agreements for better supply visibility and predictability. The company has seen an increase in customers willing to commit to longer-term deals and is now facing requests from customers for three to five-year agreements.

A：The strategy for staying engaged in the NAND market involves making sure the company has a supply available to play in various markets, including those with customers who want to continue doing business as they did in the past. This ensures the company remains relevant and continues to engage with customers on their terms, maintaining a presence across the entire market.

A：It is important to have a clear relationship with the customer base in the NAND business because the company has an exclusive set of customers, including the who's who of technology companies worldwide. Engaging with these customers is vital for maintaining and strengthening the business relationship.

A：The rationale behind the company's strategy to change relationships in the NAND market is to bridge the gap between the three-month visibility supply side and the ten-year investments in fabs, which previously led to dissatisfaction from both sides. The goal is to reach a more balanced and sustainable situation where everyone involved can achieve their requirements.

A：The company plans to achieve and sustain growth in the NAND market by committing to mid to high teens volume growth and expanding its supply and volume accordingly. This growth strategy is seen as sustainable over long periods and aligns with a broader fabric strategy that ensures a planned approach to development and production.

A：Innovation plays a central role in the company's growth strategy for the NAND market, with a focus on developing scalable technology and increasing productivity. Through consistent innovation, the company has been able to increase output per wafer by 54% with each new node, which translates to a yearly productivity growth CAGR of 27%. This underscores the company's commitment to investing in intellectual capital over financial capital to drive growth.

A：The company manages investment in technology to avoid over-saturating the market with supply by carefully planning the introduction of new nodes and not releasing them yearly. This strategy allows the company to maintain high productivity and avoid the risks associated with excessive supply, like price erosion. The focus is on the application of intellectual capital rather than financial capital for growth.

A：A detailed roadmap is crucial for the NAND business as it outlines the company's strategy for product development and technology scaling. It serves as a guide for future investments and innovations, ensuring the company remains competitive and can meet customer demands. The presentation by Alper will provide insights into the current and future directions of the company's NAND roadmap.

A：According to the company's recent earnings call, the market for NAND is expected to grow to $300 billion in the current year and $500 billion in the following year. The company provided this figure as a market outlook and it includes the impact of China, which can skew the numbers when trying to back into other regions' growth. This forecast highlights the company's confidence in the pricing and market dynamics of the NAND sector.

A：Understanding market dynamics is significant for creating value in the NAND business as it allows the company to make informed predictions about pricing, supply, and demand. Insights from market intelligence, such as the work done by Eric's team, enable the company to adjust its strategy and ensure it remains competitive in a rapidly changing playing field. This understanding helps unlock value creation by aligning supply with market demand and optimizing revenue and profitability.

A：The speaker indicates that NAND flash is the only semiconductor technology that can match the exponential growth of AI, which is driving the increased adoption of NAND in AI architectures.

A：The three main scaling methods mentioned are lateral scaling, logical scaling, and architectural scaling. The company is prioritizing the more capital-efficient methods of scaling over the financially more challenging vertical scaling.

A：The company's capital intensity is lower than the industry average. The data provided indicates that in 2025, the industry spent on average 2.66 times more capital than the company to generate the same output.

A：The company's product roadmap includes an emphasis on increasing both performance and power efficiency in every generation through device and design innovation.

A：The CBA technology, or hybrid bonding technology, allows the company to combine two different wafers to make a single wafer, which in turn can augment the performance and power efficiency of existing technology nodes. This enables the company to quickly upgrade its product portfolio to meet customer needs without extensive capital spending, providing operational flexibility and a competitive advantage.

A：The Bics 9 product leverages the company's CBA technology to deliver significant performance gains with minimal incremental capital spending. It is based on specifications from large hyperscale customers who required high performance and rapid delivery. Bics 9 is expected to power the next generation of storage SSDs.

A：The Big 10 product line features the 1 TB TLC die and the 2 TB QLC die, both of which have achieved substantial improvements over the previous Bics 8 technology. The 2 TB QLC die, in particular, is the highest density memory chip in the world, with more than 60% density improvements, over double the read and write bandwidths, and a 75% improvement in power intensity and efficiency.

A：The historical growth rate of bit capacity for the company's technology is 27% CAGR per wafer annually. The company's production plans are based on a long-term demand forecast of about a high teens growth rate, which means that the technology productivity is sufficient not only to meet production needs but also to exceed them by a significant margin.

A：The speaker expects the flash market to reach 1.2 zeta bytes of shipments in 2026, driven by growth in consumer and edge devices, the transition of PC storage needs to flash, and the increasing demand from AI data centers. The revenue from the flash market is expected to grow to over $300 billion in 2026 and nearly $500 billion in 2027.

A：The industry's supply and capacity were influenced by a targeted over 30% annual growth rate, a peak in wafer capacity at 800,000 wafers per month in 2022, and a dramatic reduction in demand due to Covid-related inventory digestion. This led to structural capacity reset below peak levels, yet the industry still managed to achieve mid to high teens production growth via nodal migrations.

A：The projected growth of the Flash market by 2026 is over $300 billion and can approach half a trillion dollars. This growth is being driven by a shift in the supply side towards nodal migrations that meet bit growth targets, as well as a transition by Oems to produce more premium offerings and Asps to continue to grow.

A：The stages of the AI data cycle include: 1) Raw Data Archive where organizations store everything they collect, 2) Model Data Preparation where raw data is extracted, transformed, and loaded to make it machine-ready, 3) Model Training using massive GPU clusters to process and build a model, 4) Inference where trained models are used to answer questions and generate outputs, and 5) New Content Generation where model outputs are stored for ongoing consumption and can be fine-tuned over time.

A：Flash plays a critical role in the AI infrastructure, being proliferated through all layers and not just sitting at the edge. It is involved in various workloads such as fast data lakes, staging for training checkpointing, and ephemeral and persistent Kv Caches that are essential for efficient AI operation.

A：The infrastructure built for training, which focused on data storage, preparation, and presentation to GPUs, is not sufficient for inference. As the focus of AI data centers shifts to inference, there is an increasing need for longer context length, richer conversation, and more parallelism, which translates to a greater demand for working memory or the Kv cache.

A：The two emerging trends in inference when it comes to NAND flash are data augmentation, which increases the relevance and accuracy of responses by providing external data, and context remembrance, which involves the model maintaining a memory of the user's history to provide smarter and longer conversations.

A：Kv cache functions as working memory or a notebook in inference. It is used to store the context of previous tokens during the decode process, where one token is generated at a time. The context length can grow with longer conversations, making the entire AI process much more efficient. The Kv cache allows the model to remember past interactions and generate more relevant responses.

A：The common theme that emerges from using the ephemeral Kv cache in AI systems is the need for persistence in the memory hierarchy. This is because ephemeral caches like HBM and system DRAM have limited capacity, and as AI systems scale and the number of interactions grows into the billions, a persistent layer is required to provide extended capacity.

A：The persistent Kv cache supports AI systems by storing data that cannot be stored in the high bandwidth of DRAM and HBM. It acts as an additional layer in the memory hierarchy that provides durable storage for AI models and user contexts, enabling billions of interactions without having to rely solely on volatile, ephemeral caches.

A：Persistence is important in the memory hierarchy for AI systems because it allows for the retention of user context and session data even when power is lost. This is crucial for AI systems that require maintaining session state across interactions, whether it's for a few hours or indefinitely, to enable efficient and relevant responses.

A：An infrastructure planner should consider the number of sessions that will hit the infrastructure, how long the sessions should be retained, the cash base ratio based on existing infrastructure's Kv cash pools, and the amount of storage required per session, which is a function of session length and the size of the tokens used in the context.

A：The projected 2030 install base for persistent Kv cache is estimated to be one zettabyte. This is derived from considering variables such as session length, number of users, and growth in context length and average session size, suggesting a substantial opportunity for growth in the persistent Kv cache market.

A：Within an AI data center, flash is placed in various layers including data lakes (large capacity QLC drives), GPUs (high-performance TLC SSDs for active training data sets and checkpointing), network-attached persistent Kv cache (lukewarm storage), and close-coupled persistent Kv cache (warm and hot storage).

A：Integrating Flash into AI data centers results in reduced energy consumption, higher throughput, and the ability to store more data without requiring re computation, which is more energy-intensive. Flash acts as a 'token battery', storing intelligence that can be utilized later, and is essential for the efficient operation of AI systems.

A：The Sandisk AI Lab is an initiative started by the company to understand the bottlenecks in persistent Kv cache by running server and rack scale systems with AI workloads. The lab aims to provide insights by simulating real-world conditions to complement customer learnings and enhance product understanding.

A：Persistent Kv cache contributes to sustainable value creation by enabling efficient and relevant responses in AI systems through persistent storage of user contexts and session data. This efficiency and relevance lead to better performance and ultimately value for both customers and shareholders.

A：The market the company operates in is changing from an edge-centric market to a data center-centric market, which brings different dynamics.

A：The company reported a revenue of $20 billion, up 175% from the prior record, with gross margin improving every single quarter to 71.6% from 30.3% the year before. The final quarter closed with a gross margin of 84.6%, enabling an EPS of $39.25 compared to only 29 cents the year before.

A：The company generated $8.7 billion in free cash flow, excluding new business model prepayments, and improved its run rate to $20 billion, indicating robust business growth and the ability to capture market value.

A：New business models were introduced to strengthen relationships with strategic customers and to create a fast-growing, profitable, and less volatile business. These models are based around supply and demand uncertainty with multi-year agreements, growing at a fast pace, featuring fixed and variable pricing components, and having financial guarantees. They involve conversations at the highest levels of the companies, requiring board approval.

A：The new agreements are win-win conversations with deep technical and commercial integration, involving multi-year engagements that have seen high-level integration. Some customers have already expanded their contracts within a short period after initial签订.

A：The company projects to allocate a significant portion of its business to the new business models because they are fast-growing, attractive, and less volatile. The goal is to make these the predominant way of doing business for the company.

A：Pricing in the new business models is fixed in some contracts, while some include variable portions with floors and ceilings. Financial guarantees are provided by third-party financial institutions and customer deposits and credits. Financial guarantees hold a strong protection level, with the ratio between financial guarantee and remaining performance obligation (RPO) evolving over time to provide protection as a percentage of revenue to come.

A：Financial guarantees include third-party financial institutions providing financial guarantees and customer deposits and credits. Of the total guarantees, $16.5 billion has been secured, with $2.5 billion already in the bank. Financial guarantees will stay constant and low throughout most of the time.

A：The company plans to execute multi-year agreements (MVBs) with excellence, ensuring quality products on time and fulfilling their part of the bargain. They will also increase safety stocks and continue to evaluate new deals with a selective approach to ensure long-term value for products customers are willing to pay for.

A：The company expects revenue growth in the mid to high teens, non-GAAP gross margin around 80%, and non-GAAP operating margin at 70% going forward. This includes expecting to spend about 5% in Opex and not expecting significant contributions from other income and expense.

A：The company expects a capital intensity of mid single digits as a percentage of revenue and plans to continue investing in business operations, maintaining a strong balance sheet, and using excess cash for share buybacks.

A：The company plans to return excess cash to shareholders through its share buyback program, having already executed a $6 billion program and authorizing an additional $14 billion program. The board has authorized a total of $15.5 billion for share buybacks, with the belief that returning cash to shareholders is essential.

A：HBF stands for High Bandwidth Flash, which is a technology introduced by the company. It is not included in the revenue projections because the company is funding it out of its operating expenses (Opex) and capital expenditures (CapEx), not as a revenue-generating projection.

A：The Memory World Problem refers to the inability of DRAM to keep up with the demands of compute and AI due to scalability issues. Sandisk addresses this problem with two new technologies: 3D Matrix Memory and High Bandwidth Flash (HBF). Both technologies are highly scalable and are designed to solve the memory bandwidth issues.

A：3D Matrix Memory is a technology developed by Sandisk that started in 2017 and involves developing memory layers. By 2024, development moved to a 300 mm facility. The company has delivered development vehicles and demonstrated multi-gigabit level functional memory arrays. The technology is approaching product spec performance levels and is a long-term project with continuous progress updates.

A：High Bandwidth Flash (HBF) is a device that was introduced by Sandisk, offering the same read bandwidth as HBM but with 16 to 8 times the capacity. It was designed with AI inference in mind, leveraging mixture of experts and sparse models, and allows for long context lengths. This makes HBF significant for AI workloads due to its high capacity and performance.

A：In the memory-centric AI paradigm, large model sizes and long context lengths drive up memory capacities while mixture of expert type sparse models drive down compute needs. This shift results in an increased focus on memory requirements, which HBF is positioned to meet.

A：Workload simulations showed that a single HBF GPU could match the performance of 8 HBM GPUs in a multi-turn genic workload. This indicates that HBF can deliver the same token output with fewer GPUs, improving capital efficiency by up to 8 times and matching or surpassing the performance with half the number of GPUs.

A：HBF is envisioned to enable sophisticated AI models, including those with over 100 billion parameters, to run on edge devices, offering an AI experience that 'never forgets' and is always available to the user. This is expected to significantly change the way people experience AI in their daily lives.

A：The HBF for Edge is a second-generation HBF device being actively developed with multiple customers. Its purpose is to dramatically change how AI is run on edge devices, enabling the operation of sophisticated AI models on these devices and providing an AI experience that remembers everything about the user.

A：The strategy for HBF includes creating an open ecosystem with a partnership approach. Sandisk has announced partnerships with SK Hynix, and established a consortium under OCP with Google and 10 other members. Meta has recently joined this consortium. The company is building a technical advisory board with experts like Jim Keller and David Patterson to guide its direction and is committed to open standards and collaboration.

A：Jim and his team are pursuing a vision of AI integration that sees most of the data center being taken over by AI and the emergence of heterogeneous computing, with AI compute and general-purpose compute being built on fundamental principles. They are creating high-end AI processors and high-end RISC-V processors, with two businesses: one licensing IP for various projects such as autonomous driving, robotics, and server projects, and the other developing high-end server designs.

A：The scalable AI computer Galaxy is a high-end server designed for AI computing with a focus on scalability. Unlike traditional computing setups, Galaxy allows for scaling up to 20 servers with 5, 6, 800 Gb Ethernet ports per server, which can be interconnected in quad servers. This results in a system capable of running from a single chip to up to 36 servers. Galaxy's design incorporates flexibility in memory and processing capabilities, allowing for a wide variety of models to be run on the same hardware.

A：Current HBM-based models face limitations such as being constrained by the local size of DRAM and often lacking sufficient network bandwidth. To address these challenges, Jim's company has built a terabyte of RAM per Galaxy server, with 36 galaxies providing a total of 36 TB of DRAM. They also ensure high network bandwidth to serve memory flexibly across the machine, tackling the issues of memory scalability and network bandwidth.

A：Jim's company's processors are distinguished by their balance of SRAM, local DRAM, host DRAM, and flash memory. GPUs were initially built for graphics and require a significant amount of HBM memory, leading to an imbalance. However, their processor architecture has 200 MB of SRAM per chip, allowing for a better balance. The changing ratios of compute to memory components like Kv cache and prefill in AI architectures are important considerations when building machines for different tasks, and the company's architecture is adaptable to these shifts.

A：HBM (High Bandwidth Memory) is significant in AI because it brings high bandwidth to play in high-bandwidth systems, overcoming limitations of Flash which lacked sufficient bandwidth. HBM enables the trade-off between compute and memory to be balanced in new ways, as it is effective in saving computation results in large format, which was not feasible with Flash due to its lower capacity. This leads to more efficient processing and potentially radical changes in applications like robotics and autonomous driving.

A：HBM has significant potential applications in edge computing and robotics, where it can transform operations by providing on-device processing with augmentation of Kv cache for everything seen in Flash. This will lead to more efficient autonomous systems that can understand their environment and surroundings, which can be revolutionary for applications like robotics and autonomous driving. The reduced compute power consumption compared to traditional computing makes HBM a suitable choice for edge devices where power efficiency is crucial.

A：The company has taped out the first HBM memory die, marking a significant milestone towards the market introduction of HBM. Although the full product will not be available immediately, customers can expect to see samples of inference devices with HBM next year. The company's approach to ensuring customer availability involves a co-development strategy with some of the largest customers in the world, which is anticipated to provide valuable insights into the technology's future from both a market and financial perspective.

A：Rapid market changes are making it difficult to understand what product to build and how much to build. There is a premium on staying very close to customers because the market is changing at an incredible pace and will continue to iterate.

A：The scaling of inference is incredible and represents the most significant technology transition in the speaker's career. There is a constant focus on driving requirements down, using less power and space, and increasing efficiency to scale faster and more economically.

A：Staying close to customers is crucial because it provides insights into product development and architectural choices. This knowledge helps in understanding the right products to build and where to allocate resources for success.

A：Customers provide companies with data that influence product development. Insights gained from customer deployments, such as the use of HBM andDDR memory, can guide the development process and the scaling of inference technologies.

A：The experiments showed that for running complex workloads, having just HBM SSD was sufficient, with a minimal amount of DDR required. This was a revelation as it suggested that system DRAM capacity could potentially be reduced.

A：It is feasible to reduce system DDR requirements for inference, but the extent of reduction, whether it be by half, one, or eight, depends on various factors and is workload-dependent.

A：The company plans to align with customer needs, understanding demand and adjusting production accordingly. This approach involves a wait-and-see strategy regarding product launches and customer demand, followed by adjusting the production plan based on actual usage.

A：Customization in HBM aligns with potential advanced packaging production. The company is prepared for the compute-to-memory shift and is confident in their capabilities to perform compute next to or within the memory. This transition will be dictated by customer discussions around performance preferences. The impact on CapEx will depend on the successful integration of these capabilities into the market and the potential for higher production volumes in future years.

A：The company's strategy is to optimize the percentage of NBM dedicated to certain applications over time. They will continue learning from data and customer feedback to define the optimal number, which is currently estimated based on existing data. The goal is to create maximum shareholder value on a sustainable basis, and the specific percentage for future years will evolve as more data becomes available.

A：New technologies are expected to provide continuous progress, with high endurance metrics, and learning curves that apply to core technology and HBAs, which allows for phenomenal endurance numbers, as seen in the warm cache technology. The endurance is workload-dependent, with higher read rates being necessary, and the ability to offer a much higher drive per day compared to conventional SFDs.

A：The balance between data center and edge computing is expected to change dramatically, with customers increasingly seeking deals that favor edge computing. The company wants to keep all parts of its portfolio robust, including data center operations, and ensure that engineers remain busy building products across all platforms. This shift is expected to continue rebalancing the company's focus and priorities.

A：The company is focused on engaging with customers on longer durations through various contract types of different durations, which makes it challenging to predict future segments. There's an emphasis on flexibility in the system to maintain a balanced approach. They are judicious in their business approach, following customer needs, and aiming for better decision-making with increased visibility. The company intends to ensure the robustness of all segments - Edge, Consumer, and Data Center - and is actively involved in structuring potential engagements with customers to ensure alignment and value creation over various time horizons.

A：The introduction of HBM and HBF has made conventional DRAM capacity tight and is expected to continue doing so as these new technologies become more prevalent. HBF, specifically, is anticipated to be more capital intense than conventional DRAM and may eventually become even more so, as it utilizes different technologies and has higher performance capabilities. However, the company believes it has enough headroom to grow through innovation and maintain robust scaling capabilities.

A：The company views Nand as a very scalable technology and believes there is at least a decade of visibility on scaling capabilities. They recognize the importance of getting the economics right and the challenges associated with scaling due to visibility issues in the market. The key lies in better understanding demand and supply dynamics to avoid sudden shortages, as exemplified by the transition from 8 to 10 wafers showing 65% growth. The company is confident in their ability to scale significantly, potentially reaching 60-70% growth in a single wafer generation.

A：The company is on a path to inventing new business models, moving away from traditional contracts. They are engaging with customers on longer-term engagements ranging from three months to four years, seeking to understand true demand and create structures that benefit both parties. The focus is on co-development and helping customers understand what's possible. The objective is to provide a completely hand-crafted structure that aligns with the customer's needs and the company's capabilities.

A：Translating RD productivity numbers into a semiconductor fabrication plan involves years of detailed planning. This includes monthly and yearly planning for which nodes are running where, in what fabrication facility, and what the yield is, all of which contribute to the output.

A：The significance of scalability in the semiconductor business is that it enables the technology to be produced efficiently and on a large scale, leading to cost-effective solutions. The semiconductor business benefits from 19 generations of innovation, and there is still a long roadmap ahead. This scalability also attracts market attention and can lead to increased usage of the technology.

A：The potential risks of scalability in the semiconductor business include the possibility of overproduction if the market forecast is wrong. The speaker highlights that the idea of making 10-year investments and conducting quarterly auctions to sell products does not make sense. The challenge is to find a balance between having enough capacity to meet market needs without overproducing.

A：The speaker believes continuous improvement and innovation are essential because they drive efficiency and productivity. By focusing on these areas, a company can align its investment horizons and create a positive future. Waiting for the past to reoccur is seen as a waste of time; instead, there is an opportunity to invent new solutions that work for everyone involved.

A：The semiconductor industry's approach to cost reductions has evolved from focusing on frequent cost down announcements to a more internalized approach. The speaker notes that the company's job is to drive efficiency, and cost reductions are part of the company's productivity. The cost of products is determined by the market, not by the internal costs of production.

A：The speaker believes that the forecast for the semiconductor business is better now than it was during the hyperscalers' demand forecast during COVID-19. He argues that being skeptical is easy, but it is necessary to have confidence in the ability to change outcomes and pursue different methods to achieve success. Despite potential challenges, there is an emphasis on moving forward with strategic changes and investments.

A：The speaker thinks skepticism can be beneficial in business because it can lead to a deeper examination of risks and uncertainties, which can also indicate where the biggest returns and opportunities lie. The speaker acknowledges the value of skepticism but stresses the importance of having the confidence to challenge the status quo and pursue meaningful change that can yield different outcomes.

A：The semiconductor business's interaction with customers has evolved from a purely transactional relationship to strategic partnerships. The speaker mentions that conversations with CEOs and CFOs have become more frequent and significant, and while there are no guarantees, the company works hard to make these interactions successful. The business is committed to the physical AI market and sees a growing potential there.