A：The forecast for Q2 2026 consolidated revenue is approximately $22 billion, representing 47% year-on-year growth.

A：The projected revenue growth for the AI semiconductor business in Q2 is an acceleration to 140% year on year, with revenue reaching $10.7 billion.

A：The custom AI accelerator roadmap is alive and well, with shipping activities taking place now, and plans to scale to multiple E1s in the future.

A：The revenue forecast for non-AI semiconductors in Q2 is approximately $4.1 billion, which is up 14% from a year ago.

A：The growth in infrastructure software for Q1 was 13% year on year, with bookings continuing to be strong. For Q2, the focus is on infrastructure software revenue to be approximately $7.2 billion, which is 9% year on year.

A：Consolidated revenue for Q1 was $19.3 billion, marking a 29% increase from the previous year.

A：Revenue for the semiconductor solutions segment in Q1 was $12.5 billion, with a gross margin up 100 basis points year on year to approximately 68%.

A：Free cash flow in the quarter was $8 billion, representing 41% of revenue.

A：The Board of Directors has authorized an additional $5 billion for the share repurchase program, effective through the end of calendar year 2026.

A：The non GAAP tax rate for Q2 in fiscal year 2026 is expected to be approximately 16.5%, primarily due to the impact of the global minimum tax and the geographic mix of income.

A：AI chips are critical to creating and productizing platforms across various sectors including enterprise, cloud services, and consumer subscriptions. This demand is driven by the need for constant compute capacity for training and inference to productize these platforms, with many customers, including hyperscalers, creating their own custom accelerators.

A：The company's current generation solutions outperform competitor offerings in terms of performance and complexity. They are 12 to 18 months ahead in terms of chip design complexity, packaging complexity, and IP.

A：Competitive AI chip development requires a best-in-class silicon design team, advanced packaging, and expertise in networking clusters. The best technology IP and execution are essential, with Nvidia being a formidable competitor. To compete effectively, companies need to produce chips that are not only good enough but better than their competitors, including Nvidia. The ideal partner would possess this best technology and execute at the highest level.

A：High volume production is critical in the development of AI chips to ensure a quick time to market. The ability to produce 100,000 chips or more at yields that are cost-effective is crucial, and very few players in the industry can achieve this. The company in question claims to have significant experience in doing just that, which is a key competitive advantage.

A：The demand for networking components in AI revenue is driven by the new generation of GPUs and their capabilities, such as running 200 Gb and 400 Gb bandwidth. The Tomahawk 6, introduced a few months prior to the speech, is the only solution that offers this level of bandwidth for hyper-scalers. Additionally, the demand is bolstered by the need for high-bandwidth scaling-out optical transceivers. This combination is driving the growth of networking components within the company, and they expect continued momentum with the launch of the next generation in 2027.

A：The architecture of AI accelerators in GPUs is evolving to become more specialized for particular workloads. General-purpose GPUs are being supplemented with more customized designs, such as XPs, which are tailored for specific tasks like training and inference. These specialized designs are increasingly effective for workloads that do not fit the traditional matrix multiplication focus of GPUs, like mixture of experts. This trend is expected to continue, making XPs a more preferred choice for workload-specific performance optimization.

A：The anticipated impact of the company's product mix on its overall business is not considered substantial by the speaker, indicating that there is no need for worry regarding this matter.

A：The speaker's company plans to substantially exceed a $100 billion revenue goal next year by shipping approximately 10 GW of products, which would equate to content per gigawatt in the $20 billion range, as indicated by the speaker's calculation and expectations.

A：Looking at gigawatts is significant when assessing the company's chip sales because that's how the chips are sold. According to the speaker, the dollars per gigawatt can vary dramatically depending on the customer, but it's important to note that this metric is not far from the dollar amounts being discussed.

A：The company ensures supply visibility for the next few years by anticipating significant growth, locking up key components like T class and tea class, and having deep strategic partnerships with customers and suppliers for the necessary elements. They invest with their partners to develop not just more capacity but also the appropriate technology for the future demand.

A：The distinction between the revenue from chips versus the revenue from the racks in the company's projects is not directly answered. However, it is implied that the company may not differentiate between the two revenue sources, suggesting that the entirety of the revenue for the projects, including Anthropic's racks and chips, is not explicitly delineated in the provided information.

A：The company manages visibility and market share with multiple fragmented customers by focusing on a small number of strategic customers with significant revenue generation. They have a clear understanding of each customer's strategic plan and their needs for custom silicon, which allows for strong visibility into market share across a fragmented set of customers using multiple cloud service providers.

A：The preferred method for connecting GPUs in scale up scenarios is direct attached copper, as it offers the lowest latency, power, and cost.

A：The speaker believes Ethernet is the de facto standard for scale out in the cloud and the right choice for scale up as well. This is supported by engagements with partners and the industry, indicating that Ethernet is the preferred protocol for achieving necessary latency and scale requirements.

A：The anticipated trend for customers using the speaker's custom xpu technology is that they will develop two chips each year, one for training and one for inference, as they progress towards complete xpu. This is because there is a need to invest simultaneously in training to achieve higher levels of intelligence and in inference for product testing and deployment.

A：The speaker indicates that visibility and strategic partnership with the customers have improved over time. They have gained better insight into the customer's plans and long-term strategic deployment of the XPS they develop together. The speaker emphasizes that the investment these customers are making is long-term and that the partnership goes beyond transactional to be part of the strategic roadmap.