A：In the third quarter of fiscal year 2026, Broadcom reported revenue, operating income, and free cash flow that exceeded prior records. The revenue was $29.6 billion, up 86% year on year. Operating income grew even faster with a year-over-year increase and a record operating margin of 68% revenue, reflecting strong operating leverage.

A：AI semiconductor revenue grew 220% year over year.

A：During the quarter, Broadcom delivered Ironwood TPU v7 to both Entropic and Google and began production shipments of the Next generation TPU version 8 for Google. The new TPU generation has more memory and bandwidth compared to Ironwood and is optimized for inference workloads, performing comparably to or surpassing the performance of the Vera Rubin GPU.

A：For the fourth quarter, Broadcom expects AI revenue to grow to $21.7 billion, which is up 2236% year over year based on preliminary numbers. The guidance for Q4 is expected to result in an increase of 56% year over year, with AI revenue reaching $18.6 billion for the year.

A：The relationship between Broadcom and Google in terms of TPU development is strong, with a long-term agreement to develop and supply future generations of TPU and AI networking. The partnership is expected to fulfill growing demand with successive generations of complex CPUs developed, with plans to deliver multi tens of billions of dollars of GPUs annually over the next several years.

A：Entropic is expected to deploy another script gigawatts of TPU version AI in 2027, and an incremental 10 GW in 2028. OpenAI is on track for a planned deployment of 1.3 GW in 2027 and is in development for the next generation xpu beyond Jalapeno, aiming for over Ed gigawatts of TPU deployment. OpenAI also has a line of sight to become Broadcom's second-largest xpu customer after surpassing Google in 2027 and sustaining that position through 2028.

A：OpenAI and Broadcom are deep in development of the next generation xpu beyond Jalapeno, which is approaching kbu in Ed. OpenAI is projected to deploy over Ed gigawatts of this xpu and its successor. The new generation outperforms the Grace Blackwell Altra in performance, latency, throughput, and power, and is comparable to very Gpus in running OpenAI workloads, demonstrating the ability to operate at half the cost of a GPU.

A：Broadcom is extending its leadership in AI networking with the introduction of Tomahawk 7, the industry's first 200 TB per second Ethernet switch for scaling, continuing to lead in every generation of PCI Express switching, and expanding its capacity and market position as a leader in leading edge optical DPs and CW lasers for optical interconnects.

A：AI semiconductor revenue is expected to grow to approximately $230 billion in 2028. The extremely strong demand and the company's ability to secure supply for this outlook are the primary reasons for the growth trajectory.

A：For Q4, the projected consolidated revenue is $34.8 billion, an increase of 93% year on year. The forecasted Q4 Ei revenue is $21.7 billion, with an operating margin expected to be approximately equal to the revenue figure.

A：Operating income increased 92% from a year ago to a record $20.1 billion, with non-GAAP EPS up 96% year on year to $3.32 cents.

A：Revenue for the semiconductor solutions segment was a record $20.8 billion, up 127% year on year, and the gross margin was approximately 76%.

A：Revenue for infrastructure software was $8.8 billion, up 29% year on year, with a gross margin of 94% for the quarter.

A：Capital expenditures in the quarter were $532 million, and the company paid down $5.6 billion of long-term debt. Subsequent to the quarter, an additional $1.5 billion of senior notes were paid down. The weighted average coupon rate and years to maturity of the company's gross principal fixed-rate debt are 4% and 7.4 years, respectively.

A：The company's strategy includes empowering leading AI labs with cost-effective and sustainable growth solutions. This aligns with their existing capital allocation framework and is facilitated by strategic financing. Through the AI xpv platform, the company partners with financial partners to underwrite and capitalize AI infrastructure assets, potentially providing modest residual value guarantees.

A：The guidance for Q4 is consolidated revenue of $34.8 billion, up 93% year on year. The forecasted AI revenue is to double again to $115 billion in fiscal 2027 and then double again in fiscal 2028 to $230 billion. This guidance is provided to indicate the trajectory of growth without the intention to update it on a quarterly basis.

A：The timely deployment of chips is important as it aligns with customer expectations and is a critical factor considered when providing an outlook on shipping more chips. It affects customer satisfaction and the company's reputation for reliable delivery.

A：The projected demand for AI semiconductors is $115 billion in fiscal year 2027 and an additional $230 billion in 2028, totaling $350 billion. The company believes with a high degree of confidence that they will ship $350 billion of AI semiconductors to customers in the next two fiscal years.

A：The company's supply bottlenecks are being addressed by starting to deploy a Singapore fabrication facility for substrates starting in fiscal 27. This new capacity in Singapore is expected to help in resolving the existing supply bottlenecks.

A：The Google number provided refers to the capacity for in-procurement and entropic components, not a singular figure. The 10 GW figure mentioned pertains only to inproce and entropic components, not the overall capacity.

A：The Tomahawk switching family has had a phenomenal success, particularly with the Tomahawk 6 ramp, which has been the fastest of any switching family. The company is almost sold out for Tomahawk for the next year. Tomahawk 7 with 400 GB GPUs is also being utilized by Base XP customers. Additionally, the adoption of the Tomahawk Ultra scale-up platform using low latency Ethernet is increasing and expected to be deployed starting this quarter and in fiscal year 27.

A：Revenue per gigawatt can be evaluated by considering the outlined gigawatt capacity and the projected deployment and revenue generation over the next two fiscal years. It is mentioned that the company does not specify the number of gigawatts that will come into actual deployment within the next few years. They only outline the journey into deploying XPs and provide guidance on future revenue expectations based on the deployment of these chips in data centers.

A：The actual number of gigawatts expected to be deployed within the outlined period is not explicitly stated. However, it is mentioned that the company outlines the journey into deploying XPs with the intention that over time, the deployed gigawatts will reflect the demand forecasted. The company's revenue forecast is based on the belief that they will ship $350 billion of AI semiconductors to customers in the next two fiscal years, which encompasses the period mentioned.

A：The maximum off-balance sheet risk for the backstop agreements was not specifically quantified in the transcript but was mentioned to be about 29 billion for the first tranche. The speaker did not provide an order of magnitude for the maximum exposure for each of Anthropic and OpenAI's gigawatts over the next few years.

A：The speaker indicated that there was nothing new to add regarding residual value guarantees or backstop agreements, as they do not have anything to announce in this regard. The numbers mentioned previously around these topics remain unchanged.

A：The impact of the XP mix and rise in number cost on gross margins in 2027 and 2028 is not directly quantified in the transcript. However, the speaker suggests that gross margin will be affected by the revenue mix between semiconductors and infrastructure software and the product mix within semiconductors, which includes increasing memory content.

A：The speaker did not directly address the reason for Google's supply constraints impacting their ability to provide silicon. However, they did mention that they create financial vehicles for startups and ensure that these companies, which depend on suppliers like Google, are financially secure and stable to fund themselves. Additionally, the speaker suggests that they have ample supply of technology and are happy to help these companies succeed.

A：The speaker indicates that hyperscalers like the ones mentioned are planning to create their own first-party compute capacity and develop silicon that is cost-performance optimized. They eventually aim to operate their own data centers and offer AI generative API access models to the world. The speaker also notes that this transition is already happening and that they are in the process of enabling this.

A：The speaker expects that as the product mix changes with an increase in xPU revenue, it will impact the gross margin. However, they emphasize the importance of focusing on operating margin, as revenue growth far exceeds operating spending, resulting in accretive operating leverage. They expect to sustain operating margin despite potential dilution from the product mix.

A：The attachment rate for scale up using Tomahawk Ultra in XPS deployments is not explicitly mentioned, but the speaker notes that customers who build XPS with them are deploying Tomahawk Ultra, in addition to Tomahawk Ridge and other Ethernet-based solutions. This deployment is observed in both XPs and GPU clusters.

A：The forecasted outlook is realistic and takes into consideration potential constraints in land, power, and shell. The company engages with customers on the availability of these resources before believing it will happen due to the long lead times for projects. They perform non-development analysis and reflect the forecast outlook considering these potential constraints.

A：Not all of the financing for the 10 GW expansion and the 5 GW for OpenAI in fiscal 2028 is expected to be through the xpv, and the specific needs will be evaluated case by case. The speaker mentions that not every site will be the same, and financing may or may not be provided based on whether it is within their framework.

A：Land power and shell are mentioned as significant constraints impacting the company's outlook heading into fiscal 2027. Furthermore, other supply chain constraints such as memory substrates and components are mentioned as potential challenges that could impact the company's outlook. These constraints are intertwined with the delivery and availability of AI data centers, and a collaborative effort with customers is necessary to address the issue.

A：The performance improvement and capacity expansion of xpu are expected to lead to an increase in revenue per day. With each generation of xpu, performance increases, leading to higher production costs and an increase in the price per unit. Despite this, the number of advanced xpu per gigawatt remains relatively stable, keeping dollars per gigawatt in the range of 20 to 30 billion. As there is an expected acceleration in the number of gigawatts to meet customer demand, the revenue per day is anticipated to rise in line with the content in dollars remaining stable.