The paragraph is from a conference call for Broadcom Inc.'s first-quarter fiscal year 2025 financial results. The call is introduced by Gu, the Head of Investor Relations, and includes company leaders Hock Tan, Kirsten Spears, and Charlie Kawwas. It details that Broadcom released a press statement and financial tables post-market, with additional information available on their website. Hock Tan and Kirsten Spears will discuss financial results, future guidance, and the business environment, while questions will follow their remarks. The call includes caution about risks with forward-looking statements, mentions both GAAP and non-GAAP financial reporting, and highlights that Q1 2025 revenue reached a record $14.9 billion, marking a 25% increase from the previous year.

The paragraph discusses the company's semiconductor business performance and outlines its strategic investments in AI technology. The semiconductor revenue reached $8.2 billion in Q1, driven by a 77% increase in AI revenue. The company maintains its AI revenue guidance and highlights investments in R&D for next-generation accelerators and expanded cluster capacities for hyperscale customers. A focus is on developing two-nanometer AI XPU packaging and scaling clusters to one million XPUs. The company is also preparing to deliver a new 100 terabit Tomahawk 6 switch to customers. These efforts align with the roadmap of three major hyperscale customers, projecting a serviceable addressable market of $60 to $90 billion by 2027. Additionally, the company is working with other hyperscalers on customized AI accelerators, with plans to tape out their XPUs within the year.

The paragraph discusses Broadcom's collaboration with hyperscalers, highlighting its strength in hardware while hyperscalers excel in software. Broadcom has secured new partnerships for developing custom accelerators to optimize large language models. The trend towards using XPUs is ongoing, with increasing demand anticipated by 2025. Broadcom's Q1 AI revenue was $4.1 billion, projected to rise to $4.4 billion in Q2, reflecting a 44% year-on-year growth. In contrast, non-AI semiconductor revenue saw a 9% sequential decline, although broadband experienced a recovery, and server storage is expected to grow in the coming quarters. Enterprise networking remains stable as customers manage existing inventory.

In the outlined financial performance for Q1 and Q2, wireless revenue experienced a sequential decline due to seasonality, remaining flat year on year, with similar expectations for Q2. Although industrial resales dropped significantly in Q1 and are anticipated to remain down in Q2, non-AI semiconductor revenue is expected to stay flat. Overall, Q2 total semiconductor revenue is projected to rise by 2% sequentially and 17% year on year to $8.4 billion. Infrastructure software revenue saw significant growth in Q1, influenced by deals delayed from Q4, as VMware now features in year-on-year comparisons. This growth is driven by a shift from perpetual licenses to subscriptions, with a focus on upselling VCF to virtualize entire data centers and create private cloud environments. Approximately 70% of the largest customers have adopted VCF. Future growth is anticipated with the adoption of the VMware Private AI Foundation, which will virtualize GPUs alongside CPUs to facilitate AI workload management on-prem.

The paragraph discusses the collaboration between VMware and NVIDIA, noting that VMware now has 39 enterprise customers for its Private AI Foundation. The demand from customers is attributed to an open ecosystem and efficient load balancing and automation capabilities, which reduce costs. VMware projects Q2 software revenue to be $6.5 billion, a 23% increase year over year, and total consolidated revenue to be $14.9 billion, a 19% increase. They expect Q2 adjusted EBITDA to be 66% of revenue. For Q1 financial performance, revenue was $14.9 billion, up 25% from the previous year, with a gross margin of 79.1%, exceeding expectations due to higher infrastructure software revenues and a favorable semiconductor revenue mix. Operating income reached $9.8 billion, a 44% increase, and adjusted EBITDA was $10.1 billion, or 68% of revenue. The semiconductor solutions segment generated $8.2 billion, or 55% of total revenue, with a 68% gross margin, up 70 basis points year over year.

Operating expenses increased by 3% to $890 million due to R&D investments, resulting in a semiconductor operating margin of 57%. Infrastructure software revenue rose by 47% to $6.7 billion, largely driven by VMware, achieving a gross margin of 92.5% and an operating margin of 76%, up from 59% the previous year. Free cash flow was $6 billion, impacted by expenses related to the VMware acquisition and taxes. Capital expenditures amounted to $100 million. Day sales outstanding decreased to 30 days, while inventory increased to $1.9 billion. The company ended the quarter with $9.3 billion in cash and reduced debt by a net $1.1 billion, resulting in $58.8 billion in fixed-rate debt with an average coupon rate of 3.8% and 7.3 years to maturity.

The company has $6 billion in floating-rate debt with a weighted average coupon rate of 5.4% and a maturity of 3.8 years, alongside $4 billion in commercial paper at an average 4.6% rate. They allocated capital by paying $2.8 billion in dividends and spending $2 billion to repurchase 8.7 million shares. For Q2 guidance, they anticipate $14.9 billion in consolidated revenue, with semiconductor revenue reaching $8.4 billion, AI revenue at $4.4 billion, non-AI semiconductor revenue at $4 billion, and infrastructure software revenue at $6.5 billion. They project an adjusted EBITDA of about 66% and a consolidated gross margin slightly down due to revenue mix. The non-GAAP tax rate for Q2 and fiscal year 2025 is expected to be around 14%. The Q&A session is set to start, with Ben Reitzes from Melius asking the first question.

Hock Tan discusses the custom silicon trend and emphasizes that the company is not the direct creator of XPUs but rather collaborates with hyperscaler partners to develop these chips and compute systems. He clarifies that these partners are working to build systems similar to those of the three existing major customers who have deployed at scale. Although these four new partners are not yet customers, they are aiming to run their own large Frontier models. The process of developing these chips is accelerated, taking around a year and a half, due to an established framework and methodology that has proven successful with the current customers.

The paragraph is part of a conversation between Harlan Sur, an analyst from JPMorgan, and Hock Tan, who seems to be an executive discussing strong quarterly results in the AI sector. Harlan Sur acknowledges the momentum in the AI business and inquires about the expectations for the second half of the fiscal year, particularly regarding AI programs leveraging three-nanometer technology. Hock Tan responds by expressing the challenge of predicting customer behavior but notes that the company is exceeding expectations in Q1, with encouraging signs for Q2, driven by improved networking shipments that support AI accelerators for hyperscalers.

In the article paragraph, Harlan Sur asks Hock Tan about the progress of the three-nanometer technology ramp in the second half of the fiscal year. Hock Tan prefers not to speculate on that. William Stein from Truist Securities then poses a question about potential disruptions in the industry due to tariffs and the Deepseek issue, which have caused decision-making challenges for some companies. He acknowledges Broadcom's growth, especially in AI, and asks whether these disruptions might lead to significant changes for the company. Hock Tan responds by acknowledging these issues but states that it is difficult to predict the impact of tariffs on the company, as their structure and effects are still uncertain.

The paragraph discusses the positive disruption in the semiconductor industry driven by generative AI, which is accelerating advancements in semiconductor technology, including process, packaging, and design for higher performance accelerators and networking functionality. It emphasizes the ongoing innovation occurring monthly, particularly with XPUs, as companies aim to optimize performance for various models and partners. Optimization involves considering multiple factors such as compute capacity, network bandwidth, memory capacity, and latency, as these elements are crucial in addressing the distributed computing challenges associated with training, post-training, and inference processes.

The paragraph discusses the current trends and challenges in the tech industry related to generative AI and hardware development. It highlights the disruption caused by the need to create advanced chips and hardware infrastructure to support AI workloads. There is an emerging trend of enterprises considering on-premises solutions for AI workloads due to data privacy and control concerns, instead of relying on public cloud services. This has led to upgrades in data centers and a focus on VMware Private AI Foundation. Additionally, it touches upon the uncertainty regarding tariffs and indicates that more clarity might emerge in three to six months. Following this discussion, the conversation transitions to Ross Seymore from Deutsche Bank, who intends to ask a question about the XPU side of things.

The paragraph discusses the approach to design wins and deployments in terms of product scaling and partner selection. Hock Tan explains that, unlike many peers, his company defines a design win as a product that is already in scale production and deployment, which can take 1.5 to 2 years from initial design to implementation. They are selective in choosing partners who require large volumes, particularly in framing and training large language models, and they do not consider small-scale production as true deployment. The focus is on long-term, large-volume relationships rather than short-term projects.

The paragraph consists of a discussion on ASIC business strategies and concerns about potential regulatory impacts on AI design wins or shipments. Ross Seymore and Stacy Rasgon ask questions about the company's customer base and any worries regarding new regulations or AI diffusion rules potentially affecting their AI design wins. Hock Tan acknowledges some level of concern due to geopolitical tensions but assures there are no direct worries affecting their current three customers. When probed about whether any customers are based in China, Hock Tan avoids confirming specifics. The conversation then shifts to Vivek Arya noting Hock Tan's emphasis on AI training workloads, while others perceive the AI market may lean towards inference workloads, especially with new reasoning models.

In the paragraph, Hock Tan discusses the market dynamics between training and inference in AI chipsets, addressing how they contribute to a combined total addressable market (TAM) of $60 to $90 billion. He clarifies that while both training and inference are essential, most revenue comes from training. On the topic of server infrastructure, Harsh Kumar from Piper Sandler inquires about the importance of Ethernet in large clusters and how customers decide between different switch ASICs. Tan explains that hyperscalers prioritize performance when scaling AI accelerators, whether XPU or GPU.

The paragraph discusses the focus on high performance in the development of hardware for hyperscalers, particularly in networking. The company leverages its decade-long experience in switching and routing to enhance its AI-capable technology, aiming for advancements in bandwidth from 800 gigabits per second to 3.2 terabits per second. They are rapidly developing new products like Tomahawk 6 and planning for future generations, primarily for a few large customers. This signifies a substantial investment in anticipation of large market opportunities. Harsh Kumar thanks Hock, and the conversation shifts to Timothy Arcuri, who inquires about the growth in XPU units projected from two million last year to seven million by 2027-2028.

In the paragraph, Timothy Arcuri asks Hock Tan if four new engagements will increase their current market of seven million units. Hock Tan clarifies that these four are engagement partners, not yet customers, and thus do not contribute to the served available market count, which currently only involves three customers. Following this, CJ Muse inquires about how the company is enhancing its portfolio for six mega scale frontier models and how it aids partners in achieving high performance like exaflops per second per dollar of capitalization per watt, while respecting confidentiality where partners want to differentiate. Hock Tan responds that they provide basic semiconductor technology, which partners can optimize for their specific needs.

The paragraph discusses the complexities of optimizing XPU (likely cross-processing unit) models, particularly focusing on balancing performance and power requirements to control total cost of ownership. It mentions working closely with partners to optimize for various tasks, such as pretraining and inference. There is a transition in the conversation where Christopher Rolland from Susquehanna asks a question, seeking insights from Hock and Kirsten on new greenfield scale-up opportunities in connectivity technologies like optical or copper, and how these could benefit the company. He also inquires about the increase in operational expenditures (OpEx) related to AI opportunities.

Hock Tan discusses his company's portfolio and its significant opportunities in the area of networking connectivity, specifically through optical means. The focus is on expansion and next-generation technologies, with products such as multimode lasers, VCSELs, single-mode lasers, and various networking switches like Jericho and Tomahawk. While the company's products contribute to a considerable portion of AI-related revenue, averaging around 30%, the majority, about 70%, comes from XPUs and accelerators. Tan highlights the broad range of products and architectures the company offers, contributing to their growth and success in the AI market.

In the paragraph, Christopher Rolland and Kirsten Spears discuss R&D spending and focus areas for their company. Kirsten notes that they spent $1.4 billion on R&D in Q1, which is expected to increase in Q2, focusing on next-generation products like a two-nanometer AIXPU package in 3D and enhancing Tomahawk 5 for AI scaling on Ethernet. Hock Tan mentions that the networking side saw a surge in Q1 but expects the normal mix to return to a 70-30 split in favor of compute over networking. He also states that the company is not currently considering mergers and acquisitions (M&A) due to its focus on AI and VMware. Vijay Rakesh from Mizuho asks about the AI side and potential M&A, and Hock responds that their current focus is not on M&A.

The paragraph discusses the conclusion of a question-and-answer session and an earnings call. It mentions that Broadcom plans to report its second-quarter earnings for fiscal year 2025 on Thursday, June 5th, 2025, after the market closes, with a public webcast scheduled at 2 PM Pacific. The call is then concluded, and participants are thanked for joining, with the operator confirming the end of the program and allowing participants to disconnect.

This summary was generated with AI and may contain some inaccuracies.