ARM: Most of the future growth will come from royalties, not licensing revenue (Q3 FY25 conference call)

ARM (ARM.O) released its Q3 FY2025 financial report (ending December 2024) after U.S. stock market hours on February 6, 2025, Beijing time. The key points are as follows:

The following is a summary of the ARM Q3 FY2025 earnings call. For the financial report interpretation, please refer to "ARM: AI Drives Performance to Meet Targets, High Valuation Remains a 'Sweet Burden'".

1. $Arm(ARM.US) Core Financial Information Review:

2. Detailed Content of ARM's Earnings Call

2.1 Key Information from Executives:

1. Financial Performance

Total revenue for Q3: $983 million, exceeding the expected range, up 19% year-over-year.

Royalty revenue: A record $580 million, up 23% year-over-year, exceeding expectations. Growth was primarily driven by the continued adoption of the Armv9 architecture and chip shipments based on the Compute Subsystem (CSS).

License revenue: $403 million, up 14% year-over-year, above expectations. License revenue fluctuates significantly due to the timing and scale of high-value licensing agreements; it is recommended to focus on Annual Contract Value (ACV), which grew 9% year-over-year in Q3.

R&D expenses: Non-GAAP operating costs reached $522 million in Q3, a historical high, in line with expectations. The company will continue to increase investment in R&D to support long-term growth.

Non-GAAP operating profit: Reached nearly record levels at $442 million.

2. Business Progress

AI Demand Drives Growth: AI demand continues to drive the development of the Arm ecosystem, with the company reporting record total revenue and royalty revenue. The adoption of Armv9 architecture and CSS technology fueled this growth.

Smartphones and Data Centers: OPPO and vivo's new flagship smartphones adopted MediaTek's Dimensity 9400 system-on-chip, which is based on Arm's CSS. Market share for Arm-based chips in data centers, such as AWS Graviton, Microsoft Cobalt, Google Axion, and NVIDIA's Grace, is steadily increasing.

Ecosystem Investment: Arm has over 20 million developers, making it the largest developer community in the world. NVIDIA's announced Project DIGITS combines the Arm-based Grace CPU and Blackwell GPU to form a new GB10 super chip, supporting the world's smallest AI supercomputerKey Projects: Arm has participated in projects such as Stargate (AI infrastructure deployment) and Cristal Intelligence (enterprise-level AI), further consolidating its core position in the AI ecosystem.

3. Future Outlook

Fourth Quarter Expectations: Revenue is expected to be between $1.175 billion and $1.275 billion, a year-on-year increase of 32%. Non-GAAP operating expenses are expected to be $590 million, and non-GAAP earnings per share (EPS) are expected to be between $0.48 and $0.56.

Fiscal Year 2025 Full-Year Expectations: The midpoint of the full-year revenue expectation has been raised to approximately $4 billion, a year-on-year increase of 24%, exceeding the long-term target of 20%. It is expected that the year-on-year growth rate of royalty revenue will be in the high teens, and license revenue will grow by approximately 30% year-on-year. Non-GAAP operating expenses are expected to be $2.1 billion, a year-on-year increase of 21%. Full-year non-GAAP EPS is expected to be between $1.56 and $1.64.

2.2 Q&A:

Q: Regarding the growth points of license revenue in the coming year. This question is also related to some recent news we have seen in the AI field, and it also involves Arm's capabilities. Two weeks ago, we saw the announcement of the Stargate project, with a budget exceeding $100 billion. Recently, we also heard about the new collaboration project Cristal Intelligence between SoftBank and OpenAI. We are trying to understand Arm's position in these projects, as it seems that Arm is both a driver and has a first-mover advantage in the AI era. Could you please outline Arm's opportunities more clearly, including the possibility of selling new products to these new projects and the potential long-term benefits for Arm?

A: The Stargate project is an extremely important infrastructure project, with the U.S. immediately investing $100 billion and planning to invest $500 billion in the future. This is a project in collaboration with OpenAI, Oracle, and SoftBank, with technology partners including us, Microsoft, and NVIDIA. For Arm, we are very excited to be the preferred CPU for such a platform. With the Blackwell CPU and Grace, Arm will be the preferred CPU for the initial configuration. Looking ahead, there is tremendous potential for technological innovation in this field. This is a very exciting project, and we believe it will have a transformative impact on the industry.

The Cristal Intelligence project focuses on agent AI, where agents will move across every node in the hardware ecosystem. From the smallest devices (like earbuds) to data centers, this essentially means that agents will become the core interface or driving force for AI within devices. For Arm, this is a significant opportunity, as AI workloads will run on every endpoint I mentioned. Moreover, given that Arm is the most widely used computing platform globally, these AI workloads will run through Arm. Through our KleidiAI library, we will enable developers to easily optimize these agents for the Arm platformTherefore, the announcement of these two projects is of significant importance to both the industry and our company.

Regarding the drivers of license revenue growth, this quarter's license revenue exceeded our expectations by approximately $27 million, a year-on-year increase of 14%. The main drivers over the past few quarters remain the demand for AI, the continued adoption of v9 technology required for AI chips, and of course, CSS. As Rene mentioned, we have also signed some additional CSS contracts, and we are continuing to advance.

In the upcoming fourth quarter, we expect license revenue to grow by approximately 60% year-on-year. We have mentioned since the beginning of the year that we plan to sign some large contracts in the fourth quarter. Most of these contracts are proceeding as planned and are in line with expectations. These contracts are very much aligned with AI and CSS-driven categories.

Q：You mentioned that the demand for CSS from partners is stronger than initially expected. So, how should we think about the growth momentum of royalty revenue as we enter fiscal year 2026 compared to your previous expectations?

A We are not yet ready to discuss the situation for fiscal year 2026. However, based on last quarter's exit rate, we are very satisfied with the current momentum. For example, last quarter's royalty revenue reached $500 million, approximately 13% higher than our previous record of $514 million. Of course, this is not surprising, as this is also the first quarter we have seen substantial revenue from CSS, primarily driven by the Dimensity 9400 chip we mentioned. Additionally, Cobalt CSS has also started shipping and deploying, which is translating into royalty revenue. With the increase in CSS deployments over the next few quarters, we expect the momentum to continue. As for quantifying this momentum, we will need to wait until next quarter.

Q：Last month we saw news reports about a lawsuit involving one of your major clients. According to media reports, there seems to be some disagreement regarding the contract expiration date. They believe the contract expires in 2028, while we believe it is 2025. I know these are all comments based on calendar years, but could you share your current perspective? For example, do you expect any revenue that will not be realized in the next calendar year, or can we quantify that?

A This question pertains to the impact of the lawsuit with Qualcomm on revenue. We have predicted since the IPO that we would not win this lawsuit. The main reason for this lawsuit is to defend our intellectual property, which is very important. However, from a financial perspective, we have always assumed that we would continue to charge royalties at the same rate as in the past, and that they would continue to pay.

Q：I did not see the contribution of v9 to royalty revenue in the shareholder letter. I may have missed it. If you have the numbers from last quarter, could you share them?

A In terms of total royalty revenue for last quarter, the percentage was 25%, consistent with the previous quarter.

Q：The 25% figure has been consistent for several quarters. Do you expect it to rise again in the coming quarters? I know the answer may be affirmative, but... could you help us understand why it has stagnated at 25% over the past few quarters?A First of all, this is a total percentage, so we see a growth of 23% this quarter, which means it is flat as a total percentage. But if you look only at the dollar revenue from v9, its share has increased from 15% last year to 25% this quarter. Therefore, absolute dollar revenue has grown by three digits. So, the relative flatness is actually a good sign, indicating that we still have a long growth runway. We still expect v9 to grow to 67% to 70% of total royalty revenue. Moreover, being able to meet or exceed the royalty growth rate without an acceleration in v9 growth should give us more confidence in our ability to drive further royalty growth in the coming quarters and beyond.

Supplement: Perhaps just to give you a sense of how to think about these transitions, they are mainly driven by the transition of OEM products and when they are launched. For example, the MediaTek 9400 has already been designed into OPPO and vivo phones, which are now in mass production. Therefore, you will see peaks as production ramps up, followed by stabilization. But with the release of new versions, two things will happen. Broader adoption of high-end products, as well as the transition from high-end to mid-high-end and mid-range products from v8 to v9. Therefore, the transition rate we are seeing is completely in line with our expectations and is very consistent with the overall growth trajectory we anticipate. Thus, we are very satisfied with the current progress. As you mentioned, there is room for growth, but this is mainly driven by OEM changes in chip portfolios rather than new licensing companies.

Q：Regarding the question of royalty growth, can you speak more specifically about the situation in fiscal year 2026, as well as the contribution of v9 throughout the year and the evolution of the product portfolio?

A We will provide guidance for fiscal year 2026 next quarter, just like we did last year, when we typically do so during the announcement of fourth-quarter results. So we will provide more details. I just want to say that we have previously stated that we expect royalty revenue to grow by about 20%. We do not have an update on that forecast today.

We will achieve this through the combination of v9 and CSS. Therefore, what you saw last quarter was that v8 actually had strong growth. That’s good, but overall, you should expect v9 to continue to grow. But also remember that v9 is just one factor of growth. CSS is a larger growth factor because the royalty rate for CSS is about twice that of v9, and of course, the royalty rate for v9 is about twice that of v8. Therefore, you have to consider all factors comprehensively, but given the momentum from last quarter and our expectations for the next quarter, we believe this will lay a good foundation for fiscal year 2026.

Supplement: The CSS royalty rate for fiscal year 2026 may not necessarily be the same as that for fiscal year 2025. These CSS rates will change over time, and they will vary year by year as new solutions are brought to market. Therefore, these are subject to change over time and will increase over time.

Q：License revenue is obviously difficult to predict for a specific week, day, or month. But are there some structural changes happening, such as the complexity of your customer agreements, which means these agreements will take longer to complete?Should we expect this situation in the coming years?**

A: You need to categorize the contracts mainly into two types. One type is new contracts signed with companies we have never collaborated with before. This may take longer. The other type is the renewal of existing ATA (Technology License Agreements), which usually has a shorter cycle. However, larger transactions generally require more time. When the transaction amount reaches hundreds of millions of dollars, they may involve different approval processes, such as the board of directors, which can take many months. But overall, regarding your question, will there be any changes next year or the year after compared to the past year? No, there are no changes.

Q: Regarding AWS, you have achieved great success with AWS, with nearly 90% of the top 1,000 customers using Graviton and Arm IP. Can you talk about the progress of Cobalt 100 and how fast it is developing? I have a follow-up question.

A: I will leave specific comments about Cobalt shipments to Microsoft. But overall, what we are seeing is not just an increase in the deployment momentum of these products in the cloud. Moreover, the transition from x86 plus H100 to Grace Blackwell GB200 is also facilitating Arm's development in data centers. Because the main CPU for these AI data centers is Grace, from the perspective of host control, this is a good tailwind. Therefore, we are very satisfied with the overall momentum. I will again leave specific comments about Cobalt to Microsoft, but the momentum has been very good.

Q: So regarding related parties, I see a quarter-on-quarter growth of 48%. Do you see any changes in the long term? Will this growth continue?

A: I think this will be quite stable. The largest portion of related parties is Arm China, so we expect this to remain relatively stable. Arm China may account for a smaller percentage of revenue. I think this quarter is about 25%. I have previously mentioned that we expect it to decline to a mid-teens percentage over the next few years. However, there may not be significant changes in the coming quarters.

Q: A question about mergers and acquisitions. There has been speculation in the media about mergers and acquisitions related to Arm or SoftBank, particularly regarding whether you might move beyond the traditional pure IP model and further into the silicon space. Could you please talk about this, even though you won't comment on specific mergers and acquisitions, can you discuss the internal discussions you have had with customers and their needs that might lead you in this direction in the future?

A: As you might guess, we cannot comment on any speculation or rumors. Most of my time is spent thinking about growth and the future. I think we are all trying to consider the pace of AI development and the speed of changes in software models, which puts tremendous pressure on our ecosystem to develop products faster and better.

As Arm, we are at the core of all this, and we strive to solve our customers' problems. But unfortunately, I cannot provide more details, and of course, I cannot speculate on any merger rumorsQ Your team has always led with a system-level strategy to drive demand for computing IP solutions, right? AMBA is a great example, and so is CSS. It seems that your team is now expanding this strategy into the complex SoC market with a rapid shift towards this more heterogeneous chiplet strategy.

You are building an ecosystem around CSA (Chiplet System Architecture). How has the progress been so far? Have you seen an acceleration in CSA activities? More importantly, how will the Arm team achieve profitability through this strategy?

A We have indeed seen an acceleration in demand for CSS. Customers clearly see its advantages. We are now involved in CSS in almost all major markets we participate in. We have also launched what we call the Arm Total Design Partner Program. This allows end partners and design companies to adopt our CSS and potentially develop chiplets, and the demand for this initiative is very strong.

What is really driving all of this is the increasing complexity of these designs, and the tight coupling between hardware and software. Since Arm is at the center of the software ecosystem for these chips, there is a significant demand for us to do more, faster, to help products get to market quicker.

We are in a great time in the industry where computing demand is outpacing the silicon to serve it. We are receiving many inquiries about the smartphone market and AI capabilities to leverage what is happening. You have to remember that the chips for these smartphones were designed 2 to 3 years ago, with everything like memory subsystems and power already predefined.

Therefore, adapting these small language models or anything else into phones, given that you still need to run displays, operating systems, and applications, is quite a challenge. So helping to solve this problem and getting products to market faster is exactly what we are very focused on.

Q：I want to return to the issue of v9 adoption. Of course, in absolute numbers, it represents more dollars, but as a percentage of royalties, it has stagnated at 25%. I wonder if you could explain what has changed compared to the assumptions at the beginning of the year. Since we haven't seen it grow in the last quarter of this year, does this create an acceleration or upward driver for next year? Or is this not the right way to look at the issue?

A：So, what you really want to ask is whether this creates conditions for growth next year? I think it can. I just don't know if it will be next year or the year after, or exactly when that will be. As you mentioned, our previous assumption was that it would grow from 10% to 15%, then to 20%, and then to 25%.

Now it has stabilized at 25% for several quarters. The reason we provide this metric is to help people understand the adoption of v9. The slowdown in the growth of v9 as a percentage of total royalties is actually a good thing because it gives us a better growth outlook, as we still believe it will reach 60% to 70%Why do we believe this? Well, we have signed contracts, and we already know about the upcoming V9 products. So, the question is when will they be shipped, and what is the product portfolio? Ultimately, these are the driving factors. To emphasize again, the reason we provide this number is to help people understand the leading indicators of royalty growth.

Therefore, being able to meet or exceed the royalty numbers even in the case of a slowdown in the V9 portfolio growth is actually a good thing. But I think you should focus more on the overall royalty growth rather than the adoption of V9. When we provide this metric, the rate of royalty growth is much slower than it is now. So, I think you should focus more on the overall royalty growth rather than the adoption of V9.

Q This is a more macro question. Recently, many investors have been discussing training optimization issues, and there has been a lot of discussion about DeepSeek. Assuming these are all valid, does this change your view on Arm's opportunities in the AI space? More broadly, I would like to hear your thoughts on this topic, not just about Arm.

A First of all, if you consider DeepSeek's achievements, whether it's their V3 general model or R1 inference model, clearly, they have done a lot of work based on existing work (i.e., cutting-edge models created by others), and then DeepSeek has done very creative work to build a very efficient inference model.

I think this is good for the industry because it increases efficiency and lowers costs. By doing so, it expands the demand for overall computing. So from a general perspective, this is a good thing. When you consider Arm's applications, given that AI workloads need to run anywhere, more efficient inference reduces costs, making it easier to run these applications in power-constrained and compute-constrained areas.

This is very good for Arm. Products like Grace Blackwell are great, but you can never put them in a phone, in earbuds, or even in a car. But Arm is everywhere. Therefore, I think when the overall inference costs are reduced, this is good for Arm.

I also think we are far from being good enough because if you look at the spending announced by the big players in this market, I think Google just announced it will spend $60 billion in its earnings call. Satya (Microsoft CEO) talked about $80 billion. Meta mentioned $40 to $50 billion. No one is backing down. The reason is that we are far from the transformative capabilities that AI can achieve. So, I actually think this is a good thing, as it will increase overall computing demand. For Arm, this is even better because it allows us to operate in areas where efficiency is critical, which is exactly where our strengths lie.

Q Following up on the ACV (Annual Contract Value) question. You mentioned a slowdown in growth. I know you will provide full-year guidance later. Just curious how you view ACV in the long term. Are there any large data center or mobile projects coming up that might restore it to double-digit or high single-digit growth?How much of the ACV comes from Arm China business?

A Regarding our forecast and where we think it should land, if you go back to when we went public, we basically provided the ACV, believing it could grow in the mid-single digits to high-single digits. The reason is that if you consider all our deals as allocable, most of our license revenue and ACV actually comes from ATA transactions.

And almost all ATA transactions typically have a 7% annual growth. So it should be... and all non-ATA transactions won't have this. So it could be around 7% up or down. So that's our forecast. Then over the past six quarters, due to AI, we've seen an acceleration in license revenue, primarily around the adoption of v9 and now CSS. So this gives us upside from that 7% range all the way up to a high point of 14% or 15%.

Will we return to the mid-teens range? I don't know. This is really something we've never truly forecasted. When I provide guidance next quarter, I'll try to give you more insight. But in terms of long-term growth models, you should expect most of the growth to really come from royalties. So, of course, the royalty rate for v9 is much higher. Both v9 and CSS do not provide a one-time increase in royalty rates.

They are a stepwise change, but typically there will also be annual increases with the release of new versions. In the future, you should expect most of the growth to come more from royalties rather than licensing revenue.

Q How much of the ACV comes from Arm China?

A The Arm China portion is about 20%, so quite close to the overall mix. Last quarter it was 25%, so it's in that range.

Q A question about CSS, is the CSS licensing activity primarily in data centers, or is it balanced between data centers and mobile? And Rene, I just want to confirm what you mentioned earlier. When AI enters smartphones, does that mean these processors will adopt chiplet architecture? So in the long term, the attach rate of CSS in smartphones could be very high?

A I think every endpoint you described, earbuds might not be applicable due to their small size. But this chiplet approach will become prevalent in almost every SoC (system-on-chip). In other words, inside the package, you will see multiple small chips everywhere. It has almost become a standard in high-end.

But I think you will see it everywhere, and this is a huge opportunity for us because it not only allows us to provide computing CSS from a CPU perspective but also enables us to have the right combination, whether it's NPU (neural processing unit) or combined with GPU, along with the right CPU combination to maximize performance.

You actually touched on a very key demand driver, which is precisely why we see strong demand for CSS types in all these endpoint markets.

Q Then in terms of IoT, will earbuds also become CSS devices? Or is it a single chip wafer with more processing power on Arm chips?Regarding the combination of CSS, we have discussed it in the past, and we have announced it in the automotive sector, but it has not yet arrived. So based on the dozen or so CSS licenses we have sold in the past, let's assume it is basically a 50-50 split between infrastructure and customer business.

Additionally, I want to emphasize that this chiplet architecture will be ubiquitous. This presents a huge opportunity for us, as we can provide computing CSS from a CPU perspective, and also combine NPU or GPU as needed, along with the right CPU combination for optimal performance. This architecture will permeate almost all SoCs, which is exactly why we are seeing strong demand for CSS across all end markets.

Risk Disclosure and Statement of this Article: Dolphin Investment Research Disclaimer and General Disclosure