4% in Q2 and 18.5% a year ago. R&D as % of sales in Q3 was, I think, 9.6% as compared to 10.2% in Q2 and 8.9% a year ago. SG&A spending to sales was 10.6% compared to 11.2% in Q2 and 9.6% a year ago. Thanks to the GP expansion, OP in Q3 improved by 38% year-on-year and 25% quarter-on-quarter. OP margin in Q3 reached 14.6%, marking a new record. Among all segments, power electronics saw the fastest sequential growth in Q3, while mobility and infrastructure showed modest quarter-over-quarter increases, and automation experienced a slight sequential decline. Year-over-year, all segments except mobility, which faced challenges from last year's high base and sluggish market demand. Other segments show varying degrees of improvement. Earnings-wise, power electronics and infrastructure showed strong improvement compared to Q2 and a year ago.
With infrastructure benefiting from a low base, the earnings of mobility remained stable compared to Q2 but dropped from the high base of a year ago. Automation suffered profit contractions in Q3 due to weak demand. Q3 non-operating profit was 1.2 billion NT dollars, dropped from 1.9 billion NT dollars in Q2 and a year ago. The foreign exchange loss in Q2 was mainly due to the strong Thai Baht versus the U.S. dollars. Since the portion of our production costs, including labor and overhead, are in Thailand, the appreciation of the Thai Baht can negatively impact our foreign exchange income. In Q3, the Thai Baht appreciated by nearly 10% against the U.S. dollar. In Q3, we had 17.7 billion NT dollars profit before tax, up 28% year-on-year and 18% quarter-over-quarter.
Q3 EBITDA was TWD 24.2-24.3 billion, up 24% year-on-year and 12% quarter-on-quarter. Q3 tax expense was about TWD 3.5 billion, representing a 20% effective tax rate. Net profit after tax was about TWD 12.4 billion, up 32% year-on-year and 24% from Q2, reaching a new record. So Q3 EPS was 4.75. Now we have a look at the accumulated numbers for the first three quarters of the year. The revenue was TWD 306.9 billion in the first three quarters, up 2% from a year ago. GP was up 17% year-on-year, while GP margin was 33% versus 28.8% a year ago, thanks to the better mix and some meaningful reversal of the inventory provisions. The operating expense in the first three quarters was up by 15% year-on-year, with SG&A up 13% and R&D up 16%.
OpEx ratio increased to 21% from 18.7% a year ago, with the SG&A expense ratio expanding to 11.1% from 10% a year ago, and the R&D expense ratio increased to 9.9% from 8.7%. OP margin in the first three quarters improved to 12% from 10.2% a year ago, thanks to the improvement of GP margin. For the first three quarters, power electronics and mobility grew by 7% and 1% respectively, and the revenues of automation and infrastructure contracted by 7% and 4% due to the sluggish demand and the high base of last year. Earnings-wise, both electronics and infrastructure show strong profit improvement, while the profit of automation and mobility shrank from a year ago due to the high base and slow economy. For the first three quarters, we had about 4.5 billion NT dollars non-operating profit. In total, we had NT$ 41.5 billion pre-tax income.
Our EBITDA was NT$ 60.9 billion, up 16% from a year ago. So the tax expense was around NT$ 8.4 billion, representing a 20.3% effective tax rate. So therefore, the net profit after tax in the first three quarters was NT$ 28 billion versus NT$ 24.4 billion a year ago. So the EPS in the first three quarters of the year was 10.80 versus, sorry, 10.8 vs. 9.4 a year ago, representing a 15% year-on-year growth.
So, can you give us some idea regarding your view on the data center demand? In terms of the AI data center demand, I think, is mainly coming from the AI data center deployment and construction. If you look at all those leading hyperscalers, especially in the U.S., if you look at their CapEx planning, which continues to increase, continues to increase in the near term. If you just look at Q4 of this year, I think their CapEx, I mean, per quarter can reach $40 billion per quarter. In terms of the opportunities we can embrace in this data center booming era, I think, if you look at our product offering, I think we can, I think we can embrace the business opportunities, especially, on our, I mean, with our power technology.
So if you look at all those, I mean, data centers, especially for the AI applications, they are actually extremely power-hungry. So basically, those data centers are consuming a lot of electricity. So whenever, actually, you need to, I mean, use the electricity, you always need to have a device to convert the electricity, which is exactly where we can help our customers, not just on the cooling side, but also on the energy management side. So I think we may probably benefit from the demand of liquid cooling projects as well as the energy management products. So in terms of the liquid cooling products, can you please elaborate on that? So do you ship more component-level products or system-level projects?
I think we actually supply both components, cooling components, and system products and system-level products. However, the majority of revenue is sourced from the system-level business. As we previously explained, I think we enjoy some competitive advantages, including our relationship and track record with those leading customers, I mean, leading CSP and hyperscalers, because for those, I mean, major projects, I believe the customers they would not be willing to take a risk just because of lower pricing for one or two components. I think the real reliability is still their top priority in terms of whenever looking for the suitable, I mean, suppliers for their liquid cooling projects. Can you share your view on the outlook for Q4 of this year?
In terms of the Q3 revenue momentum, we actually are expecting to see some sequential increase.
Sorry, we are expecting to see some year-on-year improvement for the Q4 revenues. But the business mix might be a little bit different from Q3. So for example, we are expecting to see power to be flat or maybe slightly down, and the component business is likely to be down a little bit. And also, we are also expecting ICT business to improve. And then, for EV business, it's more likely to be flat. So in terms of the liquid cooling products versus the air-cooling products for next generation AI products. So I think it's really subject to clients' requirements. But even though, within the system level or within the rack level, it's going to be the liquid to maybe purely liquid to liquid solution.
But at the end of the day, if you want to dissipate the heat outside of the building, you still need to use air-cooling technology to help you to dissipate the heat. So my question is, regarding, are you seeing any differences between the AI customers versus the traditional power supply customers? The reason why I was asking this is because, I mean, for other or traditional power supply customers, they may actually concern more about the cost and also as well as the power efficiency. But is it the case, I mean, for the AI customers, they actually care more about the energy efficiency instead of the cost? So also, another question is, how do you see the capability of mass production of those liquid cooling products?
I think the capability or ability to mass produce the products are always going to be one of the top concerns when customers are choosing or selecting their suppliers. And also, the lead time is another top priority when they are selecting their suppliers. So, because I think you are probably one of the very few suppliers being able to provide both the power products and the cooling products. And do you see any advantages when you are able to, I mean, supply both types of products? I think it's actually not that easy to really, I mean, integrate the power solutions and the cooling solutions, because if you look at the power supplies, they are essentially not the biggest heat generator. Actually, the chips generate the most heat.
Is it really cost-effective to, I mean, to combine both, I mean, power solutions and cooling solutions? I'm not quite sure. But our advantages are we are able to, I mean, help our customers to reduce the power consumption by improving the energy efficiency per product. I think maybe some of you may have noticed the news. Microsoft recently is planning to reactivate a nuclear plant on an island, which was hard to imagine in the past. But it can kind of, I mean, somehow, or sorry, somewhat indicating, I mean, how much electricity will be needed for the AI data centers. Because the electricity resource is always limited, it's actually pretty hard to, I mean, to increase the electricity supply, so customers may actually turn to focus on how to save more energy.
So can I say that is actually one of your key advantages? Yes. And then I think it's not just about the energy efficiency, but also about the power density. So I think, yeah, of course, in terms of the power conversion, efficiency, that is something that we can work on. But in terms of the processing or computation efficiency, that is not something that we can address. So that actually requires the efforts from the AI software companies. So can you share your product offering for the liquid cooling solutions? So I think the bottleneck in terms of supply chain isn't really our top concern by far. But what we really concern is actually the manufacturing process, because it's actually quite different from how we make the cooling fans.
For example, for the liquid cooling products, you need to solder, and then also you need to have some sort of, I mean, some sort of mechanism to prevent leakage. So I think the manufacturing process is actually more critical, in our view.
Yeah, but I don't really, I mean, I haven't really seen any significant issues in terms of the components. So I was told that there is still some leakage issue, I mean, yeah, in the market. I think it's still not. We are not really there yet. I think, right now, I think the most critical thing is actually the system design, whether or not your system, your system design can efficiently and effectively to dissipate the heat.
I think in terms of the leakage is actually the next stage. I mean, when you try to or plan to mass produce the products, because there are going to be a lot of connecting points, I mean, within the whole systems. It's challenging, but I think, for every company, I mean, we are going to see some learning curve. What are the implications of the lawsuit with, I mean, a U.S. peer? I think, if you look at the power market, it is a long-standing market for tens of years, for a couple of decades. Many or every power company, they actually has a lot of patents.
In our case, whenever we try to or plan to develop a new product, we always do the IP study. Sometimes we also invite the external experts like the professional lawyers to help us to study and review the IP study. I think for this case, we did, I mean, have done the IP study for our products. At the end of the day, I have to say the thing is, we can only do the best, whatever the best we can do. You can never really prevent anyone to, I mean, to sue you, I mean on certain technologies. I think the case is still under review by the U.S. ITC, with the final decision is vetted by February next year.
And then for the initial determination, I mean, made by ITC, I would consider is actually a mixed result. But after receiving the initial, I mean, determination by ITC, we actually have already, I mean, redesigned our products and proposed new designs to our customers. And then we also submitted our new designs to the judge and being proved that our new designs actually are free from any controversial or IP issues. So, I mean, when it comes to the IPs issue, actually whenever you ask maybe different, especially the different senior engineers, they may just have different point of views on this. But at the end of the day, I mean, we do not expect it to have a significant impact on our revenues next year. So my next question is regarding your AI-related products.
So in terms of the AI-related products, do your customers have any kind of concerns in terms of the manufacturing locations? And then in terms of the CapEx, can you share the percentage in terms of how much you invest into the CapEx capacity expansion for next year, especially for the cooling solutions? I think yes, especially for the U.S. customers, they would not be willing to, I mean, to see their products to be produced in China. So in that case, we will have to produce the products in other countries. And then speaking of the CapEx for capacity expansion, I don't think that we actually allocate a lot of, I mean, capacity expansion. So when can we see the meaningful contribution from liquid cooling products? Is it going to be one Q?
Yeah, we actually are shipping some of our products, I mean, this year, last year, this year. But in terms of the ramping of contribution, I think it's going to be next year, so can you share your views, I mean, regarding liquid-to-liquid solutions and liquid-to-air solutions? I think for those two different solutions, especially for liquid-to-liquid solutions, it actually requires a new data center, and then for liquid-to-air solution, actually you can just do some maybe renovation, some, I mean, which can enable you to, I mean, adopt or employ, deploy such liquid-to-air solutions without, I mean, building up or a new data center from scratch. If you want to pursue a pure liquid-to-liquid solution, then you have to build a new data center from scratch.
So, does that mean that actually you are not really seeing a lot of, I mean, new data centers? No, no, of course not, because if those data centers are for AI applications, I think they are actually building, they are actually aggressively building new data centers. So I can add some points on that. So in terms of the liquid-to-liquid solution, we do see some demand for liquid-to-liquid solution. But for our customers, they also have the concerns on the OpEx, I mean, and their CapEx. So it is definitely not a mainstream solution yet. And it is actually a highly sophisticated solution. So can you share your view on the 2025 outlook? So in terms of the 2025 outlook, I mean, I think that we do definitely expect continued growth for the overall company.
In term and then AI is definitely going to be one of the key things. But I think many of you have heard about our DC-DC, I mean, converters. It is not related, I mean, by our customers. I think the main reason or main concern of our customer, I mean, regarding the DC-DC converters, actually, our DC-DC would be the sole source if they choose this solution. So they decided to make the DC-DC, I mean, conversion on board. So for next year, I think DC-DC, I mean, the revenue of DC-DC converter business may drop. But the whole power supply business group will continue to grow.
And then for the ICT segment, I think because of the increasing demand, I mean, from the data center, so we do expect to see the growth, I mean, of, I mean, ICT segment. So I think I can also add some points on this question. I think especially from the technical level point of view. So I think even though, I mean, one of our DC-DC converter customers decided to use this discrete solution to replace the modular solution. But that is actually a reference designs for their customers, for its customers. So but who they really, I mean, determine the system designs are actually the OEMs. And also some of the CSPs, they also have their own opinion on how to design their, their, their own system. So because there are actually different pros and cons, I mean, for those two different solutions.
It's more of a reference designs. So it doesn't mean that all customers, I mean, they are using the new Blackwell series, and then they are going to use the discrete solution. I think both solutions can exist. So it's really subject to customers. So how do you see the changes? I mean, is there any change in terms of the competitive landscape? Yes. I mean, there are a lot of companies, I mean, claiming that they are also able to produce or provide or supply such products for customers. But I think, because as I said, I mean, those products are highly engineered, sophisticated. So because if you look at those major customers, they are actually colos, either colos or CSPs. I think they would, I mean, rather choose more reliable and well-established companies for their products and solutions.
So because we actually have, I mean, long-term relationship with those leading, I mean, customers, especially during the stage of very early design cycle. So I think that is one of our advantages. So can you share, I mean, your view on maybe how much percentage in terms of the liquid cooling as a percentage of total revenues in Q4 and next year? Because we actually don't really make provide official, I mean, financial guidance. So I can hardly answer, I mean, this question. But we do expect to see, I mean, meaningful increase on our, I mean, from our ICT segment in Q4. Can you elaborate on your exposure to the AI applications? I think not just our ICT segment, but also our power supply power supplies business can benefit from the AI data center deployment.
So can you share, I mean, any parameters on some less discussed businesses, just like your UPS products, and networking products, and so on and so forth? So I think what we have been doing is actually, for many different markets or many different businesses, we are trying to review and to see whether we are able to provide more system or solution level products for our customers. But because we are still, I mean, in this process, so we will, I mean, share with you when we have made more progress. So we have some questions from the online audience. For the first one is, I mean, did Q3 gross margins also significantly benefit from the reversal of inventory write-downs? Yes, much like in Q2. This benefit was still quite evident in Q3.
So, for Q4, the gross margins, I think, is expected, actually subject to the business mix. And then also subject to the inventory and also the shipment. So what is your outlook on the expense ratio for 2024 to 2026? I think our budget for this year projects around a 15% increase in expenses. Next year's budget is still being finalized. And then we do expect some growth. But it's difficult to predict precise expense ratio due to many variables in revenue. And then for next year's CapEx, I think due to the purchase of some office buildings, this year's capital expenditures may increase by NT$6 billion-NT$7 billion, compared to last year. Although the expected figure will depend on the progress of new facilities in the final months.
Next year, capital expansion will focus on factory automation, capacity expansion, and construction of new facilities likely amounting to somewhat less than this year. So why has, I mean, the royalty payment to the parent company shown an upward trend in Delta Thailand financial statements? Since most of Delta's R&D is conducted in Taiwan, the rapid expansion of capacity in Thailand has increased reliance on Delta's, I mean, R&D, resulting in gradually rising royalties. So this upward trend is expected to continue in the foreseeable future. However, the exact calculation depends on per product-by-product evaluation, so it cannot be explained with a simple formula. So I can also add some points on that. So for some of the orders, actually, customers place orders on Delta Electronics. Delta Thailand also only does the manufacturing, contract manufacturing part.
And then we do all the sales marketing and R&D. That's the reason why they have to pay the royalty fees to Delta. The main reason is because, actually, Delta Electronics, I mean, takes the most responsibility of R&D. The upward trend is expected to continue. The next question is, do you, I mean, expect to see the capacity in China continue to grow, I mean, in the future? Yes. And then because we are actually still in the process of diversifying our capacity. So, I think, currently, our China capacity accounts for like around 60% of our total capacity. But because we are also building capacity outside of China.
So I think, China capacity as a whole, as a percentage to the whole capacity is going to, I mean, be somewhere between 50%-60%. So what is Delta's approach to mergers and acquisitions for business expansion? I think the company actually considers M&A as an essential part of long-term growth. We actually always have a lot of, I mean, targets, I mean, or projects, in the pipeline. But we are always very, more cautious and more conservative whenever we are considering a new target or a deal with a new target.
So, in the future, when we are looking for the new targets, I think we will definitely consider the integration potential between the new target, the company, I mean, the target company's products and our product portfolio because we always, I mean, we always have to go to, I mean, to integrate more of our products into solutions. So, can you share your view on the EV business? I think for next year, I think it's going to be stable. I mean, for Q4. And next year, we expect to see some growth for next year. So my next question is regarding the OP margin because next quarter, as you said, you are expecting softer gross margin, but with maybe expanded economies of scale. Can we expect some further improvement on your OP margin?
I think I would tend to be more conservative.
Okay. So thank you for coming today. And we will see you next year. Thank you.