Let me start by saying that the first half 2026 represent another milestone for Technoprobe. Our performance demonstrates not only outstanding financial execution, but also strength of our positioning in what we believe is one of the most attractive structural growth markets in the semiconductor industry. We delivered record revenues of EUR 464.1 million, growing 42.4% year-over-year and 48.2% sequentially as a result of a broad-based demand across our key customers and applications, confirming that the long-term AI investment cycle continues to accelerate. As volumes increased, we captured significant operating leverage across the business. Gross profit grew almost 66% year-over-year, with gross margin expanding to 53.8%, while EBITDA increased nearly 94% to more than EUR 206 million, delivering an outstanding margin of 44.4%. These results demonstrate the scalability of our business model and our ability to convert strong market demand into superior profitability.
Looking at the key drivers behind these results, execution has clearly been one of our greatest strengths. We achieved record revenues because of the manufacturing capacity improving and, at the same time, leveraging our cost structure, allowing profitability to grow significantly faster than revenues. The combination of execution excellence, operating leverage, and margin expansion is exactly what we aim to deliver as our business continues to scale. From a market perspective, AI data center remains by far the most important growth engine for Technoprobe. We continue to see exceptional investment across the AI ecosystem, with customer increasing spending to support ever more powerful computing platforms. We believe this is not a short-term phenomenon, but a structural transformation of the semiconductor industry. Within this environment, several trends continue to work strongly in our favor. First, data center applications remain the primary driver of demand for our advanced probe technologies.
Second, the rapid evolution toward increasingly sophisticated AI models, including the emergence of agentic AI, is pushing semiconductor complexity to new levels. Finally, every new generation of advanced devices requires significantly higher test intensity, increasing the value of our solutions and creating additional content opportunities for Technoprobe. These structural trends reinforce our confidence in the long-term growth trajectory of our business. Looking ahead, we remain very optimistic about the second half of the year. The positive demand trends we experienced during H1 are continuing. Customer engagement remains extremely strong. To support this sustained demand, we are continuing to expand our manufacturing capacity, ensuring that we can meet customers' requirements while preserving the operational excellence and profitability that have characterized our performance so far. Overall, we believe these results demonstrate that Technoprobe is executing exceptionally well, benefiting from powerful secular growth drivers and further strengthening its competitive position. Next.
Let me now provide some context on the market environment and our expectation for the second half of the year. Overall, the picture remains constructive, with the trends and momentum observed in the first half expected to continue into H2. However, the pace of growth remains clearly differentiated across the markets. Starting from AI, demand remains exceptionally strong and continues to be the main structural growth engine for the semiconductor industry. Recent industry commentary confirms sustained investment in the AI infrastructure, continued strong demand for leading-edge technologies, and increasing requirements for high-performance computing, advanced memory, and packaging. For Technoprobe, this environment is particularly supportive. The development of increasingly powerful accelerators, custom processors, HBM solutions, and complex chiplet architectures is increasing both device complexity and test intensity. Each new technology generation requires more sophisticated testing, tighter performance requirements, and greater reliability, reinforcing the strategic importance of advanced probe card solutions.
We therefore expect strong AI growth to continue throughout the second half, with no material change in the underlying momentum. Visibility remains solid. Customer engagement is high, and the broader AI ecosystem continues to invest aggressively in next-generation computing capacity. Moving to the consumer market, the trend is more moderate. We are seeing slight growth supported by a gradual normalization in smartphones and personal computers, selected product refresh cycles, and the progressive introduction of AI-enabled devices. Nevertheless, we are not assuming a broad-based consumer boom. The recovery remains measured and differs by customer, application, and product category. Our expectation for H2 is therefore for continued slight growth, broadly in line with the first half. The market appears healthier and more stable than in previous periods, but the magnitude of the opportunity remains significantly below what we are currently seeing in AI and data center applications.
Finally, in automotive and industrial, the key message is recovery. Inventory correction and customer destocking have progressed. Order patterns are gradually improving and an increasing number of applications are beginning to contribute to the upcycle. Within automotive, semiconductor content continues to increase as vehicles become more digital, connected, and software-defined. The near-term recovery is still progressive rather than uniform, and visibility can vary across geographies and customers. As a result, our H2 assumption for automotive and industrial is continued recovery rather than a sharp acceleration. We believe the direction of travel is positive, we remain disciplined and do not yet assume a full return to normalized demand across every application. To summarize, we expect the second half to reflect the same market structure seen in H1, strong and sustained growth in AI, moderate growth in consumer, and gradual recovery in automotive and industrial.
This mix is favorable for Technoprobe, given our strong exposure to the most advanced and test-intensive semiconductor applications, and supports our confidence in the continuation of the positive trends into the second half of the year. Let me turn to Stefano Beretta, who will give you more colors on our figures.
Good afternoon, everyone. As you may have read in our press release, revenues recorded in the second quarter marked another all-time record for Technoprobe, exceeding EUR 277 million, above the high end of the outlook range, registering an increase of 64.3% compared to the same period of prior year, with a sequential increase of more than 48% compared to Q1 2026. Gross profit has practically doubled compared to the same period of 2025, up to EUR 158.5 million, representing a 57.2% margin, also exceeding the high end of our outlook range. Even more extraordinary is the increase in the EBITDA, which in just one quarter reached EUR 137 million with an increase of 135% compared to the same period of last year, representing an impressive margin of over 49%, well above the upper end guided during our prior call.
Moving to the year-to-date figures already showed by Stefano Felici, total revenues were just above EUR 464 million, with a year-on-year increase of 42.4%. The gross profit was EUR 249.6 million, 65.7% higher compared to the same period of 2025 and representing a margin of 53.8%. The EBITDA closed at EUR 206.2 million, almost doubled compared to the same period of 2025, representing a margin of 44.4%. On this page, you can see a summary comparison between the financials at the end of the first six months, 2025 and 2026. To comment that further, revenue year-on-year increase was entirely driven by an organic growth of almost EUR 170 million, largely sustained by artificial intelligence volumes, as well as a soft recovery in consumer segment, together with a slight recovery in automotive and industrial. Dimension growth has been partially offset by the FX headwind for approximately EUR 30 million.
The revenue expressed at constant currency used in H1 2025 would have been approximately EUR 30 million higher than reported revenue, meaning an unfavorable impact of more than 6%. On a gross profit level, the increase in the margin of 760 basis points from 46.2%- 53.8%, as already mentioned in our prior press releases, confirm once again the ability to rapidly increase the production capacity and implement internal efficiency in our production processes together with the operating leverage across the entire P&L. All of these impacts have been partially offset by the increase in depreciation following the investments in fixed assets made during the latest months to expand our capacity and increase the automation. Gross profit at constant currency would have been approximately EUR 24 million higher by using the H1 2025 average rate, bringing the margin well above 55%.
At the EBITDA level, the effect of the operating leverage is even more evident, also favored by spending discipline in SG&A, and the integration of the R&D departments of the various divisions, showing an increase in the margin, even including a negative Forex impact for approximately EUR 23 million, which would have brought the margin above 46%. The net financial position has remained almost stable over this last six months, during which the cash generated by the operating activities for approximately EUR 93 million was practically all used to fund the capacity expansion made in the same period. As already anticipated in the previous press release, the group has recently embarked on a path to double its production capacity from 2025 benchmark by the first quarter of 2027 through an ambitious investment and hiring plan.
The results exceeded our expectation, and we now expect to be able to increase the capacity up to 140% compared to the run rate at the end of 2025. The complete success will depend on many factors, especially on the ability to quickly adapt to changes in global dynamics and maintaining focus on our technology, on our people, and on an ethical governance. The increase in the production capacity, combined with the stronger than expected volume growth driven by the artificial intelligence, leads us to expect a third quarter with another strong sequential growth in both revenue and profitability, and supported by a meaningful operating leverage effect. As shown in the chart, our investment mix is also evolving significantly.
Compared to prior estimate, we have farther accelerated the total investment to be deployed by the end of Q1 2027 to approximately EUR 350 million, of which approximately 80% of total investments are now concentrated in Italy, where the construction of the new greenfield manufacturing facility in Cernusco Lombardone has already begun, and planned to be concluded by the end of Q1 2027. That said, third quarter of the year is expected to show another sequential record revenue, together with robust growth in gross margin and EBITDA margin. Revenues to be about EUR 314 million, ±3%. Gross margin in the range of 61.5%, ±200 basis points, and the EBITDA margin in the range of 52%, ±200 basis points.
Considering that we are more than halfway through the year, and based on the volume and capacity information available to us to date, we believe it's appropriate to update the revenue and the EBITDA margin targets for 2026, revise it again upwards as follows. Consolidated revenues in the range between EUR 1,050 million- EUR 1,100 million. The EBITDA margin between 46%- 48%, this is all I have for the moment. Thanks everyone for your attention. Now we can move to the question-and-answer session.
Thank you to the speakers today. We now have an opportunity for questions. As a reminder, if you would like to ask a question, please use the raise hand function on your screen, or for those dialing in, it's star nine on your keypad. Once your name is announced, please remember to unmute your line and state your company name before asking your question. Thank you. The first question today comes from Alberto Gegra. Please, Alberto, go ahead.
Hi, good afternoon, congratulations for this another strong set of results. I have one question on the implied second half. If you have anything to flag, just to better understand the reason behind the phasing between the third and the fourth quarter since you are having a very strong third quarter than the last one, slightly down sequentially. The second question on the capacity increase, just to reconciliate the previous messages of around EUR 1.4 billion run rate of manageable sales by the end of first quarter. An update on this figure, consider the updated CapEx, also considering DIS and potential Chinese revenue. The third one, if you can update us on the new market, what are you seeing in the discussion with your potential customers? What drivers do you expect in 2027, in particular between CPU, ASIC, HBM, and silicon photonics? Thank you.
Thank you, Alberto. Let me start for the H2 trend. For the moment, we guided Q3, of course, and Q4 is still under examination. What we show now is an upgraded trend for the year end, but we still have some lag of information. As you know, the visibility is very short and not yet completed for the Q4. What we can see right now is that a significant mass production and cyclical trend, especially for the GPUs, that led the significant growth the first half of the year would take a kind of pause for the second part, especially for the last part of the year, and then to restart in terms of volumes in 2027. In the second part, especially in the last part of the year, we expect a bit of slowdown in the GPUs, softly replaced by CPUs and ASICs.
Volumes will be different in that case. This is something we expected usually in the Q3. It's a kind of cyclicality in the past years. The cyclicality has been moved a bit farther along the year, so we expect that in the Q4. About the capacity increase, as I mentioned before, we have increased the investment expected by the end of Q1 2027 from EUR 250 million- EUR 350 million, EUR 100 million more. This additional increase that is almost entirely addressed in the equipment and automation, not only in Italy but also in all our facility, especially in Taiwan and Korea and China as well, will allow us to increase again our capacity from what we mentioned in the prior press release.
That was EUR 1.4 billion at a run rate, at least to EUR 1.6 billion run rate at the end of Q1, meaning $1.9 billion in terms of run rate. This is the most important upgrade we want to show and we want to disclose. Lastly, for the new market. Leave the floor to Stefano Felici.
Yes. As far as market segments for next year, we don't expect major changes in the ratio between GPU, CPUs, HBM. There might be a slight correction in the CPUs, maybe a little bit more CPU demand related to AI, but this is still early to say. We don't expect major difference compared 2026 . For an emerging market that maybe I can comment is, you mentioned silicon photonics. This can be really considered a new segment. The situation here is that right now, there are solutions in the market that are more lab oriented. It's because the customers are approaching this type of test for the first time, and it's still a little bit unclear what will be the final strategy, how to test these type of devices in volumes. This year, I would say that it's still a characterization phase. Okay.
There are solutions there that are not for HBM and not for volumes. There are solutions more for characterization. What we are preparing for is a solution that can address high volume. Okay. We have already developed the core technology for this type of testing, and will be up to the final customer to decide when to switch from a characterization phase to a mass production, to test in high volume for these devices. Probably this will happen during 2027, the start of these volumes, but it's still early to predict exactly how much will be the business. For sure, we will update you in the next calls.
Okay. Very clear. Thank you.
Thank you, Alberto, for your question. We now have a question from Oliver Wong. Please, Oliver, go ahead. Oliver, can you hear us?
Hello. Hey. Thank you for taking my question. I have a question about HBM. You mentioned that the next year you don't expect a significant change in composition of different chips. Has there been a delay in HBM revenues of any sort?
No, there are no different comments compared to prior press releases. We still expect for 2026 to have a very small portion of revenues in the range of EUR 10 million, more or less, no more than that by the end of the year. Again, we have been qualified by one of the three customers, and the situation is pretty stable. We are still in the qualification phase for the other two main players of the segment. We do not expect to have additional information by year-end. We remain stable and consistent.
Sorry, I meant for next year, for 2027. Since you said you're expecting a big change in the composition of types of chips, I guess that means HBM will continue to be a pretty small part of the revenue. I was wondering if there's any sort of delay compared to what you previously may have expected.
Yeah, got it. Thank you. For 2027, in terms of magnitude, is still to be defined. As you know, the definition and the adoption of the technology is yet to become what is more suitable for the production of Technoprobe that use a vertical MEMS product rather than cantilever product. The adoption of this technology is what drives the volumes and the potential campaign assigned to Technoprobe. For 2027, we don't have a particular visibility on that. It doesn't depend on us. What it counts very much for us is to have a technology ready for the moment when it will be necessary. We are very committed on that, and we know that our product will be successful for the customer. One of them is already adopting in terms of, let me say, prototypization, and it works.
We are very confident it will be successful also in 2027. Again, it's still too early to define which will be the volume for next year.
Yeah. I want to add on this. Stefano mentioned about our approach, which is vertical MEMS approach. We're leveraging basically our technologies used for logic. This is important to say because the vertical, we think that, also the HBM will evolve and can be customized to perform better in the future. The better technology will give more flexibility basically to the HBM design. The other very important things for us is that we want to base this technology on our best core technology that can also produce the same level of profitability that we have also with logic. This is typically in also the past, the memory market was not the best one as far as profitability, but again, because of a different application.
Our main point is to address this market with a better performance, with a different technology that can produce the same result of the technology we use for our logic. As said by Stefano, doesn't depend on us when the adoption will happen. We already going in the right direction. There is this interest in this technology, but it's still early to say when exactly it will be adopted in mass production.
Thank you, Stefano.
Thank you, Oliver. We will now move on to the next question. The next question comes from Alexander Duval. Please, Alexander, go ahead.
Yes. Hi, everyone, and congratulations on the strong results. I had a couple of questions. Firstly, you have raised your CapEx investment this year. I wondered if you could give us some thoughts on the level of investment you might do beyond 1Q 2027. Clearly, there's an uplift you're talking about in your cumulative spend up to that point, would be very helpful to get a sense given the extremely robust AI market. Second of all, just to double-click on that question about differentiation within memory, I wondered if you could elaborate a bit more on the extent to which technological differentiation that you offer will allow you to surmount some of the sticky relationships one would think that competitors would have in the memory space. Thank you very much.
Thank you, Alexander. Let me take the first point about the investments. Your read is very correct. EUR 350 million to be spent in 15 months, starting from January 2026- March 2027 is very challenging. We are literally committed to deploy all these investments. Right now, at the end of June 2026, we spent approximately EUR 90 million. We have EUR 250 million, more or less, left to be spent in the next nine months. Very challenging because most of this investment relates to machineries and equipment, depending on the ability of our suppliers to deliver what we have ordered to them. It could be the case that, if not completed by March, something can be delayed for one month or two months. Overall, the commitment is to spend this big amount for this capacity increase.
This is exactly to deploy the run rate capacity at EUR 1.6 billion. That is already more than double, largely more than double compared to one year ago. Overall, do not forget the fact that the expansion plan that was launched in the fourth quarter 2025, is being implemented through two parallel work streams. Not only purchase of equipment. Each of the two work streams has a different execution timeline. The first one that is the most important for the moment in the short term has been quickly delivered, and that is quickly delivered great results. Involving greater execution complexity and focused on optimizing the existing manufacturing processes, including redesign, production workflow, reducing machine processing time, increasing the number of daily shifts, reconfiguring selected factory layouts, and all to accommodate additional automation equipment.
The resulting capacity today is expected to progress steadily month by month through the end of 2026. We will not have a big bang at the end of Q1. This is what is concretely visible right now in our numbers. We are growing consistently quarter by quarter and month by month. The second work stream is, of course, the new 4,000 sq m manufacturing facility, which is already underway and expected to be completed by the end of Q1 2027. This project has a longer implementation timeline. It carries lower execution risks. At the greenfield facility can be designed and configured in line with the group's specific operational requirements. The contribution to the productivity and capacity will only materialize once the facility become operational.
I take the second question. I hope not to be too much technical to answer to your question, but feel free to ask again more question if you need. The simplest way to answer your question is to check also and start from the beginning what happened and tell you what happened with the logic devices. If we go back, 15 years, 20 years ago, all the logic devices were tested, most of them were tested with cantilever probes, not vertical. I'm talking about logic now, not memories, but then I'll make the point why I'm saying this. Focusing on the logic, they were tested with the cantilever probes. Cantilever means probes coming from the side, and they cannot cross each other. You basically with that type of geometry, you can contact just a line of paths. Okay. Typically in the periphery of the devices.
There are still some devices with just pads on the periphery of the device. Cantilever probes can, of course, contact this line of paths. This was for logic, became a very big technical constraint for the performance of the chip. Because the designers needed more paths to be probed in the area, and they needed to maybe to put some power supplies in the middle of the device or ground in the middle of the device to get more better power integrity, signal integrity, to get a better performance. What happened is that for the logic chips, the designers had to put paths covering the whole area of the chip, paths everywhere. That was really the reason, the technical reason why cantilever technologies couldn't at that point be suitable anymore for the application.
Because if you have paths everywhere also in the middle of the device, everywhere, the only way to contact those paths is in vertical. That was really why the vertical technology basically was the winning technology at that point. From cantilever, everybody switched to vertical, and basically 99% of the logic market and any logic chip is tested with vertical technology because of this reason. It is more flexibility for designers to cover and to put all the paths wherever they want, basically. Okay. Now coming back to memory
Typically, the memory chips, the memory devices, were simpler than logic devices. Till now, the pads, to be contact for testing, are still in line. Can be two lines of pads, very similar to the configuration that I just described for logic. This allow for, as far as the geometry for a probe card, still allow to reach these pads using micro cantilever probes. From a side, you can contact a line of pads. Again, our vision, but is also supported by customers, is that would be much more beneficial now to have pads everywhere, because now we are going the direction where also the HBM is not anymore a very simple memory chip, but is a very high performance chip. Everybody want to push over the limit the performance of these devices.
This can be really the reason why at certain point, the designer will prefer to change the layout and put pads everywhere, and at that point, vertical will be the only solution. I don't know if I hope this can explain the differentiation here of the technologies. This clarify your question?
That's super helpful. Thank you so much.
Thank you, Alexander, for your question. The next question now comes from the phone number ending in 893. Please go ahead. I see that you're currently on mute. Please remember to unmute your line. Thank you.
Hi, sorry about that. Some technical difficulties. It's George here from Deutsche Bank. Thanks for taking my questions, congrats on a great quarter. I have two questions. Just firstly, on the new guide for this year, I think maybe you touched on this earlier, just to double-click. If I take your implied Q4 guide, even at the top end of the sales range, you're growing maybe 3% sequentially in Q4, your implied margin in Q4 at the high end of the margin guide for the full year, I think the Q4 margin's around 49%, which is down from Q3 despite the implied higher sales. Just wondering how to think about the drivers here for the margin specifically. Maybe it's visibility, maybe it's a degree of caution, but any commentary there would be helpful. I have a follow-up.
Thank you for the question. Basically, if you see the profitability guideline or trend we show it, this is a full year profitability that is not in a decrease compared to Q3. Because you should remember that Q1 was a much lower profitability on the EBITDA, you have to accumulate all the profit of the year. We expect for the moment to have Q4 in a range between second quarter and third quarter revenue. We don't know yet which will be the guideline for that, the real orders backlogs we will see in Q4, we expect that in a range between Q3 and Q2 2026. If the range will be that, we will have a similar profitability equal to the same quarters we have just passed. There could be some small dilution in the second part of the year, not significant.
In general, the current mix of customer we expect in the second part of the year will be more addressed with higher agency fees or distribution fee that could impact a bit on the profitability. There will be some probably additional labor award for our colleagues, depending on the results that will be achieved. There will be some R&D and SG&A extra charge related to the Chinese factory ramp-up, especially. As you know, we are building a new factory in China to serve the domestic market, and this is ongoing, and the vast majority of the expenses for the ramp-up are expected in the second part of the year, Q3 and Q4. We talk about EUR 3 million- EUR 4 million ramp-up cost in the second part of the year.
All these elements combined together can give you a very small dilution on the profitability, but the profitability overall across the year will be consistent with the second quarter, with the level of revenue of the second quarter, and consistently Q3 with the level of revenues of Q3.
Brilliant. Thank you very much. Just secondly, wondering if you have any update on the custom ASIC market share or any further engagements with customers. With these new markets like ASICs, HBM, silicon photonics becoming more material, I guess in 2027, maybe 2028, how should we think about the potential for margin dilution or margin gains from these new products? versus maybe what you're doing today in the AI GPU, I assume is driving the margin today.
Let me answer from the financial standpoint, then I leave the floor to Stefano. From a financial standpoint and profitability, we do not expect to have any dilution on this product, unless maybe for the initial ramp-up of the industrialization. As Stefano mentioned before, for example, silicon photonics is still on a lab phase, so the cost is higher and the profitability is lower, but we talk about very little volumes. At their mass volumes, we do not expect any dilution on profitability in any of our products.
I would add that for especially for AI products, we're still investing, developing, and improving our technologies. As we always said, basically every generation of GPU also we update our technology, the new generation of technology, probing technology. Typically, our model for profitability is very good. It's not the same product. It's always an updated product or even completely new product. We need to address different challenges year-over-year. The power of this chip is become very high, so there are other challenges we need to address, and this is a very good opportunity, of course. As said by Stefano, this can give us the chance to be pretty stable in the profitability, even grow even more with new technologies. For market share, let's say we prefer not to give exactly the percentage of each segment.
I can tell you, by the way, that the AI, all of these, the mix of products relating to AI, I mean, GPU, CPU, ASICs, now represent more than 50% of our total revenue. It's very healthy. We are growing. The numbers are there to see. It's our record quarter and Q3 even more. I can definitely tell you that we are leading the market. We are very strong position, very strong market share in each segment.
Brilliant. Maybe if I could just add one quick question to my two questions. Just touching on the question earlier around the CapEx beyond Q1 2027, obviously, you raised to EUR 350 million out to Q1 2027. Are you thinking about going beyond this currently, beyond the 140% capacity increase? I assume you're speaking to your customers about 2027, 2028, maybe forecasts even beyond that. I'm just wondering about CapEx beyond the Q1 2027.
Thank you. This is another very key point for Technoprobe. We, of course, have a lot of conversation with our customers. Everybody knows that everyone is investing in capacity, not only in testing space, but also in the semiconductor industry in general. We do not give up on investments. The big investment we are doing right now will allow us to serve the current production, the current demand we have now in our estimates. What we expect on 2027 and 2028 is to make another round of investment, a significant round of investments. The amount and the timing of this new capacity investment is still to be discussed and approved by the board. For sure, it will be interesting to see what we will be able to deploy for 2027 and 2028.
Of course, the capacity will increase even further in the next couple of years. Thanks also not only to the new investment, but to the increase of internal efficiencies.
Yes. I want to also say that typically, the way that we plan is incremental investments and CapEx is not in the reaction mode. We just don't wait and see what happen. Typically, we get very good feedback from our main customers every quarter. We are in front of them, all the main, the major customers and founders, and they really tell us what to prepare for the future. They give us good visibility. We're not talking about orders or commitment, but pretty good visibility about what to do. I think this is a course when there will be the next round of investment, we will for sure immediately share with you what we're going to do. We are, I would say, ahead of the game in order to be prepared for the next wave.
Brilliant. Thank you so much, guys.
Thank you. We will now move on to the next question. The next question comes from Harry Blaiklock. Please, Harry, go ahead.
Hi there. Thanks for taking my questions. First one is just around gross margins. I know they've been very strong. I wanted to ask specifically how much of that is kind of related to customers shifting over to turnkey solutions? I guess more broadly on that, are you seeing more customers shifting over to turnkey as testing complexity is increasing?
Turnkey solution is, of course, the most profitable product we can sell to our customers. This is very exposed in the second quarter, especially in terms of percentage on the total of our revenues. We cannot, of course, disclose exactly the percentage of our mix. For the two main customers that relate to GPUs is, of course, the most important portion of our revenue. This is another reason of our gross margin increase across the quarter.
Yes. Basically, you can see from all our main customers, the announcements they are making, about very big, very fast acceleration even starting from this year. Even not expected for them as well. What happened is that all of these main players were very hungry for capacity, they would take any capacity available. The game here was how to grow this capacity as fast as possible. As you can see, we did a very big jump forward in Technoprobe, and I can explain this also, and this was said also in some past calls, that a big, big factor here was also the fact that we are very vertically integrated. This helped us a lot to be faster in growing the capacity, faster than other players, I would say. To explain this, maybe I can make a very simple example.
Imagine that Technoprobe is a race car and you want to go faster. How you do? Of course, you can modify the car, you can try to do something, but if you own the design of the engine and you are the one developing the software for the engine, and all the parts of the car are developed by you, of course you know what to do, and you will modify the car the way you want very quickly. This is really what is happening. We own and develop all the main equipment we use for our production, including the software for the equipment. We develop many. We are very vertically integrated. This allowed us to push on the efficiency and use our machines in a much better way. This you can see a very big jump in efficiency and the profitability.
Of course, to grow even more, you need more equipment, more CapEx and so on. The vertical integration helped us really to move faster here than other players, let's say.
Got it. You could say you're the Ferrari of the probe card world.
You said.
We are in time. This was important because right now every suppliers we see, this in our industry, they are short of capacity. No matter what supplier we are considering, can be a PCB supplier or any supplier of machines in the semiconductor world, typically now is fighting for capacity. Of course, owning the IP and the design of the machines and so on, it's easier, of course, to do things without relying on suppliers that also they have problems with capacity. Of course, if we can do by ourselves, we can push more and be faster. This was really an important point in our playbook.
Got it. I just had one last question, which was on kind of a clarification around the capacity investments that you're completing by Q1. I think you mentioned earlier on the call that it would be a run rate of $ 1.9 billion annual revenue. I don't know whether I misheard that, I guess an additional question on that, you're obviously talking about further capacity additions beyond that after Q1. Looking at that initial investment that's ending in Q1, how long would you expect it to take to ramp that to kind of close to full utilization?
When you mention $1.9 billion, I want to reiterate the message. This is U.S. dollar.
Okay.
This is EUR 1.6 billion, $1.9 billion . Just to clarify. The utilization, it depends who will answer to you because if you ask to sales, they will say, "Now we hope to have 100% utilization." If you ask to production, maybe they would say 80%. In general, it's not always healthy to have a full utilization. We believe that the utilization will be largely used by the end of Q1. With the EUR 1.6 billion run rate, we expect to be able to serve all our main customers unless something unexpected will happen for, I don't know, HBM adoption or silicon photonics adoption, or any other unexpected events so far. The run rate, the capacity will increase again gradually all over 2027. We have, for the moment, a preliminary plan of investments that will be, of course, disclosed as soon as ready, that will increase across the year.
We are very confident that our goals and our plans will be consistent with the demand growth.
Great. Thank you, guys.
Thank you. We will now move on to Giovanni Selvetti. Please, Giovanni, go ahead.
Hello, everyone, congratulations for the results. I think most of the questions were actually answered. Just maybe a follow-up on the questions from both Alberto and George about the implied Q4 numbers. If I didn't just say misheard previously, I think you mentioned also different in the mix in Q4 in terms of sales, I was wondering if you can kind of provide a difference in the gross margin between CPUs and GPUs, just to have an idea.
Yes. On the mix on sales, as I mentioned, we expect Q4 to have a decrease in terms of GPUs because the big campaigns have been already almost completed in the course of 2026, Their cyclicality will restart at the end of Q4 with the deliveries and revenues restarting in 2027. In the meantime, the shift will be more addressed to CPUs and ASICs, for which we don't see any particular difference in gross margin in profitability. The complexity of this different product is almost the same for the GPUs, volumes are a bit different because what we have experienced in the first part of the year is that industrialization of probe cards for GPUs have been massive, you have additional operating leverage when you produce more product of the same type.
We do not expect to have this similar mix on the last part of the year. We expect same volumes in general, composed by more designs, more projects. That's why the profitability overall could be Not the profitability, sorry. The operating leverage could be a little affected, but this is largely expected in our model, this is also guided at an increase at the fiscal year overall gross margin up to 200 basis points on average, despite this cyclicality. For us, it's a very encouraging second part of the year.
Okay. All clear. Thank you very much.
Thank you, Giovanni.
Thank you.
I see that we have a follow-up question from Oliver Wong. Please, Oliver, the floor to you.
Yes. Thank you, guys, for taking my follow-up. Just wanted to piggyback on, you guys sounded pretty confident on continuing to increase capacity. I was wondering if maybe you could share a little teaser on what's driving that. Obviously, GPU volumes are going to be strong, out of the emerging opportunities, would you be able to say if there are any particular ones, anything in customer conversation that's giving you that confidence to continue to expand capacity? Thanks.
What is driving is the AI growth that is really still predicted very strong course. This is not only so the expectation we get is not only a higher number of chip to be tested, but with a longer also test time. This is valid for GPU, CPUs, all the type of AI devices will grow in number of volumes to be tested and also in test intensity. This is really what is driving here the growth of this market. I don't know if this was the question.
Yep. Sounds good. Thanks so much.
Thank you, Oliver. As we have no hands up at the moment, I will now give the word back to the speakers for any final comments before bringing this presentation to a close. Thank you.
Thank you for everyone for joining us tonight. We're very happy to announce this record quarter. We hope to hear from you. The next call will be, we think, a very good call. Bye-bye.
Thank you.
Thank you. This presentation will now come to a close.