Session at the BofA Global Tech Conference. I'm Vivek Arya from BofA Semiconductor, Semicap Equipment, research team, and really honored to have the team from Microchip join us this afternoon. Steve Sanghi, Chairman and CEO, and Eric Bjornholt, Vice President and Chief Financial Officer. We will start with some opening remarks from Steve, then I'll go through my questions. Please feel free to raise your hand if you'd like to bring anything up. With that, really warm welcome to you, Steve and Eric. Really happy that you could join us at the conference. Steve, maybe let me turn it over to you.
Thank you, Vivek. Before I begin, I wish to remind you that during this presentation, we may be making some projections and other forward-looking statements. These are predictions and the actual results may vary materially. I refer you to our filings with the SEC regarding some important risk factors about the company. What I'm going to talk about is two things which were both in the press release yesterday. You must have read them, just going to expand on it a little bit more. The first is the data center. We have been getting numerous requests from investors to give them more information regarding our data center business. The challenge is that in our end market breakdown, we have been breaking out data center and compute as one of the end market, which is about 18% of our business.
On a calendar year basis where we provided the information based on our calendar year 2025 revenue, that 18% comes out to be about $787 million. There's a ±2% error band because there's a lot of suppositions and estimates in there. Out of that business, the challenge is that we have a Data Center Solutions Business Unit, which is pure data center, 100% of their products going to data center. That business unit produced $302.7 million in calendar 2025 out of that $787 million total. That doesn't mean that our data center business was only $302.7 million. There are a lot of additional products, some of our microcontrollers, analog, timing products, security products, memory products, and others, which go into multiple markets, data center being one of them.
Many of these products are catalog products that also get sold through distribution and contract manufacturers, which will bundle the total sale on a given product. Some portion is going to data center, but some could be going to industrial. There are a lot of estimates involved, and our total data center business is somewhere between that $303 million-$787 million. Since that is much more difficult because of going into multiple markets, what we decided to do is give you our revenue for just the Data Center Solutions Business Unit, which is 100% into data centers. That was $302.7 million and expected to grow to about $500 million in calendar year 2026.
We'll continue to do more work on it and see if we can take the difference between $787 million and $303 million and break it out further into what is in compute and what is in data center. We were not able to do that by yesterday. That was the one part of the press release, and as I had various one-on-ones, I think some people were still confused where they thought the total size of our exposure to data center was 302.7. That is not the case. That is only the Data Center Solutions Business Unit. The second item in the press release was pricing. Till last week, we were essentially saying we're not increasing prices to our customers.
Essentially, some of the operational cost increases we have incurred with people, chemicals, gases, transportation, energy, gold, copper and others, so far, we have absorbed them without giving a price increase to our customers. Despite all that, our gross margin in the last year has gone from 52%-61.5% the last quarter and we're guiding to about 63 point, I don't know what the exact guidance is this quarter.
62 and three-quarters.
62 and three-quarters this quarter. That's happening despite us absorbing some of the cost increases. So far, we were able to push back on our suppliers, foundry and OSAT and other suppliers, that we can't take a price increase, and we didn't incur them. Now it's becoming much more difficult because essentially, they have been able to pass those price increases to all of our competitors, and they're passing those price increases on us, and there are dates and times and percentages by technology, by foundry, by OSAT, that'll be coming up where our costs will be going up.
Finally, and we've been doing this analysis for some time. A decision was finally made two days ago, Sunday evening, in a worldwide call where we put all this data together that we have been putting it together for some time and made a decision that we will have to increase prices to take the cost increase we're incurring and margin it up and give it to our customers. Now, at this point in time, we have no estimates on the amount of cost increase or price increase on our customers, which products will see it and which customers will see it, how we will deal with distribution, how we will deal with direct customers. All that work has to be put together in the next two to three weeks, and then our salespeople and business unit leaders will take those price increases to our customers.
We don't really have any more detail on that. We will not be telling you the average price increase for obvious reasons. One, that's a competitive information. Second, if a customer knows what the average is, then nobody will take more than the average. That will create a problem if we want to exempt some customers where we are still repairing our relationship. We have to make that up by giving more cost increase to somebody who's a non-partner. If you come up with an average number, then nobody will allow you to do that. We will not be disclosing how much price we are increasing and just execute it in the marketplace. With those two updates, I'm happy to take any questions.
Thank you, Steve. Maybe continuing on that pricing theme. As you mentioned, Microchip has been somewhat more resistant to taking pricing, whereas many, I would say, in fact, all of your analog and kind of power semi competitors have taken pricing. How much of that, Steve, was just because you were still in the customer relationship-repairing mode? Now when they see this, what do you think is going to be their reaction? Does it impact that progress that you have been making?
I think you have to look at through cycles. In the last cycle, we took pricing up a little too much, one time longer than probably they did, and we were much harsher on our PSP program, in the non-cancelable, non-returnable, non-reschedulable, than our competitors were. We upset some customers where our competitors treated them better. Now, fast-forward in the last year and a half, we have done very good with our customers, repaired a lot of relationships, and this time, our competitors took the very first opportunity and raised the price on them while we told our customers that we're absorbing the cost increases and you're not getting a price increase. Net flow of the customer has been towards us. We won many, many designs and customers don't like our competitors today, they like us. Some of them have done two price increases already.
We haven't even done one. When we go back to them saying we have held a price increase now all year and we have treated you better and all that, and they have seen two price increases from the competitors. When they hear from us that we're doing a price increase, number one, they will not be surprised, and if the amount is reasonable, then I think they would be understanding. Our relationship will still be very good.
Got it. Just to confirm, this is not happening because of shortages, right? I mean, your lead times are still at reasonable-
It has nothing to do with shortages or excess. It has nothing to do with increasing the price because we can. I mean, if we really wanted to, we could increase the price more. It's just purely have to driven with, we don't want to gouge the customers, but we don't want a headwind on our gross margin. We'll take an estimate of our cost increases, margin it up, and give it to them and be able to explain it.
Got it. Historically, Steve, has this resulted in some kind of pre-buying or extra buying, anticipating that, "Oh, look, there is this another price increase coming, so let me buy more now?" Or do you think-
Depends on how you implement it. If you wait too long, give them too much of a time to buy the product, they will try to buy some product in advance at the lower prices. We will increase the price on the backlog.
Right.
Any incremental orders we see over and above their run rate, we'll schedule them at the new price after the price increase was effective. Therefore, there will be no incentive, and if anybody places the order, we'll schedule them after, and it'll be all at the new prices. There should be no games, really. I mean, it shouldn't affect our revenue. There should be no pulling effect-
Got it.
Because we get to schedule it, and just based on the lead times, today have seven to eight weeks. The window we'll probably give them will be shorter than that. Essentially, all that will fall outside of the price increase.
Got it. Makes sense. Is there a scenario where this is a tailwind, albeit a modest one to gross margins, or do you expect it to be neutral, or how should we think about the benefit?
A couple of things. We have absorbed some of the internal operational cost increases thus far.
Right.
When we're increasing the price, we're including in it. There is a little bit of the recovery that we have incurred the cost, and now we'll pass that on, which will make it slightly accretive. The second part is, our target is to be margin neutral. You can never hit the bullseye.
Right.
I don't want to have the risk of being on the left side, so we want to be on the right side. Given at all, it will end up being slightly accretive, but only because of we can't be so precise.
Got it. This kind of bridges into the next few questions about your data center. Is it fair to think that most of your data center sales are coming from parts that you get from foundry? If that is the fastest-growing part of the business that you are always exposed to this inflationary impact.
Well, inflationary impact is not only on foundries. We are seeing the similar inflationary impact inside because foundries, their cost is going up on their people and chemicals and gases and energy and transportation and all that. That shows up in their wafer cost, and they give it to us at a higher price. Our internal manufacturing is doing the same. They have the same inflationary cost increases. In answer to your question, our Data Center Solutions Business Unit products are all outside.
Right.
Not our total data center exposure. Some of our analog products, our microcontroller products, some of our security products, timing products, that go into data centers are produced inside.
Got it.
Yeah.
Okay. The growth rate that you described for the data center business, I think 60% +. How do you benchmark that? Do you think it's the right growth rate? There's such a wide range of growth rates that we see across the data center. Can that be sustained over the next few years? How should investors think about just your data center? First, maybe let's start with what is in that data center number. What specific end markets, applications, and then the growth rate?
In the Data Center Solutions Business Unit, we make three product lines with the fourth one coming. The first product line is our storage controllers, storage accelerators, and storage expanders. In that product line, we have about 3X performance advantage to our nearest competitor, so it's very state-of-the-art and it's doing very well. The second product line in that is PCIe memory controllers and CXL memory controllers. There again, we're kind of leading the industry. The third product line is PCI Express switch. Now, in PCI Express switch, we did very well when we bought the business from Microsemi. They had done the Gen 2, Gen 3, and Gen 4. I'm sorry, Gen 2 and Gen 3. Gen 4 happened after we had bought the business, and Gen 4 happened on my clock when I was CEO last time, and we did very well in those.
We screwed up on Gen 5. That didn't happen on my clock, but it happened under Microchip clock. We were two years late to the market on Gen 5, and therefore, all those Gen 5 designs went to the competition. When we eventually came up with the Gen 5 product, that again happened in the last year, we won a lot of the designs again, but we got a fairly small share in the second source position, not in the primary position. The commitment we made is that we got to regain our mojo, and that's why we produced a Gen 6 product on time on one of the best products in the industry. The only 3-nm product, significantly higher performance than the competition, and 30%-40% lower power. That part is winning designs like crazy. Every single customer who did business with us is re-engaged.
We have publicly said in the last conference call, we have won six designs, and now we have won two more, so we have eight designs on Gen 6 switch, and we have one design on the retimer. Retimer doesn't go to production till January, so this is very early win. I think if we had not lost all that on Gen 5, then it will be a little bit easier to make assessments of how fast it will grow or what it will accelerate. We know the market is very large. The upside is just totally incredible. If we keep winning these designs, they go to production next year, we can acquire the capacity, then this could be very big.
I think we are cognizant of the fact that we are coming from behind, having missed the Gen 5 and we're competing with a formidable competitor, with a B name in it. We are just a little cautious in trying to forecast anything. Now, one other thing that's happening on the switch side is, and you may have heard about it or read about it, During the training phase of large language models, it was the era of GPU, and the GPU to CPU ratio was as much as 10: 1. Now as you convert from training phase to a inference phase, it is favoring the CPU. GPU to CPU ratio is going as much as 1:1 , and I'm even starting to read one GPU and two CPU.
PCI Express connects a GPU to CPU, CPU to CPU, GPU to memory, CPU to memory, CPU to anything, a RAID card, a memory controller, a GPU. As the new industry is calling it CPU renaissance.
Right.
If there is a new CPU renaissance, then you're seeing Intel is capacity constrained, others are too. As these CPU volumes explode, it is very, very positive for PCI Express because GPU to GPU connection is through NVLink.
Right.
GPU to CPU connection is through PCI Express, which is standard industry protocol. The market strength and conversion from training to inference favors PCI Express. If that market grows more than anybody expects, then we should really get a fair share of the market. I think demand is going to be strong enough with probably overall TSMC capacity limited, that everybody will get their fair share to what the capacity they can acquire.
All right. Is there some simple math, Steve, that you have in mind? If let's say the CPU market, people used to talk about $60 billion, then they went to $100 billion, now the new number is $200 billion. What does that tell you about the PCI opportunity associated with it?
It is probably calculable. I haven't done that yet. I think that CPU renaissance is relatively new.
Got it.
I don't think I've seen the PCI Express estimate because of that change. That will only describe the TAM, total available market. It still wouldn't tell us what we could win. Like I said, we are making up the loss.
Is it as simple as your share of 3-nm allocation at TSMC? Is that the first order way of looking at your share, or there are other factors?
What we need on 3-nm is 0.1% or 0.2% of what Jensen needs. I just tell them, "Give me all I need," and Jensen even won't feel the change, and it's working so far.
Is it deterministic what your share is?
It's not deterministic. The part goes to production at the end of this month. Winning eight designs before the part is in production is unprecedented.
Right.
Because the part is that good and it's doing that good in customer tests when they put them in the board. We think we have a winner on our hands, but we are shy in forecasting till we actually start to see these wins going in production and then multiplying like crazy.
Got it. Is there a scenario where these kind of growth rates are sustainable or one could think about?
I hope they're not only sustainable, but they accelerate. I'm not really willing to put a number on it yet.
Okay. Understood. Next thing, Steve, is just the broader analog cycle. Where are we? Usually these cycles have sort of the inventory replenishment phase and then the end demand improvement phase. Where are we? Just because the size of these markets, especially in the data center, are getting so big, so it's hard to separate out the cyclical recovery from what are the secular drivers.
That's a very good question, and I've been talking about it. The first phase of our growth in the last year has predominantly come from customers and distributors that were buying product well below their consumption level because their inventories were high. Asymptotically, they're trying to reach where they're buying equal to the consumption level. That fed the first part of the growth, which is typically inventory depletion-driven semiconductor cycle growth. What's being layered on it now is two of the things. One is the new designs that we have won in the last two years, and even in the data center, since we were so late to Gen 5 PCI switch, we lost all those designs.
When we came back with it, we won a lot of designs where we are in the second source position, and so that PCI Express Gen 5 is ramping, waiting for the Gen 6 production to start. We're getting some growth because of that. Now the third phase really is the innovation-driven growth. There is innovation happening in four of our end markets. It's happening in, obviously, data center, it's happening in automotive, it's happening in aerospace and defense, and it's happening in industrial. I'll spend a minute on each of them. Data center, we have talked about. Lots of innovation going on. It should grow. In the automotive, so number of automotive units per year grow in low single digits. That has never been the story. The story is the consumption level inside.
On some of the high-end automotive cars, we have 81 chips from Microchip in one car and 63 chips in another car. There's a large amount of Microchip content in the cars. Today, there are so many different protocols inside the car. There is a Controller Area Network . There's a CAN bus, there's a LIN bus, there's an entertainment bus called the MOST bus, there's USB connection inside, there's Ethernet, there's RS-232. In the past, all these things don't need to talk to each other. Your microcontroller or tire pressure sensor doesn't need to talk to anything. Today, you could press a button and say, "What's my tire pressure?" It will show you a picture of your car, four tires, and tire pressure in each of them. My Tesla does that.
Now the car is talking to every node that is outside in the car. In the current car setup, you have to go through bridges to convert from USB to Ethernet, Controller Area Network to USB or to Ethernet, MOST bus to Ethernet, and these bridges are expensive and bulky and extra chips that need to be added. Where the cars of 28 and 29 and 30 are going is to consolidate all those protocols into a single Ethernet-based network requiring no conversions. Everything is basically that. We are leading that charge. That standard is called a T1S, which run on a single twisted pair, Ethernet wire, not the eight-strand wire you have in your office, but two-strand wire, single twisted pair. That should drive significant growth, and there's a lot of innovation happening on it.
We talked about Data Center, automotive, and quickly the third one is aerospace and defense. Aerospace and defense has aviation, has military hardware, and space. We're growing in all three, like Boeing is building planes like crazy again, where MAX 7 production is back up. Military hardware, they're trying to replenish all the hardware they have used in the wars. Trump is asking primes to quadruple their production. That's a long-term growth for us because primes are adding capacity with brick and mortar and all that, and that revenue should do very well. Space, U.S. as well as the world, has a new fascination with space, and the fascination that hasn't been seen since the Apollo days. We should do very well with all our space content. The fourth final market is industrial, with what's happening in industrial with onshoring.
When the factories are being onshored, they're more highly automated, there's a lot more robotics in them than the factories they're moving from in China, so there's more content from us. Sensors, microcontrollers, PCI Express, and all these parts in the factory, and then all our parts going into humanoids and robotics and moving factory line and all that. I think all that is innovation-driven growth that's layering on the top of the inventory-driven growth, and therefore future looks bright.
Got it. One maybe question on margins. I'll bring Eric into the discussion. Gross margin, you mentioned 62.75% in June. The target is to get towards 65%. I imagine most of that is just an underutilization kind of recovery.
Yeah, let me have Eric address that.
Sure. Last quarter, or the March quarter, we had $46.6 million of underutilization charges, and if you add that back to our gross margin last quarter, we're essentially at 65%. It's just growing back into our capacity. It's not all going to happen at once. There's going to be a steady state that we're increasing production in the factories. We're essentially ramping our Oregon factory as hard as we can this quarter. We are shipping at a level that is well above what we're producing at, and so that's why inventory has been coming down, and we want to make sure that we bridge that gap so inventory doesn't get too low. Those underutilization charges will get much, much lower by the end of the fiscal year.
Got it. What about the OpEx leverage side of it? Is there a certain revenue run rate where you can achieve your target operating margins also?
I think it very much depends on the slope of the revenue curve. Today, we've brought back a lot of programs that went away when we kind of went through the doldrums of the downturn. Bonuses are back to high levels, our variable comp programs, merit increases are back for employees. There's going to be growth in OpEx dollars, but a significant portion of that happened over the last two quarters as we are bringing these programs back. This quarter, we are guiding to OpEx as a percentage of revenue at about 29%. Our long-term target's 25%, and you should expect as revenue grows, that OpEx is going to grow at a slower rate as we drive towards that model.
Got it. Last question, Steve, maybe on, we always kind of look back at history and say, look, if the last peak of Microchip was here, now you're still 30% below that. Many of your peers have caught up now. Your last peak was kind of unique because of its own characteristics. What is the right way to frame what your growth trajectory? Is looking at the last peak even useful? What is the right way to think about the next?
It is not useful. In the normal past semiconductor cycles, you could do the analysis of any end of the curve, you ship to higher, and then that has to be below, and then when you consume it, then it returns back. That analysis is much more difficult to do from the last cycle, because when we were shipping at the peak of the cycle, we were shipping hundreds of millions dollars into inventory. Last peak, really, it was GAAP revenue by the definition of GAAP, but it was not going into consumption, it was going into inventory. Customers, in many cases, didn't want the product, but because of the LTSAs and PSP and all that, we jammed that product down their throats.
Right.
Therefore, that peak was very artificial. Secondly, in the history of semiconductors, other than the memory, there haven't been price increases, right?
Right.
I mean, microcontrollers, analog, you don't go through price increase cycles.
Right. Always deflationary.
Price goes down low single digits every year for the last 40 years. Therefore, the analysis is apple-to-apple. When there are so many price increases in the post-COVID cycle, then when you do the revenue-based analysis, it gets distorted. You kind of got to go to the unit-based analysis. Even in a unit-based analysis, the mix shifted a lot because during post-COVID, people gave them the low end of the businesses, so they can take their limited capacity and give it to the high end of the businesses. In high end of the businesses, there are more dollars but less units.
Right.
The unit-based analysis becomes very difficult. I think pretty much in the last year and a half, every analyst or investors that have done that analysis on Microchip was really disappointed, because that analysis didn't work.
Right.
We told them that it won't work, but nobody listened. Don't do that.
What is the right analysis then?
There is none. No.
How do you know whether you're at the right trend line?
How does it matter? We're growing from here, and take the number and look at the opportunity ahead of us with all these four markets I talked about, and that's where it is. I'm already saying that the inventory depletion-driven growth pretty much comes to an end in the next quarter or so. Rest is an innovation-driven growth. Where was the past peak? How does it help you?
Right. Makes sense.
It only confuses you.
Right. Okay. On that note, thank you so much, Steve. Thank you, Eric.
Thank you.
Really enjoyed that discussion.