← 返回

2026.10.07 | TB | Networking | Amphenol: Connecting the AI Data Centre

原标题：Amphenol – Connecting the AI Data Centre

#networking #datacenter

Third Bridge · 2026-10-07

Analyst-led • 1 Oct 2026

Companies: NVDA US, APH US, AVGO US, CSCO US, GLW US, DELL US

Matthew O'Connor (MO) Advisor, Sports & Entertainment at North AI

Colin Duncan (CD) — Third Bridge Sector Analyst

Agenda

Key Insights

Introduction

CD [00:00:04]: Welcome to Third Bridge's Interview entitled Amphenol – Connecting the AI Data Centre. I'm Colin Duncan and I'll be facilitating today's Interview with Matthew O'Connor, former industry Adviser, AI, Emerging Markets Business Development at Amphenol.

Matthew, before we start today's Interview, please state I agree or I disagree to the following statement: You understand the definition of material non-public information and agree not to disclose any such information, or any other information which is confidential, during this Interview.

MO [00:00:30]: I agree.

CD [00:00:31]: Thanks, Matthew. Could you start us out with a brief introduction to your background?

MO [00:00:37]: My background is predominantly in the technology space. I was a go-to-market executive at a global system integrator called Dimension Data for many years. Dimension Data was a subsidiary of a Japanese telecoms company called NTT Group. I had several roles following Dimension Data's integration into the NTT Group family. I was a Corporate Vice President that looked after large deal pursuits. I worked in corporate M&A, and I also worked in the innovation business within NTT Group's international business. Then when I left NTT several years ago, I set up an advisory business, which mostly advises US public companies. One was a joint venture that CommScope held with Charter Communications called ActiveVideo Networks. Another is Cognizant. Then more recently, I advised Amphenol or subsidiaries of Amphenol within the optical business on the go-to-market strategy.

Questions

00:01:37] How has Amphenol's business and competitive positioning evolved over the past 2-3 years, particularly as IT datacom has become a much larger part of the company?

MO [00:01:56]: My involvement in Amphenol coincided, I guess, with the kind of the middle phase of this growth around, ie, datacom, so AI datacom, sorry. Generally, it was a diversified connector company across a bunch of different areas, including the likes of automotive, infrastructure, industrial as well as the traditional telecoms business. I would say today, it's become a leading part of the supply chain for the AI infrastructure industry, which explains the rapid growth in not just revenue related to IT datacom, but IT datacom as a proportion of its business. Then specific to that area of business, it's well documented that it's operate a holding company model, so it's highly acquisitive. They have moved Amphenol from selling pure-play components, so through OEMs or ODMs, but also perhaps more recently in the acquisitions of the subsidiary businesses that I was involved in that they've moved into a more direct to data centre operator type proposition. That's the CommScope cabling acquisition and also the acquisition of Halo, which is a compatible optical transceiver roll-up, which they acquired several years ago. That's a major difference because it gives Amphenol a position on both sides of the data centre boom.

00:03:40] Amphenol's organic revenue grew 30% in Q2 2026, including 63% growth in IT datacom. How much of that performance reflects underlying market growth vs increasing interconnect content and market share gains?

MO [00:03:57]: I think content is important because every time there's a reference architecture enhancement that moves up in speed and power density, it does mean that the per rack content is going to increase. That's having a material impact across the Amphenol supply chain. Primarily, though, I think this has to, we have to say that there are several macro market factors. One clearly is that there's the CAPEX of hyperscalers, which is driving Nvidia's business and pulling Amphenol along with it and therefore, the increase in the volume of GPU rack shipments. The other is capital. There is, not just in hyperscalers, the huge growth in capital that is coming from the Nvidia-related ecosystem around neoclouds, new infrastructure businesses, sovereign AI businesses, that's consistently driving opportunities across Amphenol's group of companies in the IT datacom space. I would say, you could say that there is, to an extent, a degree of risk because a lot of this is linked to the Nvidia ecosystem. The Nvidia ecosystem is becoming so diverse when it comes to the end sales opportunities, when it comes to the OEMs who are participating, when it comes to the ODMs who are participating, that does mitigate the risk a little bit.

00:05:38] What do you see as the biggest risks right now in Amphenol's IT datacom segment?

MO [00:05:43]: Clearly, any kind of pause on hyperscaler CAPEX is going to have a material impact across Amphenol's IT datacom business. That would impact both the components that would potentially also impact the cabling and the optics businesses. I would say, though, that there is a lot of possible mitigation around that because Nvidia's strength, certainly as we were seeing it in the market is so dominant that, and there is so much demand vs supply that even if that were to happen, that would not be an instant pause because there's been a proliferation outside of hyperscaler around AI infrastructure build, which I alluded to earlier, which would mitigate that, I think.

00:06:44] Given IT datacom represented 43% of Q2 2026 revenue and AI-related products generated approximately USD 2.5bn – around 30% of Amphenol's revenue – do you see any concentration risk? What non-AI businesses could provide meaningful diversification if hyperscale activity slows down?

MO [00:07:18]: I think industrial and particularly the move around power, electrification and so on, I think those areas would provide a degree. Perhaps you could argue defence is kind of a macroeconomic. Certainly, when I was there, there was lots of noise around the expectation that opportunities in defence would continue to grow. I'd say some of the traditional businesses, though, like automotive and the traditional telco, telecoms business, I would not consider them as reliable for any kind of risk mitigation.

CD [00:07:57]: Why is that?

MO [00:07:57]: Sorry, outside of electric vehicles, I think I would not see them as…Amphenol has the ability to then cross-sell some of its components with EV, which they didn't currently do in traditional automotive.

00:08:20] How is Amphenol's AI data centre business distributed among chip designers, hyperscalers, system manufacturers and data centre operators?

MO [00:08:41]: The answer to that question is related to how they are structured from a communicate. There's a division of Amphenol, which is called Amphenol Communications Solutions, which is essentially an aggregated portfolio of acquisitions that's loosely grouped into the different areas. The strongest relationships are with the likes of Nvidia and AMD and Broadcom. There's been a lot of time dedicated to providing, to building relationships to ensure that they are a core part of the supply chain. There are then also relationships which I was not super close to with the system manufacturers, so the Foxconns of this world. I think the hyperscaler relationships historically were indirect. What has significantly changed in that scenario is the acquisition, those two acquisitions that I mentioned earlier and to a lesser extent, the acquisition of the Andrew Group [sic], where effectively, they were buying in products that could be consumed by hyperscalers. Particularly CommScope had a bunch of both ongoing and historical relationships with hyperscalers that were kind of, I think you could say they were accelerated by the Amphenol acquisition.

00:10:19] Amphenol acquired CommScope on January 2026. How important has this acquisition been for Amphenol?

MO [00:10:27]: It gives them a hedge against the current routes to market, which are predominantly linked to Nvidia, or AMD, Broadcom, and also the OEMs. First of all, they have a direct relationship with the data centre operator, with the hyperscaler, which they didn't currently have. That's a material change in go-to-market strategy. You could argue they also have the same thing when they acquired Halo, but there's a significant difference. Halo is just over USD 700m. CommScope is close to USD 10.5bn. They were buying kind of a deeply penetrated business that was particularly strong in the data centre market. I would say this that CommScope has made some changes in its business prior to the acquisition, which is one of the reasons, which Amphenol have benefited from. One of which was they effectively split the enterprise business to form a data centre dedicated division, and that division has performed very well for some of the macro reasons we talked about earlier.

I definitely also think that the CommScope's performance since the acquisition has been enhanced by obviously, some of the things we talked about earlier, the continued growth in the AI data centre space, the continued increased requirements for cabling that come from AI clusters. Probably another thing that's worth noting here is that prior to the acquisition, and this is kind of lends itself to the value of Amphenol's holding model is because Amphenol has such a strong balance sheet position. It means that when they are buying companies like CommScope, who at the time had a degree of financial jeopardy, it increases confidence in these large-scale hyperscaler customers who will then be prepared to spend more with them because they have a greater degree of certainty on the supply chain.

00:12:44] Has the CommScope acquisition created opportunities to cross-sell products from the broader Amphenol portfolio into data centres?

MO [00:12:53]: What I would say, in my professional opinion is that that's theoretically what would happen. In practical terms, Amphenol's model actually makes cross-selling a little harder to execute, certainly, when you're cross-selling effectively across the Nvidia boundary line. What I mean by that is if you're cross-selling at the component level, that's easier to execute. If you are then trying to cross-sell at the hyperscaler system integrator level, that's a little trickier to execute. The reality is that most of the companies that have been acquired that sit within that ACS unit are highly decentralised organisations. There is a very thin layer of go-to-market and sales, which sits above each of the business units, and there is not a coherent cross-selling strategy that exists across Amphenol as I saw it.

00:13:59] Is there any measurable evidence, such as design wins, customer concentration, capacity investment or cross-selling, showing that the CommScope acquisition is producing more than just financial stability?

MO [00:14:15]: No, I don't think there is. I don't think there is. I think a lot of this is positioning and a lot of it is increasing the confidence. You will see more of an acquired company's technical leadership team being more closely linked to that of major ecosystem providers like an Nvidia, for example, just because Amphenol is able to facilitate that conversation. I'd say that the revenue jumps that will come out of that will be very hard to trace back to those conversations and those introductions in my view.

00:15:02] On which level of the ecosystem does Amphenol have the strongest customer relationships?

MO [00:15:08]: At the component layer, at the component layer. CommScope has a strong hyperscaler set of relationships. Halo is smaller, and Halo's relationships tend to exist more at the channel, the traditional GSI channel system integration layer. I would say not just in terms of influence, but the components, particularly Nvidia, but not just Nvidia, the OEMs. Obviously, Amphenol supplies technology that's integrated into all of the major OEMs, Cisco, Dell, Supermicro, HP. I think the way these deals work when they get further on down the food chain towards, for example, a neocloud or a sovereign AI project is those companies backed by an Nvidia reference architecture are the ones that are dominating and the optics and the cabling are really when it comes to the Nvidia reference architecture. They're the only two major components in the networking bill of materials that you can decouple from that reference architecture. What Amphenol have very clearly done, in my opinion, is effectively tie up two major strategic assets in the only two areas where you can legitimately swap out technology when it comes to data centre networking.

CD [00:16:43]: Within the component layer, who has the greatest influence over Amphenol's design wins across the chip designer, the systems manufacturer or the hyperscaler? At which stage is the company typically specified into that architecture?

MO [00:16:58]: I'm not sure I'm fully qualified to give you a coherent answer on that. I can say anecdotally from talking to some of the sales leaders within Amphenol during my time there, it was the deep entrenched design relationships with Nvidia particularly, which were the reason why they were so strong. The second major reason was Amphenol's size and scale gives its supply chain advantages and the supply chain advantages obviously give it pricing power. Those two things combined with that kind of deep technical integration and a long-standing trading relationship are the three reasons why they're still very strong at the chip layer.

00:17:42] How dependent is Amphenol's AI opportunity on Nvidia-linked architectures?

MO [00:17:51]: I would say extremely, extremely. Certainly, in the field, when we were building, I guess, an early cross-sell motion that might include, for example, CommScope's cabling and compatible transceivers, it was very rare that we came across a neocloud or a sovereign that did not rely exclusively on an Nvidia architecture. The reality is in the field is that lots of these companies are, in effect, very new. They do not have huge deep technical teams, and they are very reliant on Nvidia's supply chain and reference architecture design, and that effectively pulled through both the optics and the cabling, but also the components there. I would say they're very reliant on Nvidia's architecture.

CD [00:18:54]: Do you think that Broadcom-based custom accelerators or other compute platforms could materially diversify that exposure over the next 2-3 years?

MO [00:19:06]: Yes. Yes, I do. Broadcom were also a customer. I think they definitely could. I guess the transitions would be what Nvidia do about that. Then when it comes to the optics themselves, what does the move towards co-packaged optics mean for some other areas of the business. Yes, I think that is possible.

CD [00:19:41]: What would need to happen for the diversification to become meaningful, whether through a greater deployment of Broadcom-based silicon, additional Amphenol design wins or production ramps at specific hyperscalers?

MO [00:19:58]: Production ramps at specific hyperscalers are the things that we saw that typically move the market, yes.

00:20:05] Is Amphenol's interconnect content per system broadly comparable across Nvidia and Broadcom-based architectures, or does one architecture create a meaningfully larger opportunity?

MO [00:20:19]: I would say Nvidia creates a materially larger opportunity just because of what it meant in the broader AI cluster deployments. They increase the requirement for other Amphenol products. If you think that there's a significant volume of cabling increase, for example, that is a result of Nvidia architectures. There is a significant quantity of, or a movement towards 1.6Tbps optical transceivers that's a result of Nvidia architectures. All of these things are pulling through other areas of the Amphenol business.

00:21:14] Which additional Amphenol product categories benefit most from the Nvidia cluster deployments?

MO [00:21:24]: I would say CommScope for two reasons that, one, as I mentioned earlier, it requires almost like a 10x on the actual cabling requirements vs a non-AI workload data centre, which drastically increases the value of CommScope's share of wallet, if you like. I think the second is what I mentioned earlier, which is where some of these other subsidiary businesses have carved out their own data centre business units, effectively catching the crumbs from these larger bill of materials. Then if you work that backwards towards the OEMs that are winning these AI clusters, so we're talking primarily Cisco a little bit, but primarily, we're talking Supermicro, HP, Dell, these are also all consumers of Amphenol components in the interconnect division. They're all being carried through by ultimately the customer preferring these Nvidia architectures.

00:22:48] Does most of Amphenol's incremental AI revenue come directly through CommScope's hyperscaler relationships or indirectly through equipment manufacturers such as Cisco, HP and Supermicro?

MO [00:23:01]: The vast majority would sit within its OEM and ODM relationships. They're materially bigger than the revenue that they will get from the cabling and the optics. The cabling and the optics given diversification, they capture the most value inside the rack.

00:23:18] As AI architectures shift towards greater optical connectivity, does that increase CommScope's content opportunity further? Could it displace parts of the existing copper cabling portfolio?

MO [00:23:34]: No. I think the businesses on the optics and the cabling, I mean, first of all, the optics business is a compatible optics, which essentially means it's competing against OEMs who manufacture their own optics and it competes on two things, price and lead times. I think when it comes to the cabling business of CommScope, CommScope, from a product development perspective is a large, mature, globally scaled business with significant supply chain advantages, which means, especially now it has a parent with a much stronger balance sheet than it currently had in its previous ownership structure. It means that it can afford to invest greater dollars in the development of a product roadmap that would keep up. No, I would not necessarily see that as an issue. I would think that while the optics and the cabling are smaller, it will grow very fast because the clusters are scaling out optically, as you alluded to earlier.

00:24:57] How does the cabling and interconnect opportunity differ between AI training and inference? Does one provide the larger and more durable revenue opportunity for Amphenol moving forwards?

MO [00:25:12]: I personally don't think that, at least in the field, that's necessarily an issue. We didn't see too many opportunities where we thought we were not going to get as much of the business because one was training vs inferencing. I think both of them require huge amounts. If you think that from a sales perspective, what you want to see is one leading to another, or the collaboration of one or another. By that, I mean, when these clusters are being sold, they're often being sold in silos. One silo interconnecting with another creates a secondary opportunity. No, I don't think there's a material difference, certainly not on the cabling and the optics side of the business on training vs inference.

CD [00:26:13]: Did you mention that the AI data centre can require roughly 10x more cabling?

MO [00:26:21]: Correct.

00:26:22] Does the requirement for roughly 10x more cabling in AI data centres refer to cable volume or dollar content, or is it connectivity per unit of capacity? How directly should that translate into incremental revenue for CommScope?

MO [00:26:35]: It's the cable volume. It's the cable volume. That equates to the share of wallet on these AI clusters. If you think an AI cluster, and this is an anecdotal number, but if you think an AI cluster, maybe USD 0.5bn in networking infrastructure, that's a materially larger networking bill of materials than an ordinary data centre. It just requires more physical cabling. CommScope's, the number of cables they will need to deliver to be able to connect the cluster interconnect, the cluster is materially larger. That's where the share of wallet proportionally is probably similar in low-digit percent. The value of these bill of materials are so much higher than a traditional data centre bill of materials that the business, that the revenue that they will earn on these AI clusters are significantly larger.

Then probably the other thing to note is that there are just a lot more of these kind of projects that are coming up and certainly in emerging markets outside of CommScope's major markets in the US and in Europe as well as the hyperscaler business. There are lots of new neoclouds that are being created. There are lots of governments that feel they have to have a sovereign AI strategy, and that requires a significant investment in AI data centres in their country. I think while that creates this opportunity, as you mentioned earlier, the potential training vs inference change, again, they win on both counts because the training creates the content wrap, but the inference is the thing that gives it the repeatability over time.

00:28:34] Across scale-up, scale-out and data centre-wide connectivity, where do you feel Amphenol currently has the strongest position? Is there an area that you think could contribute the most incremental revenue for the company over the next few years?

MO [00:28:52]: I think probably scale up is the strongest. The high-speed copper, especially when it was designed in with Nvidia, I think the scale-out as in that kind of rack-to-rack connectivity is, that's a congested space. Optics are in there. Pluggables are in there and the Halo and CommScope both play a role there. Particularly the compatible optics space is quite congested there. Then I guess, scale across by that, I'm assuming you mean as in the interconnectivity between data centres.

CD [00:29:32]: Yes.

MO [00:29:33]: I would say that's perhaps less so even though CommScope will play a role there that other vendors that specialise in DCI-related architecture, so Ribbon, Ciena, Cisco, Nokia, they are better placed. Adva, they are better placed to take that share of spend. Of course, there is an argument that some components within the Amphenol components business will make its way into those products as OEMs. I would suggest that the interconnectivity between data centres, between AI data centres is a secondary growth area, which we'll see more of in a few years' time. I think right now, the vast majority of data centres that are being built around AI, which will pull through directly and indirectly Amphenol products are single silo clusters.

00:30:29] How long do you think high-speed copper can remain dominant within scale-up networks? Is there a bandwidth, distance or system configuration at which you feel optical connectivity would begin to displace it?

MO [00:30:44]: I'm probably not best placed to comment on that only than to say that there is a high degree of risk there from optical, from pure-play optical technology companies that the technology will rapidly advance. For example, my former company, NTT, we're investing hundreds of millions of dollars a year in trying to solve that problem, that area of the industry. I suspect that that would become a bit of a risk.

CD [00:31:22]: If scale up transitions towards optics, can Amphenol retain a comparable dollar content through its optical portfolio or would that create a revenue risk for the company?

MO [00:31:38]: Yes. I would say it's potentially a revenue risk for the company. Now that said, the revenue that the optics portfolio generates probably is not big enough in the context of its overall revenue number in order for it to be material. I would also think that Amphenol at a corporate level is well aware of this as a potential problem and would be looking at companies that it thinks can solve that problem. One thing I would say is that outside of CommScope, which I think is its biggest acquisition, lots of the component businesses that sit within there are relatively small. I think the bigger challenge is in a vast complex decentralised structure, how do they keep up with technology? The way that model exists today has material benefits when it comes to revenue and its EBITDA. When it comes to constant innovation and the ability to keep pace with technology, the lack of an integrated centralised capability to do that may be an issue and maybe that is something that they would look to solve through M&A.

00:33:17] Why do customers purchase compatible optical transceivers from Halo rather than OEM-branded products? As optical pricing shifts, do you think Halo can continue gaining share while sustaining attractive margins?

MO [00:33:39]: Yes. Yes, I do. I think the reason why customers will buy compatible optics is because they are slowly… There are a number of reasons, but primarily, they've become educated over time that the vast majority of OEMs across the networking stack will outsource the production to a very small supply chain of optics. Then typically, they will apply huge margins on top of optics on the basis that historically, customers have bought whatever they bought them as a platform. They bought the networking infrastructure plus the optics because the OEM is providing a warranty on all of it. I think now that they're becoming more educated, there's a growing understanding that these things have been materially overpriced that from a technology perspective, as long as a transceiver has been coded to work in the target device and tested to work in the target device, it will work exactly the same as an OEM version. Effectively, that's resulted in the commodification of optics, which is a slow process. I think that's one area.

The second area specific to the AI and the data centre space is price. The compatible manufacturers that have scale like Halo have a material price advantage. Then the third area is lead time. Halo is, I think, the largest independent provider of compatible optics in the world. That gives it a much broader supply chain, that gives it the ability to hold stock, which has increased by being part of Amphenol's balance sheet. I think when you look at some of the actual businesses that sit behind an AI data centre cluster, certainly, what I was seeing when I was involved there was that we would have a system integrator who is buying, fulfilling, integrating on behalf of an AI infrastructure operator. They were getting lead times from Nvidia Mellanox for transceivers that were, in some cases, six, nine, 12 months after they place the order for the GPUs. Essentially, there has been a large number of deals in AI data centres where the optics have been bought as compatibles simply because the Mellanox supply chain is taken up by hyperscalers. What that means in practical terms is the end customer is left with a choice, do they wait for the Nvidia approved optics, knowing that they're a commodity or do they swap them out, put compatibles in, which enables them to turn on the cluster to whoever is leasing it straight away?

Obviously, some of the clusters are leasing for tens of millions of dollars per month. While cost, while they're going to save a lot of money doing it that way as a by-product, actually, the principal reason to do that as in CAPEX, but the principal reason to do that is to allow them to turn on the clusters quicker for the end tenants. In some cases, that might be a revenue question. In some cases, that might be because they've signed a delivery SLA to an end tenant, which is also occasionally a hyperscaler with huge penalties on late go-live dates. That drives a lot of opportunity around compatible optics, but because for the same reason, the CommScope's business is growing, because there are so many more of these data centres that are being built and because optics is such a relatively small part, for example, of the Mellanox business that quite often, you will just see opportunities where it's the best placed second alternative, which wins the day. You can deliver it at a certain price point within a certain timeline with everything fully tested against the target environment, and that's a very compelling proposition.

00:38:08] As demand moves from 400-800Gbps and eventually 1.6Tbps, does Halo benefit from higher content and pricing? Does rapid price erosion limit its revenue and margin uplift?

MO [00:38:21]: Certainly, at 1.6, the supply chain doesn't exist at any kind of scale. It's a brand-new product. It provides it with a long tail opportunity, I would suggest, because if you're going back to the 400, we talk 400Gbps, 800Gbps, 1.6Tbps, we're only really talking about AI data centres. It has a nice run rate business in traditional telco and data centre and enterprise where the devices are smaller capacity and the form factors are slightly different than that. That's more of a Cisco enterprise type business. I think what that indicates that movement towards 1.6Tbps is just a continued growth in neoclouds, continued growth in sovereign and continued spend in the hyperscaler CAPEX. What we were seeing, particularly on the hyperscalers is while someone like a Halo would not be able to sell or partner directly with a supply chain, with the supply chain of someone like a hyperscaler that on these deals where hyperscalers were the end tenants, the hyperscalers would relent on which vendors were approved on the optics, particularly because there was a recognition that these were swappable components. Halo's moat here, if you like, is the fact that it's the largest independent one. In fact, if you look at other Halo's competitors that are not part of the Amphenol family, their price earnings would suggest that Halo are not the only ones who are benefiting from this.

CD [00:40:22]: Do you think Halo can expand beyond the replacement and aftermarket opportunity by being designed into new AI data centre deployments from the outset?

MO [00:40:30]: No. No, I don't. I don't believe the business is not set up to operate in that fashion. I think they're very strong at what they do. Even with Amphenol's support, I don't believe that business is structured in order to be able to do that.

00:40:54] Do CommScope's direct hyperscaler relationships create a meaningful new distribution channel for Halo? Are there channel conflict barriers or product overlaps that limit the cross-sell?

MO [00:41:08]: I would say the primary overlap at the minute would be commercial. There's no incentive. The second thing is the size of the two businesses. Halo is probably a USD 400m-500m turnover business. CommScope is materially bigger. It's harder, I think, for Halo to participate in CommScope bids just because it doesn't have the alliance of scale to be able to build those out. There are some branding issues. I would say that in my view, Amphenol by acquiring companies and running them in a decentralised function, that does mean that the brand Amphenol doesn't really mean anything to anyone, and certainly not and when you're outside of a supply chain discussion. That also means that because it's a very thin overlay, it's not like there's a team of people that sit within this ACS who are tasked with building cross-sell programmes and incentives and then enforcing them on the operating companies. What they want the operating companies to do is to manage their own P&L. Typically, you'll find that each company, whether it's Halo or CommScope or any of the other ACS businesses or elsewhere in the Amphenol family, the vast majority of guys are focused on looking after their own P&L. Outside of the occasional reference of clients and opportunity, there is not a structured cross-sell programme that either of those two companies could benefit from.

00:42:46] Do you expect significant incremental revenue to come from cross-selling Halo and CommScope products?

MO [00:42:54]: No. No, I do not. I think the theory is sound that you would take CommScope's hyperscaler relationships, which Halo does not have and you could build Halo in. The reality is that the hyperscalers are sourcing optics components directly. They're not buying products, they're buying components. Effectively, they're building their own labs. They've replicated what a compatible manufacturer would do internally, and they're tested against their own architectures. They effectively have cut out the margin that a business like Halo builds in.

CD [00:43:29]: Where does Amphenol ownership create the greatest economic value for Halo? If the cross-sell is not there, what is the best economic value for Halo under Amphenol ownership?

MO [00:44:07]: I think, firstly, there is a degree of comfort on larger deals by Halo telling the story that they're part of a business that's significantly larger than them. I think I would not overlook how important that is when you're trying to position Halo as a real player in optical technologies, how important it is to say that your parent is a public company. I didn't look at the share price today, but it's USD 140bn-150bn business. I think that's kind of material. Obviously, general managers generally speaking, are compensated in Amphenol stock, which is very helpful. I think the other aspect which is important is that Amphenol has OEM relationships. Amphenol's OEM relationships are very large and at very senior levels. Some of these OEMs either do not manufacture their own transceivers and therefore, are more open to bundling third-party transceivers into their solutions as long as they're certified or are very open. In the case of Dell, for example, they're very open to bundling third-party components under the Dell, call it their Extended Technologies division. What that means for, the advantage that is for Halo in practical terms is that being a subsidiary of Amphenol can accelerate the certification process. The certification process effectively buys you a gigantic global sales force that will be bundling your products with theirs on these larger AI clusters. Effectively, it gives you another route to market in the data centre business.

00:46:07] Amphenol's decentralised model gives its subsidiaries considerable autonomy. Are there areas where this becomes a disadvantage when serving large AI hyperscaler customers vs the advantages it creates?

MO [00:46:32]: If a customer wants integration, whether that's across power, cabling, optics, copper, currently, there's not a mechanism to have that joined up conversation really that integrates all of those two things. There have been anecdotal projects, but it's not something that's been actively shopped. You'll find that some of the components are more important than others. For example, CommScope, while the dollar signs will say otherwise compared to optics, but CommScope is a materially more important company to a customer than Halo is, even though actually they may end up spending similar amounts of money on an AI, on their respective products from an AI cluster. It becomes, I think, a disadvantage also. Obviously, that impacts the presentation of a joined-up solution. It also potentially impacts pricing. I think in general, I guess, I would be interested to see how this plays out and how important OEMs and Nvidia come to dictating everything, almost everything that happens within an AI cluster, in which case, behaving as a series of small to mid-sized companies as opposed to one extremely large one has a disadvantage. I think the other disadvantage that you will have is just access to the deals.

The individual operating companies are having to work a lot harder to build buyer relationships by themselves, which often may include channel conflict, basically two Amphenol companies quoting on two separate things without knowing, which also doesn't feel great in the eyes of a customer. The actual buyer access is materially hard when you're doing it on a single operating company basis, and it's a lot easier when that's coordinated in the centre. In that case, decentralisation is a significant advantage, a disadvantage, sorry. I'll give you an example. If you're a government in an emerging market that is planning to invest in AI data centres for a sovereign AI project, (1) Amphenol companies will find it very hard to access those in a way an OEM, for example, by comparison, would not. (2) If you are the government ministry that's charged with doing that or the vehicle that they've chosen in order to be able to put this together, it's very hard to figure out what Amphenol can provide and what it can't because there is not a joined-up solution. I do think in that case, the decentralisation is a disadvantage.

00:49:46] How does CommScope compare to Corning or other structured cabling competitors? Is there anywhere it is most vulnerable to pricing pressure or share loss?

MO [00:50:09]: Probably the one I can think of is Corning here, and I'll come back to something I said earlier. CommScope has done well for macro reasons when it comes to the AI data centre division. Corning already had an enormous market share. In a sense, CommScope had a challenger mindset. What they did was, as I said, they split the enterprise business, which is very large into effectively a retail and campus type business and the data centre. It's the data centre business that's grown phenomenally well. There is an argument which I think you're implying at here is that when that data centre business hits a reasonable amount of scale and it encounters the same supply chain pressures that Corning has by virtue of the fact that it's been very successful. Is that going to have a material impact on its growth and therefore, the share price? I'd say possibly, there's possibly an argument that, that could plateau. I personally just don't see that happening anytime soon because there is a really strong, robust long tail of demand around AI data centres and CommScope are very well positioned to provide solutions for that. I think there is a risk against the likes of a Corning, which is perhaps the one I can give you a coherent view on. CommScope is seen as a very strong product set that's comparable to Corning. It's those two. They're in the same deals all the time.

00:51:53] Amphenol expects the acquired CommScope business to contribute approximately USD 4.5bn of 2026 revenue. Based on demand and customer dynamics, how achievable is this expectation? What do you expect for a revenue contribution moving forwards?

MO [00:52:12]: I don't expect it to be lower than that. I'm not sure I'm close enough to it to give you a figure, to give you an exact dollar figure, but I would think because there's been no meaningful pause in hyperscaler CAPEX, and because there's been a material increase in neocloud projects, especially in emerging markets and because there's been an increase in the volume of neoclouds full stop, I would think that they will hit that number comfortably. I think, just to go back to something we talked about earlier, when it comes to the decentralised model and its disadvantages. The advantages of the decentralised model is they've taken out a business that was part of a struggling parent. There's two very distinct business units and all of the general managers of those business units are comped on EBITDA performance, which obviously then rolls up into Amphenol. They're all comped on Amphenol stock. There is a very clear alignment of incentives, which would give me a high degree of confidence that they would make numbers. I saw, obviously, the source, same thing in Halo.

00:53:26] Amphenol's adjusted operating margin reached close to 30% in Q2 2026. Do you think Amphenol can sustain margins around this level as CommScope becomes a larger part of the portfolio? Is there any upside or downside to the operating margin moving forwards?

MO [00:53:45]: I would say provisionally neither. I don't think there's a huge risk there. I think that's eminently sustainable. I think I expect that the IT datacom business to rise as a proportion of its overall revenue, and I expect that the optics businesses and CommScope will be a significant contributor to it.

00:54:12] IT datacom represents 43% of Amphenol's revenue, and the company generated 30% organic revenue in Q2 2026. What would you consider a normalised company-wide organic growth rate over the next 2-3 years?

MO [00:54:46]: I think that's very difficult to say as an outsider, not being part of the company anymore. Yes, I would expect it to still be extremely healthy. They have been very smart in effectively diversifying the risk away from the component layer, moving closer to the end buyer. I think there's actually more chance of them making more moves in that area than there is on the product side of the business, the component side of the business, I should say.

00:55:29] What do you think will be the most important KPIs or developments to monitor to indicate whether the thesis for Amphenol is strengthening or beginning to break down? What should investors pay attention to moving forwards?

MO [00:55:48]: I would say I would pay attention to the companies it acquires because that gives you an indication of its strategy, and that gives you an indication as to whether it's buying a company for strategic access, whether it's diversifying or whether it's buying because it needs the compound. I would pay attention to what happens to the communications market in general and whether there is any lift there outside of the data centre business, which again is kind of a macro, so broadband and cable spending. I would pay attention to… It's difficult to say, especially in the AI datacom, it's difficult to say how important Nvidia is here and the ongoing health of the system, of the ecosystem that Nvidia is seeding. If there continues to be external capital that's moving into data centres of significant quantities, if Nvidia continues to increase the volume of GPU shipped, then that will have a material impact on the ongoing health of both revenue and margins within the Amphenol business.

CD [00:57:15]: Great. I think it's a good place to close on my end by saying thanks so much, for your input. I really appreciate it. Thanks as always to our Third Bridge clients as well.