Coherent Corp. (COHR) Earnings Call Transcript & Summary

Good morning. How are all of you. I think we're going to go ahead and get started, if I could ask you to find a seat. I know they're so scarce. We're so happy to see all of you. Thank you so much for coming. We were saying this yesterday at our booth that when we bought Finisar -- when II-VI bought Finisar in September of '19, the first OFC we would have had together was 2020. And of course, we didn't attend because of COVID. So this is a big event for us. We're excited to see you. Yesterday's -- the whole week, the booth traffic has been fun, and it's been really exciting. So I'm Mary Jane Raymond, I'm the Chief Financial Officer at Coherent Corp. We have some excellent guests with us today. We have Dr. Giovanni Barbarossa, our Chief Strategy Officer and the President of the Materials segment; Mark Sobey, who is the President of the Laser segment; Sanjai Parthasarathi, our Chief Marketing Officer; Magnus Bengtsson, who is also our Chief Sales Officer, and Beck Mason, our Head of Telecom. Today's event will run about 90 minutes. It's 45 minutes of presentations and the rest in Q&A. As many of you know, Coherent's broad portfolio of technology platforms enables many of the exciting megatrends in the world today. Today, we're going to spend time talking mostly about the communications end market. Our first speaker is Dr. Matthias Berger, Vice President of the Coherent transceiver technology. Matthias will discuss the unique position we have in telecom and the role that the Coherent technology plays in the evolution of that space. Our second speaker is Dr. Julie Sheridan Eng, our Chief Technology Officer. Julie will cover the Datacom segment of Communications, and she'll provide insight into how high-speed datacom transceiver technology drives and enables both the cloud and artificial intelligence. Matthias and Julie will share their insights into the markets we serve and how they see them evolving. We think this will highlight many of the important trends and upcoming catalysts that we believe will be inflection points for these markets. As a reminder, any statements we make today are given in the context of today only. They contain risk factors that are subject to change, possibly materially. We do not undertake any obligation to update these statements to reflect events subsequent to today except as required by law. A full list of our risk factors can be found in our Form 10-K for the year ended June 30, 2022. The presentation materials that we'll be showing will be available on the Investor Relations tab of our website, coherent.com. With that, let me turn it over to Matthias. Thank you.

Yes. Thank you, Mary Jane. Good morning, everybody. I'm Matthias Berger, responsible for the development of coherent transceivers and components along with the underlying technology platforms. I have been working in the communications industry for more than 30 years, overseeing multiple breakthroughs in system technologies during my career. It's a pleasure for me to be here to share with you some important upcoming trends in optical communication. And my talk will focus on telecom networks. I will go deeper into the focus area of my work at Coherent, which is transceiver technology for optical communication. This technology enables products that coincidently are called coherent transceivers. So in this talk, when I mentioned coherent transceivers, I'm referring to the technology rather than our company, which is also called Coherent. So by now, we are all very familiar with what is driving the growth of optical networks. Nevertheless, let me share a few interesting facts from the Ericsson Mobility Report, published in November 2022. First, the share of mobile data traffic on 5G was estimated to be just 17% by the end of 2022, growing to 69% in 2028, indicating that 5G is still early in its deployment cycle. Second, video traffic is estimated to account for 70% of all mobile data, which makes it the most important driver for mobile traffic growth. And third, IoTs. Devices that are connected to the internet and share data over the network autonomously are expected to grow faster than the number of mobile devices from 13 billion in 2022 to 35 billion in 2028. This is an estimated CAGR of 18%. By the end of 2028, 60% of all cellular IoT connections are forecasted to be broadband IoT with 4G connecting the majority. Of course, the world accesses the internet not just through mobile but also broadband wireline and increasingly by satellite services. More broadly, next-generation broadband services are enabling higher communication data rates and driving more data-intense applications. This growth will continue dramatically, increase the data that needs to be transported across global networks. And today, the best technology to support our global communications infrastructure continues to be optical communication primarily through fiberoptics. According to analysts, the market for telecom optical communications was $21 billion in 2022 and growing at a CAGR of 12%. Coherent is the largest provider of optical components for optical communications with 2.5 -- 2.1 billion in fiscal year 2022. We have one of the industry's broadest portfolio of technologies and products that power optical transport networks from the edge where subscribers are connected to the metro and core networks that interconnect them to the data centers that make up the cloud. Our products benefit our customers by simplifying the user experience and enabling them to reduce both CapEx and OpEx. In all applications, customers value high performance, high reliability, smaller size, lower power consumption and, of course, lower cost. Because of the realities of network operations and economics, we see a different emphasis on some of these attributes depending on each section of the network. For example, in access networks, cost, power dissipation and physical size play a major role. In transport networks, spanning hundreds of kilometers, performance is key. In submarine and space networks, high reliability becomes a factor of primary importance. So having set the stage that we have one of the broadest portfolios of products for optical communications and a very strong presence in the market, I would like to turn to a product line that the industry refers to as coherent transceivers. As you remember, I'm referring here to the technology rather than the company. Our vision for how this coherent transceiver technology will evolve is a reflection of transformations that we see happening in the optical network: now, next and beyond. What we see now is a growing trend in the industry refers to as disaggregation of optical networks. Disaggregation promises to help operators reduce cost and broaden the supply base supported by the standardization of interfaces and the introduction of multivendor interoperability. What we believe will come next is the adaptation of coherent transceivers in access networks when optimized for that cost and power-sensitive application. We believe that such coherent transceivers will be the ideal technology to eliminate the capacity bottlenecks and engineering challenges in that part of the network. At the same time, we expect that a similar but rigidized technology will enable communication with low-earth-orbit satellites to and from earth and between its satellites. Compared with more traditional radio-based technology, optical links will alleviate power supply and distance challenges. Beyond coherent transceivers and access networks and in space communication, we see the possibility of coherent transceivers being adapted in data centers when switches scale to hundreds of terabits, driving the demand for transceivers in the multiterabit range. So let's begin with the transformation taking place in optical networks now: disaggregation. While disaggregation is relatively new in transport networks, its momentum is accelerating. For traditional telecom equipment from different suppliers is based on proprietary designs that are not compatible with each other. Network operators have managed this constraint by awarding equipment vendors network buildout awards by regions. The other -- on the other hand side, cloud service providers have been building data centers with standardized equipment that is multivendor operable. The largest cloud service providers, also known as hyperscalers, are increasingly building their own transport networks to interconnect their huge data centers across regions and continents. In doing so, they are driving the demand for transport equipment that is standardized and multivendor interoperable. In other words, the largest cloud service providers are driving the disaggregation of transport equipment, partitioning it into more interoperable functions. Network operators are also supportive of this bigger disaggregation trend. Because eventually, it drives down costs for them as well. In fact, network operators have also been driving open standards. Open ROADM is an example of a multisource agreement that is championed by telecom service providers. Open ROADM seeks to define standardized interoperable network elements controlled through open software interfaces. Standardization bodies like OIF, ITU-T and IEEE are supporting such efforts toward seamless interoperability. In October 2022, Nokia, Windstream and Coherent jointly announced the successful demonstration of OpenZR+ an open standard for coherent transceivers aiming at enabling network disaggregation. To really appreciate the far-reaching implication of this solution, it helps to first understand how today routers are connected to each other over the transport network. Typically, a router will be connected to the transport network through a transponder card, being connected to a reconfigurable optical add/drop multiplexer or ROADM. Certainly, there can be multiple transponders on one line card dependent on the system architecture. One of the essential functions of the transponder is to convert the short reach so-called gray link from the router to a long-reach WDM interface that can travel across the optical transport network. And the signal arrives at its destination, another transponder on a ROADM does a conversion in reverse to complete the link in between the 2 routers. The joint announcement was all about eliminating transponder line cards, which represents huge savings in hardware, real estate and energy cost for operators. We accomplished this by miniaturizing the long-reach transmission function of the line card into a tiny pluggable transceiver that can plug directly into the router. Taking the transponder functionality out of the ROADM system and moving it into the router is effectively a disaggregation step. The pluggable transceiver module now has a standard interface that can plug in to any router. At Coherent, we have developed a 400-gig pluggable coherent transceiver in a small QSFP-DD form factor that meets the OpenZR+ standard. It was quite a significant achievement in miniaturization and technology evolution. To enable this disruption at the system level requires disruptions at the subsystem level. That is what we actually add breakthrough solutions with competitive follow-through, including manufacturing at scale with high quantity. By eliminating transponders, operators achieve energy and space savings that scale with the number of wavelengths deployed. The total savings in footprint could eventually amount to racks of equipment reclaimed from the data center or central office. Converting a line card to a small form factor pluggable can easily save 80% in space and more than 50% in power. This was the result of many years of work and many millions of dollars of investment in technology and integration. In early 2020, we announced the IC-TROSA, which is an integrated coherent transmitter receiver optical subassembly. The IC-TROSA combines all the optical functions that are needed in a coherent transceiver into one compact subassembly. Our IC-TROSA is built on our industry-leading indium phosphide technology platform out of our fab in Sweden. With the IC-TROSA, we leapfrogged an entire generation of Coherent optics that were assembled using discrete components. If you follow the component optics markets, you are likely familiar with these components. They include the integrated tunable laser assembly, ITLA; the integrated coherent receiver, ICR; and the coherent driver modulator, CDM. Instead of focusing on assemblies of components, we went directly into designing a single integrated optical component, integrating all those functions, including its system-related control electronics. For the IC-TROSA, we choose indium phosphide technology because indium phosphide devices enable transceivers to achieve much higher optical output powers within the electrical power dissipation budget of the QSFP-DD form factor. Another advantage is that it can scale more easily into high volume because no manual fiber routing with external amplification is needed. This simplifies the system and reduces cost. In February 2022, we announced the world's first 400-gig digital coherent optics in a pluggable QSFP-DD form factor with high output power. This product received the Lightwave Innovation Reviews Award at OFC. Then later in October, we made the announcement with Nokia and Windstream that I already mentioned before. We accomplished a lot in coherent transceiver technology, specifically for 400 gigabit in transport networks. But we have a lot more on the way. And so with that, I will turn to what's next. As I mentioned earlier, what we believe will come next is that coherent transceivers will be adopted in access networks and in satellite links once they are optimized for lowest cost and power consumption. In access networks, there is an opportunity to upgrade millions of 10-gig links to 100 gigabit using coherent transceivers. The key is to leverage capabilities of coherent technology to achieve the same or greater reach than 10 gigabits. This way, on the optical infrastructure, for example, in between central offices can remain the same. Tunable 10G transceivers have a reach of up to 80 kilometers. At 25G, the reach of the transceivers is reduced to approximately 15 kilometers only. But 100 gigabit coherent transceivers can reach 80 kilometers and even more. The key is to develop a 100-gig pluggable transceiver in a small and standardized form factor. The form factor by far best suited for this is the QSFP-28, is already widely used for 100-gig gray optics. Moreover, the 100-gig QSFP-28 DCO ZR module, that's a complicated name, enables end customers to use a lower-cost 100-gig router port, which is at about 1/4 to 1/3 of a 400-gig router port. We start to see more positive feedback from telecom carriers to deploy our cost-effective 100G ZR solution. We also believe that a rigidized version of these 100-gig coherent transceivers will enable communication with low-earth-orbit satellites to and from earth or between satellites. We envisioned the potential for such a product years ago already and set out to develop a unique digital signal processor, or DSP, optimized for 100 gigabit per second. We believe that, that would enable us to meet the very challenging power and size requirements of the QSFP-28 form factor. In June 2022, we announced that we completed the tape-out of our very own Steelerton DSP. It was the first DSP and the industry's first of its kind. The Steelerton DSP is less than 1/5 the size of a 400-gig DSP, and the module using it consumes less than 1/3 of power of a scaled-down 400-gig pluggable module. Then we unveiled the first 100-gig pluggable coherent transceiver in the QSFP-28 from form factor and called it the 100-gig ZR. This year at OFC, this product received the Lightwave Innovation Reviews award with the highest possible score. We are excited about our progress since 2019, and we believe that our core competency in DSPs hovers in a whole new world of opportunity that lies ahead for coherent. The new coherent 100-gig ZR transceiver plugs directly into existing headends, switches and routers, enabling existing network elements to transmit 100-gigabit signals over links stretching up to several hundred kilometers. The 100G ZR transceivers are offered in both [ C10 ] and [ I10 ] versions to meet a broad range of deployment options, including central offices as well as in more challenging ambient environments such as in street cabinets. In September of 2022, we announced that we won a very important DARPA contract to develop coherent optical transceiver technology for the space -- for the agency's space-based communications node, or Space-BACN program. The aim of Space-BACN is to create low-cost high-speed reconfigurable, laser-based data links that will enable communications between various government and private sector low-earth-orbit satellite constellations. Various incompatible and often proprietary optical interstate links prevent low-earth-orbit satellite arrays from readily sharing information. Coherent is specifically tasked with designing and fabricating coherent transceivers for a reconfigurable modem compatible with most existing single-wavelength communication protocols at a data rate of up to 100 gigabits per second, while simultaneously meeting very stringent size, weight, power and cost constraints. These satellite arrays offer tremendous potential for low-cost global communication, sensing, imaging and space exploration. According to a 2022 report jointly published by the ITU and UNESCO, the pandemic sharply magnified the consequences of the digital divide. And nearly 3 billion people are without broadband and not able to access public services or learn or work from home. These satellites can also reach areas on earth with less-developed IT infrastructure, providing easier access to education and helping reduce the digital divide in our society. I described what's next, coherent transceivers moving into access networks and satellite links. But we also have an eye on what's beyond. And there is the potential for coherent transceivers to be deployed in data centers. It's not likely to happen for some time yet, but the industry is already discussing some of the potential catalysts. As switches increase in capacity from tens of terabits to hundreds of terabits, a very simple mechanic constraint, the area of the faceplate, becomes a bottleneck. The greater the switch size, the more pluggable transceivers are needed on the faceplate to access the entire capacity of the switch. At some point, there is no longer enough room on the faceplate for more pluggable transceivers. The size of the faceplate can certainly be increased, but that is not scalable because of the added cost and real estate. Another challenge, specifically in data centers, is the pure amount of fibers linked to the number of optical connectors. Inside data centers, quite often, tens of thousands of optical connectors need to be handled, requiring trained personnel as we believe the job can't really be done by robots. Terabit-per-second interfaces bring the number of optical connections down, maintaining the size of faceplates and significantly lowering operational cost. In addition, higher capacity on a single laser utilizing innovative technologies brings power dissipation down significantly. For example, a 1.6 tera transceiver could consume as little as 1/10 of the power of 16 older-generation transceivers at 100 gigabits per second. This saves electricity and reduces the carbon footprint, making IT equipment more sustainable for our children's future. The development of transceivers at multiterabit-per-second data rates will require innovative technologies with a greater level of integration, such as multichip modules and DSPs on 5 or 3 or even smaller structures in CMOS. Moreover, this broadband technology enables interfaces to even cover long haul, ultra-long haul and submarine type of applications. Coherent is a leader in optical communications technology, spanning access, transport and data center networks, enabling applications from the ocean floor to space. Our 400-gig coherent transceivers are enabling network disaggregation, which is already happening now. In addition, we are already on what's next, cost-optimized 100-gig coherent transceivers for access networks and rigidized for space applications. And we are already looking beyond at multiterabit coherent transceiver technology that will become necessary as data centers adapt switches with hundreds of terabits of capacity. Long term, we believe that the winners in coherent transceiver market will be suppliers that will have developed a strong core competency in photonic integration and digital signal processing design. Coherent will bring the world closer with technology -- technologies that will deliver greater bandwidth for more people affordably and more sustainably. Thanks a lot for your attention, your interest in our transformation, and your enthusiasm for where we're headed. We welcome you to come along. And now let me turn it over to Julie Eng. Thanks a lot.

Thank you, Matthias. Hello. My name is Julie Sheridan Eng, and I'm the Chief Technology Officer here at Coherent. Although most of us use search engines every day, I doubt many of us think about what happens in the time between when you type in your search query and when you get the results back. First, the average query travels 1,500 miles to a data center and back over the fiberoptic network using the type of equipment and technology that Matthias spoke about. But then inside the data center, a single query uses, on average, 1,000 computers to retrieve and answer. And those computers are networked together with fiberoptic connections. So only 20 years ago, the state-of-the-art for optical transceivers was 1G. And now at Coherent, we're shipping more than 50% of our datacom revenue into 200G and higher data rate transceivers. And this has required an extremely fast pace of development and innovation in lasers. 400G and 800G transceivers have become a reality, and at Coherent, with our broad range of technologies, we're poised to continue to lead in this space. Coherent is #1 in the datacom market with $1.2 billion in sales in our fiscal year 2022. We've maintained this #1 position for multiple decades, even as the market has evolved. Increasing demand for higher speeds is expected. In fact, 800G and 1.6T transceivers are actually expected to be greater than 50% of the total available market by 2027. The winners in this market are those who have the deep technology expertise and the scale required to keep up with such relentless demand for higher bit rates. Today, I'll walk you through our technology platforms and products in the context of the datacom market evolution that we see unfolding now, next and beyond. We'll begin by describing our strong position in the now with our complete portfolio of lasers that are capable of 100G per lane. These enable 400G and 800G transceivers. Having a full set of lasers really matters for a number of reasons. The first is that larger customers prefer to work with a small number of suppliers. And therefore, that favors suppliers that have a full complement of products in their portfolio. The second is that by being in control of the laser technology, we control the pace of development and the progress in the industry. And finally, having all the technologies in-house, we can objectively compare them and make the best architectural decisions and system tradeoffs. In the second part of the presentation, we will address what's next, which is 1.6-terabit-per-second transceivers with 100G and 200G per lane. The commercialization of this technology is being accelerated by data center network architectures that are expanding to enable artificial intelligence and machine learning. The bandwidth required for the AI and ML portion of the network is increasing at a faster pace than traditional compute and storage part of the network. The accelerated demand combined with fewer suppliers who can master the high-speed transceiver technology is favorable to maintaining a healthy and profitable business. And then finally, we will look at what's beyond as we research and develop technologies that bring us closer to what the industry refers to as co-packaging. At some point in the future, as Matthias pointed out, it may become impossible to route the high-speed electrical signals out of the switch chip across the printed circuit board to the optical transceivers and also to fit all the pluggable transceivers into the front faceplate of the switch. Co-packaging optics with the switch chip may be the solution to this future bandwidth bottleneck. Let's start with the now. First, it's important to understand that as the data rate of transceivers has increased and given physics limitations on lasers, it's become more feasible to achieve higher data rate transceivers utilizing multiple lanes. This multilane architecture may either be implemented with laser arrays and parallel fiber, or it may be achieved with multiple lasers of different wavelengths multiplexed over a single fiber. The important takeaway is that a given technology enables a given data rate per lane and that multiple lanes can be utilized to achieve the aggregate transceiver data rate. Today, transceivers are starting to ship with 100G-per-lane components. Transceivers can use either 4 lanes of 100G or 8 lanes of 100G to create 400G or 800G transceivers, respectively. In order to have established a strong position in the market, it's been necessary for us to have a broad range of technologies to optimize for all the transceiver types. We've developed a broad range of technologies to support 100G per lane at short and long distances. And at the heart of transceivers are lasers. For the shortest reach transceivers, vertical cavity surface emitting lasers, or VCSELs, are used. These are based on our gallium arsenide technology platform. VCSELs are generally the lowest cost, lowest power consumption solution and are the lasers of choice for less than 100-meter connections. Coherent has multiple gallium arsenide VCSEL fabs in the United States and Europe. We're one of the highest volume manufacturer of VCSELs in the world and the only datacom VCSEL manufacturer in the world with a vertically integrated 6-inch gallium arsenide platform. And we're continuing to innovate. We have just introduced our 100G-per-lane VCSELs to support 400G and 800G transceivers. In order for data center operators to have more freedom to connect compute devices that may be more than 100 meters apart, single-mode lasers are used. And these devices are made from indium phosphide materials. Coherent has multiple indium phosphide fabs in the U.S. and in Europe. Electro-absorption modulated lasers, or EMLs, are used in the majority of single-mode lanes operating at 100G per lane today. We manufacture 100G per lane EMLs to support up to 400G and 800G transceivers up to 10 kilometers. Another indium phosphide product, the high-powered continuous wave, laser may be used with silicon photonics transceivers operating at 100G per lane or 200G per lane over 500-meter links, which is in the midreach range. We also recently introduced such laser products, and we're making these lasers available on the market, and we also plan to leverage them for our own silicon photonics-based transceiver designs. At Coherent, we also manufacture our own gallium arsenide and indium phosphide photo detectors that receive the light and convert it back to electrical signals. As a vertically integrated manufacturer, we also have a fabless in-house integrated circuit design team designing our laser drivers, amplifiers, and clock and data recovery chips that are manufactured in Tier 1 silicon foundries and used in our transceivers. These devices are also quite important to differentiate the performance of our transceivers. We assemble and test transceivers in-house in our multiple factories in Asia with our primary factories being in China and Malaysia. In these factories, we have very skilled in-house test equipment development and process automation teams, making our factories among the most sophisticated and automated in the industry. Our customers appreciate this geographic diversity and the resiliency of our manufacturing footprint as it provides them with additional assurances of supply in the case of a natural disaster or geopolitical event. So far, I've been talking about 100G per lane pluggable transceivers in the data center, and this is the now. But let's talk about what's next. I'm sure by now most of us have seen the news about ChatGPT, the artificial intelligence chatbot created by open AI. What you might not have thought of is how much compute power is required to do machine learning and artificial intelligence. The AI must be trained on an existing dataset, which can contain billions of parameters. This requires significant compute power that is distributed over tens of thousands of processing machines, many of which are connected by optical fiber. Larger datasets and AI cluster sizes will drive the need for bandwidth across all interfaces. Large models are very computation intensive, communication intensive because local computations require results from remote computations, which increase the number of required optical interconnects and the data transmission load. To meet the computational demands, data centers are adding an entirely new set of equipment dedicated to AI and ML. And while interconnect data for the mainstream networking, including servers and memory storage, is doubling every 2.5 years on average, the new AI and ML portion of the network demands faster bandwidth growth. And interconnect data rates for that portion of the network are expected to grow much faster. We're at an inflection point of growth in the data center bandwidth requirements, driven primarily by the mega data centers. And this is currently what is driving the urgency to commercialize 800G and further 1.6T transceivers. In 5 years, the market opportunity for 800G and 1.6T datacom transceivers is expected to be larger than all other types of datacom transceivers combined. And this brings us to what's next, which is the path to 1.6T transceivers by leveraging 100G and 200G per-lane technology. Because VCSELs are easily scalable into arrays, 1.6T short-reach transceivers can be achieved with 16 lanes of 100G-per-lane VCSELs. So while we're working on 200G-per-lane VCSELs in research, 1.6T short-reach transceivers may be implemented first as 16 lanes of 100G until 200G VCSELs are available. For longer reaches, we're leveraging our indium phosphide technology platform. In September of 2022, we introduced a 200G electro-absorption modulated laser, or EML, for which we received the Lightwave Innovation Reviews award. Our indium phosphide technology platform is one of the very few in the industry that has been proven with more than 200 million datacom lasers deployed in the field over the last 2 decades, qualified by virtually every network OEM and Web 2.0 in the world. We leverage that deep experience to bring these new lasers that will underpin the many millions of future transceivers deployed in the data center. What we're even more excited about is a new laser technology called the DFB-MZ. This stands for distributed feedback laser with Mach-Zehnder. This is an indium phosphide continuous wave laser monolithically integrated with an indium phosphide Mach-Zehnder modulator. This is a very advanced photonic integrated circuit. This laser technology will enable 1.6T transceivers with up to 10-kilometer reach. This is state-of-the-art in 200G-per-lane laser technology. We've combined into this photonic integrated circuit the functional elements of several of our most advanced devices, which are each breakthrough innovations on their own. This includes our knowledge and experience from electroabsorption-modulated lasers, modulators for coherent applications and tunable lasers with Mach-Zehnder modulators. This technology integrates decades of learning and disruptive breakthroughs. With our entire portfolio of 100G-per-lane laser technologies, we will be able to support 800G and 1.6T transceivers, which we believe will be in high demand for at least the remainder of the decade. But we also have an eye on what's beyond. What's beyond 200G-per-lane and 1.6T transceivers? As usual, the industry will come together through standards and multisource agreements to evaluate options and arrive at a few competing solutions. Either the line rate will be increased to greater than 200G per lane, or the number of lanes will be increased from 8 to 16, or more advanced signal modulation will be used. And all of these options are possible, and all of them are challenging in their own ways. But there's another issue on the horizon that we are addressing as an industry. Soon, we may face new bottlenecks. As the bandwidth of the switch chips increases, it becomes challenging to route all the electrical signals in and out of them across a printed circuit board and to the optical transceivers seated in the faceplate, even though the switch and the transceivers are only a few inches away from each other. It will also become increasingly challenging to fit the needed number of optical transceivers in the front faceplate to support the full bandwidth of the switches, which continue to grow in capacity. One option is the use of board-mounted optical assemblies or BOAs. These are vertically pluggable transceivers that sit right next to the switch chip. Fiber cables run directly from the BOA to the front panel. Since the cables run above the printed circuit board, they are also referred to as flyover cables. This can also be referred to as NPO, or near packaged optics, referring to the fact that the optics have moved from the front faceplate to near the switch chip. In this design, the only connector needed on the front faceplate is the fiber connector, which reduces or eliminates concerns about front faceplate density limitations. This implementation also preserves some level of pluggability. The next level of integration beyond BOAs, flyover cables or NPO is co-packaged optics or CPO. At some point, it is projected that it will not be possible to route all the high-speed electrical inputs and outputs or I/O in and out of the switch chip with traditional electronics packaging, even to a BOA, flyover cable or NPO nearby. The solution is to co-package optical chiplets into the switch chip package so that the I/O of the switch chip becomes optics and fiber instead of electrical signals. For very short reaches, we believe that gallium arsenide VCSELs will continue to win, owing to their low cost, low power consumption and ease of integration into parallel modules supporting many lanes. We are already working on a VCSEL-based optical I/O project with IBM through a contract with ARPA-E. Under that program, we delivered a module with 16 lanes at 50G per lane for a total data rate of 800G in a very small package size of 13 millimeters by 13 millimeters by 4 millimeters. The ARPA-E program was helpful in building our capabilities. When co-packaging does happen, we'll be ready to support it. For longer reach co-packaging, distances greater than 50 meters, indium phosphide CW lasers will be used likely with silicon photonics. This implementation is generally shown as silicon photonics co-packaged with a switch chip with a remote indium phosphide CW laser connected to the switch chip package through polarization maintaining fiber. We at Coherent are working on CW lasers and silicon photonics for our 800G DR8 transceivers, as we've already described and demonstrated. And in research, we have made a 1.6T silicon photonics chip with integrated optical multiplexer and demultiplexer, which is very challenging to achieve. So as all of these technologies evolve, from the front faceplate to the [ NPO ] to co-packaging, we will be ready to leverage them. While we do see co-packaging as a trend for the future, it's important to realize how much customers love pluggable transceivers. That's because there's a rich ecosystem of transceiver manufacturers that support these standardized interfaces, which creates competition and economies of scale. In addition, pluggable transceivers can be added as needed, enabling the so-called pay-as-you-grow model. They also offer easy replacement as needed, offering flexibility for our customers. For this reason, we believe that we will see the industry innovate around the pluggable optics paradigm as far as they can. We believe that pluggable transceivers will support the necessary bandwidth needs through the end of this decade. When NPO and CPO are needed, we will be ready with our broad range of laser and transceiver technologies to apply them to these new form factors. There's no one-size-fits all when it comes to transceiver and laser technologies. With our broad range of technology platforms, gallium arsenide, indium phosphide and silicon photonics, we have the luxury of choice for any given application. We look at the detailed requirements for the application and we choose the best technology considering the power consumption, the optical specs, the reliability and the cost target of the application. We do detailed simulations and measurements and make objective decisions based on these results. And because we have all the technologies in-house, we understand the tradeoffs between the technologies and can choose the best technology for the application on an application-by-application basis. I began my talk by describing our strong position now with our complete portfolio of lasers that are capable of 100G per lane. These enable 400G and 800G transceivers. Having the full set of lasers really matters for a few important reason. The first is that larger customers find it more efficient to work with a narrower base of suppliers, which favors suppliers that have the full complement of products in their portfolio. And the second is that, by being in control of the laser, we control the pace of development. In the second part of the presentation, we address what's next, which is enabling 1.6T transceivers with 100G and 200G per lane. The commercialization of this technology is being accelerated by data center network architectures that are expanding to include artificial intelligence and machine learning. Data centers are adding an entirely new set of equipment dedicated to AI and ML. The interconnect data rates for the artificial intelligence and machine learning portion of the network are increasing at a faster pace than mainstream networking. And then finally, we looked at beyond as we develop technologies that bring us closer to co-packaging. The same technologies that are used in pluggable transceivers may migrate either next to the switch or into the switch package. By investing in our high-speed gallium arsenide, indium phosphide and silicon photonics platforms, we will be prepared to support near and co-packaged optics when those solutions are required. As the leader in datacom for over 2 decades, we expect to continue to innovate, to be a technology leader, and to deliver volume solutions to our customers to enable data centers that meet their needs now, next and beyond. Thank you. And now let me turn it back over to Mary Jane. Thank you.

So I think you're staying up here. I think my colleagues are joining me for the Q&A. So you want to come, Giovanni? You had all of us -- Sanjai as well. All these handsome guys and lovely Julie. We have -- so just to let you know how this is going to go. We have people walking around with mics, and we might ask you to just wait until you get it so everybody can hear you. But we have at least 2 or 3 people doing that, so it shouldn't take long. Chris, why don't you also just for the benefit of everybody in the room, let us know your name and where you're from, and everybody then'll know you.

Thank you for a great presentation. My name's Chris Rolland. I'm from Susquehanna. Julie, that was great. That was one of the best presentations on some very complicated matter that I've seen. Probably my questions are going to be for you. The first one is around AI and ML architectures. And you kind of expressed what they look like. I'm interested to know what products you think they're going to be using. In particular, are they going to be using EMLs or CW? And what does the economics mean between EMLs and CW for you? And just broadly, how do you see this market evolving and what it means for II-VI?

Yes. Great. Thank you for the question. Yes. So the majority of links for AI and ML are projected to be below 30 meters. I actually think there's a position for the VCSEL, actually. That's step number one. But then as you want to have longer distances or if there's some reason which some of our customers just like to use single-mode fiber, then I think you'll see a combination of EMLs and silicon photonics. Once you're putting it into a pluggable module that goes through the front faceplate, you're really just looking at a tradeoff of technology between the EML and silicon photonics. So the EMLs generally use, as you probably know, at 100 gig at least per lane at 2 kilometers and 10 kilometers. And the silicon in general, say that indium phosphide usually has the best optoelectronic performance. And so what favors silicon photonics is when you want to integrate a lot of things because that can help you integrate the passives and other things on to sort of a photonic integrated circuit. So the place where silicon photonics is mostly being used is for 500 meters. This standard called DR8, which is 8 lanes. You actually have 8 fibers, and that happens to favor silicon photonics. So I think you'll see -- at short reach, you'll see VCSELs, which will -- may be the majority of the links. And then I think at midreach, you'll see silicon photonics, and then at longer reach, you'll see EMLs. And then for us, we make silicon photonics-based parts. They do have CW lasers in them. And interesting thing is those CW lasers, to get enough power out of them, they're actually pretty big chips. And so from our perspective, I think we're kind of neutral which one wins. We have solutions for all of them, and they're all kind of the same to us as far as economics go, yes.

Great. And my second question is around how you guys feel about DSP. I know you've kind of dipped your toe in the water here with your first DSP product. So I'd love to know how you feel about a road map, whether PAM-4 could be on there at some point, whether you've looked into this linear direct drive, which is a hot topic at this show, and whether you think that actually obviates the need for a DSP? And if so, does that make it lower cost and drive more volumes through lower economics? Would love your thoughts on all of this, your DSP road map and direct...

Yes. Maybe I could start, and then I'll hand off to Matthias, who's doing all the DSP work today. But for coherent, we started with the DSP and then -- coherent, with the little C, coherent market in Matthias' products. If you can do that DSP, you can do a DSP for PAM-4. So it's really all about economics and us prioritizing our resources on where we thought they had the most value for us. And so could we do a PAM-4 DSP in the future? We have the capability to do it. It's more a question of where do we want to invest our resources to get the best return on investment, where we can differentiate ourselves, as Matthias pointed out, relative to what's available in the outside market. So mostly for our internal capabilities, a lot of times what we do is we look at what's the price we combine it for, what's the cost if we do it ourselves? How much does it cost to develop it? And thereby, you can do a pretty simple math about where it makes sense put your money. And then also, if you have something that you can do that's very differentiated, like Matthias talked about with the 100G QSFP-28. So that's a very differentiated product. You couldn't buy that IC on the outside market, and that's where we chose to put our money. That's -- I don't know, any other comments on DSP or...

It was pretty complete. Great. Great. But I mean at least on the coherent side, we have a long-term road map, so there will be multiple DSPs coming. And for what I call direct-detect PAM-4, certainly, as Julie was saying, we're looking into the numbers. So there is really, to a large extent, cost, certainly also a performance. But specifically, when looking into higher data rates, and Julie was already talking about 1.6 tera, it makes a lot of sense to do codevelopment of the DSP and the photonics. And this is one, let's say, value we have of our strong vertical integration. So yes, we're also seriously looking into PAM-4 DSPs as well.

And I think to step back to your other question, the linear, I think it's an interesting concept. Andy Bechtolsheim from Arista brought it up. I think the DSPs, they're expensive. They're big. They take a lot of power, and they also generally add latency to the link. And those are all things that are in the wrong direction for our customers, right? And so taking the DSP out of the pluggable module and trying to make a linear interface, that's been done already. Actually, in the coherent market, there was this coherent ACO, you probably remember. And that had some success. There were challenges. The challenge is there's a reason you have a DSP in the module, and that's because it cleans up all the signal before it has to go to the next place. And so engineering that is not so easy. But it's not impossible -- and so easy and hard things for dollars and power, I think you're going to see a lot of effort into that linear interface, yes. And that's fine for us because we make the lasers. We make drivers, TIAs. So everything in the module, it doesn't matter to us how it's partitioned [indiscernible].

Ananda?

Ananda Baruah, Loop Capital. One for each of you, if I could. I guess, starting with Matthias, the opportunity around the access network, the upgrade opportunity, we talked about millions of lanes being able to upgrade to 100G. How incremental should we think of that as being -- do you think that, that's something that -- I guess, how well known is that? And how incremental do you guys hold it? And then the satellite opportunity that you talked about, what needs to be done to sort of have that occur? And what do you think the timing of that could be as well as a practical matter for that 2.7 billion people or some chunk of it? And then I have a quick follow-up for Julie.

Yes. To the first question, we are shipping today lots of what is called 10G tunable transceivers. We have a large market share. And the demand for this one is really hundreds of thousands a year. And the interesting thing is the industry was looking at moving from 10 to 25G as I had shown on the slides, but the problem with this one is there are engineering challenges because reach comes down. That was the reason for us to move away from direct detect to coherent because now we can get the same, let's say, engineering as for 10G. And again, it's hundreds of thousands. And specifically with, for example, 5G coming, mobile, the, let's say, increase in capacity at the edge of the network is enormous, right? And that's one reason why we're moving from 10 gig to 100 gig to give a 10x capacity boost, right, at the network access.

And then the timing of the satellite opportunity?

Yes, for satellites, it's kind of interesting -- as I mentioned, satellites in the past were using radio links. And of course, the energy, right, is getting reduced with the distance significantly, and that makes it very, very inefficient. And that's where the satellite makers are -- they are moving over to optical links. And this is already starting now. And of course, the satellites have certainly one challenge, and this is power, right, electricity. And here, we are working on solutions with very low power to really be compatible to what those satellites need. And this is more or less happening today already. So it's not large volume relative to what's happening on earth, but requirements are very tight specifically. Power was one, but reliability or quality is another one. So that, let's say, leads to interesting margins in this application space.

Interesting. And just quickly, Julie, just going back to the ML/AI opportunity. You had mentioned that in 5 years the thinking is that 800G, 1.6T will be bigger than any other -- all the other categories combined. How critical to that occurring is the ongoing manifestation of AI/ML. And I guess what I'm trying to understand is, does the pacing need to continue and pick up in AI/ML for that to happen? Or is it just an adjacency to that happening?

Yes, that's a great question. I think that that's baked into those numbers. But I think even if the AI/ML, say, weren't to happen, we're still increasing bandwidth everywhere, right? Everybody is using videos. Everybody -- so there's sort of a fundamental usage of bandwidth that's just driving this sort of furious pace in the data center. And there's some portion of it that is the AI/ML, but I actually think that we're seeing it sort of almost -- if that didn't happen, I think you're still going to see a pretty big growth in [indiscernible] range.

So Ananda, I just wanted to add to it. If there is regular AI/ML with X, ChatGPT, what we heard, is 3x in terms of compute and communications. So it's a big driver.

We should probably [indiscernible]. Go ahead, Giovanni.

No.

You sure?

[indiscernible].

Simon Leopold with Raymond James. I wanted to see if you could maybe unpack what you're seeing going on with the hyperscalers in the short and intermediate term considering, not just the AI/ML opportunities, but whether there's some inventory absorption going on or where we are in the cycle and sort of your outlook for the next couple quarters, next year, and what kind of products are really going to drive the revenue?

Sanjai?

So the hyperscalers, we look at them, we say hyperscalers, but they're very different. Each one is very different. Their upgrade cycles are different. Their architectures are different. The strategies are different. So it depends on how you are positioned with a particular hyperscaler and then which part of their deployment road map you're in. But long term, we're still projecting that market to grow very well. And short term, some hyperscalers are digesting, some hyperscalers are growing. So it's a complicated question to answer because you cannot take one data point and say that's it for the industry, right? I mean that's...

I guess I appreciate that. I'm looking for kind of the net of kind of the takeaway for the group.

Well, again, we talk about Coherent, we're -- as you know, we're mostly concentrated, I would say, in 1.5 hyperscalers, so this way. And we -- so there is room for growth, for sure, for us by just taking share, which maybe we lost over time. So we'll regain share. So that's good. The general trend, maybe -- I think in the next 2 quarters, there's going to be some adjustment maybe due to maybe inventories in some cases. But I think we are well positioned, as I said, to gain share at those 2, 2.5 hyperscalers that we are not in. And I think we really made a lot of targets, particularly with the higher speed, not the 200G, which is the bulk of really what we're shipping, but almost on the 400 and 800 I think we have a very, very good upside on even the short term to see some pickup because, again, the upgrade cycles are all completely offset from each other, right, depending on the hyperscalers, right? And so -- but we have the broadest offering out there. We are ready to support them all. It's all about making sure we have the right design wins at the right time. And I think we'll be pretty aligned with their needs in terms of capacity. That's not an issue. I mean, it used to be a dish on telecom with supply chain, but we don't really have that challenge with datacom. So when we talk hyperscalers, we haven't really been affected by that. So that's not the real challenge for now.

I was just going to add, the growth rates are the fastest and the higher speeds. So 800G -- we've been shipping 800G for a few quarters. I mean, that's sort of at very fast growth rates. And our own business, over 50% of our datacom revenues is at higher speeds. So...

Samik?

Samik, JPMorgan. Julie, maybe this question is more for you. When I look at the products that you're thinking of on the VCSEL side short reach and the indium phosphide more sort of midregion higher, how do you think about the sort of competitive ecosystem on both those areas because not every competitor has capabilities across both? Where do you see more of a differentiation as you sort of build out the product portfolio because typically, sort of, if you think about shorter reach being maybe less differentiated, but the ecosystem also has to be present. So maybe that. And then the -- just a quick follow-up. I mean, you laid out the road map here. Any sense of time lines around when the sort of next happens -- when the beyond happens

Well, I'll start with VCSELs. So as Julie said, for AI/ML, that's really the short reach is where the activity is. That's where the action is. If I said ChatGPT consumes 3 times the resources or loosely said transceivers, then that's application that's growing. That's all short reach. There is still innovation. I mean, Andy says the best way to go faster is to go faster. And the 100G -- so we announced a 100G VCSEL. And that technology is going to continue to drive that...

Yes. Yes. I would say, in general, there are fewer people that can make VCSELs than can make indium phosphide devices. That's a true statement. These high-speed VCSELs are hard to make.

And at the very high volumes. You know we...

Right. Right. And with the reliability and everything, so...

So Samik, were you asking about when, for example, co-package will have -- will happens? I mean, I -- again, we are -- one way to look at it is that, as Matthias said, the datacom world has always been disaggregated versus telecom, right? Telecom is disaggregating now. But the datacom stream -- if I don't know what the definition is of this aggregation in terms of availability, interoperability, standardization and so forth, the last thing is going to happen that datacom is going to aggregate now because some of our competitors are saying that's what's going to happen. It's not going to happen. I mean, it will happen when you really have run out of technology, you have no options, then you are stuck to co-package everything, and you can't have a modularity, pluggability and all that. I think that's 3 years away, at least, so we don't really see co-package OpEx happening anytime in the next 3 years. Maybe it would be 5 years.

Yes, exactly.

Because again, even if -- maybe even again in that kind of world of the datacom world, even if the technology may have reached a point of no return, security of supply, let's call it this way, our competitive landscape for supply is more important, and so most of the hyperscalers will never go to a single supplier for anything, right? So you have to have a similar solution, so -- that's something that other people can do too and compete on cost. And so it's going to be very unlikely that, in the next 3 years, you will see 1.6 tera co-package. I think it can be done with transceivers -- with pluggables, I mean, transceivers. And so it's that -- yes, I think sounds about right. Let's say, 1.6T in 3 years. Still co-package is not going to happen. Maybe 5 years, we will -- maybe will happen. If it happens, great, we have a solution. We have this -- also something called COGA co-package optics, that we have developed. It can be a very, very versatile platform to address co-packaged solutions. So we are well positioned no matter what the time line is of the co-package deployment. But I think, again, from a disaggregation standpoint, if you -- that's -- maybe it's not the right term for the datacom world, but from the standardization and availability of solutions probably is going to be pushed out because pluggable transceivers are going to be able to deliver the solution needed anyway. So -- and people -- the customers like that.

I mean, as a technology platform, we think the best application for co-packaging technology isn't a pluggable transceiver. It's at least for the next 8, 10 years.

I think to the end of the decade, I'd be surprised if you see it even before the end of the decade, just because when you really look at it, if you can get to 1.6 tera for a transceiver, you can fit a lot of transceivers. And as I tried to say that you could even go to near package optics before you go to co-package optics, and that's still at least vertically pluggable probably, right, and that can support a multivendor ecosystem. And so I think -- again, but as we said, we'll be there to support it when we need to. So -- but I think people don't -- as I tried to say, do not underestimate how much our customers love pluggable transceivers. Yes.

[indiscernible] development cycles and so [indiscernible] more than a challenge [indiscernible]?

It's getting tough, actually. It used to be that -- I mean, you probably know that we would go like from -- in fiber channel 1, 2, 4, 8, 16. We had 3 years between every cycle, but we were only doubling every cycle. And then at -- even in ethernet, we've made that 1, 10, we had a lot of time between the cycle. So the cycles are getting faster and faster because we are having to match the switch chip because, essentially, as the switch chip goes up in data rate, we become the bottleneck, right? And so there's a huge pressure to innovate more quickly as we're getting closer and closer to kind of physical limits. So it's getting harder no question about it, but that favors people like us, where we have a lot of history, a long technology, a lot of smart people. And so that's actually better for us. But it is hard. There's no question about it.

There was a chart in Julie's presentation, if you look at the market evolution of the virus data rates, there's definitely -- the challenge is that all those are happening in parallel. So during the holidays, you had like 100 and maybe 25, and maybe 10. Now you have 100, 200, 800, 1.6. You will have 3.2 very soon. So all in parallel because, again, all the hyperscalers, they have -- in terms of the deployments, the upgrades are offset by 1, 2 years, and you have to manage all those in parts. So that's the complexity you're having multiple data rates in parallel -- more data rates than probably maybe you had in the past. Not to talk about the form factors. That's another complexity, right? The form factors, the data rate, and of course, the reach depending on where it's going.

Giovanni, I assume -- the diversity or the lack of commonality in those -- in the cycle times between the different hyperscalers. I assume that naturally favors a larger organization in the optical world in terms of being able to address all the different requirements?

That's true. That's true. If you want to address them all, for sure, you have to have scale and absolutely, yes. So if you have like a small company, maybe we'll be able to address very well one particular form factor, one particular data rate. We want particular reach. But I think for the scale we have, I think we are well positioned to capture the largest -- the broadest possible segment of the market.

Go ahead.

Ruben Roy from Stifel. Julie, I had a quick follow-up on Chris Rolland's question about linear drive. And just wondering, obviously, it's new, and we're hearing about it here at the show. But are there opportunities if that were to happen at some point? How do you compensate for the DSP, I guess, is my question? Do you make better TIAs, drivers? Is there an opportunity to gain value content?

Yes. Yes, definitely. Yes. I think you have to -- our customers would have to put equalization or something in their chip that they put on their board. And then for us, there would be a lot of ability to add equalization or something on our side to make sure that the whole link worked. It will be more of an engineered link. And so again, that favors people like us because we have very detailed modelers. We have a long history. We understand all the technologies and electrical and optical. So that kind of solution becomes much more complicated to actually implement properly, which actually favors people like us, I think. So this definitely gives you, I think, more opportunity to innovate actually, but it's hard yet. But that usually favors us. So that's good.

Dave?

Dave Kang, B. Riley. So yesterday, Lumentum, when they talked about the next gen for long-haul metro, they talked about CSDM. Just wondering what your strategy is there.

Long-haul CSDM?

By carrier. They said they're going to use more carriers as you reach Shannon's and Moore's Law?

[indiscernible]. Yes. I mean that's -- I mean that's been around, I don't know, let's say, 35 years, so something like that. So -- yes, it's possible. Of course, it's one of many solutions to fight with the Shannon's limit. If you start hitting the limit, you have to go their way. But I don't -- I'm not aware of any customer or any effort in the industry that is actually thinking seriously about deploying that technology. So I don't [indiscernible].

Honestly, I haven't listened to that talk, so would need to look into this one. But I mean, some topics are in research for decades and the reason is, sometimes there are some breakthroughs in some direction. But before things can turn into real products, let's say, an entire parameter range has to be met, right? And quite often on those solutions, are there 1 or 2 things which are very difficult to implement and to reach, right? And at the end, maybe it's possible to demonstrate something, but for example, yield is not achievable, right? So of course, we want to be cost efficient. And if your yield numbers are in the 10, 20 or whatever percent range, then it will never, ever turn into a product. This is more a general comment relative to some innovations which were published but never ever made it into real products, right? But in general, I mean, multi-carrier concepts are discussed for many, many years already. They have some advantages, but is it really a breakthrough? I don't know.

[indiscernible].

I'm sorry, we need to repeat the question. Can we repeat the question?

Paul just made a comment that's probably more [indiscernible]. Go ahead. Go ahead. That's fine.

Sorry, I'm showing my age. So I think it was around 15 years ago, there were 2 dedicated SDM companies, both of which went bankrupt.

Yes.

Exactly.

Maybe it was the wrong time. Yes.

And a follow-up question is, they also talked about integrated CNL, WSS. Just wondering where you are with that.

Sanjai, do you want...

WSSs?

It's all about WSS.

So -- I mean, the...

I want to say, we can only talk about 1 of the 2 WSSs. We can't talk about the whole separate. So I suspect that they're working on it.

Yes. We don't have access to their road maps. They're completely held separate, so we can't -- but it is -- we think there is a -- there is definitely a lot of activity going through C&L. I mean, we've had C&L and extended L and extended C amplifiers and all of the other parts. But on the WSS....

We're a little restricted.

We're a little -- yes, we can't talk about. We don't know...

All right. We have a question from Sidney Ho. We have 2 questions. The first one is, can you touch on the importance of vertical integration as the industry moves towards higher data rate transceivers and how much of a contributor has it been to your recent share gains into 100G and above? That's the first question. The second question is how dependent is the growth of our telecom business this year on government incentives currently targeting 5G broadband access. And given the various opportunities that you are working on, do you think that business can continue to see double-digit growth going forward?

Maybe I could start on the vertical integration portion. There's 2 different aspects to vertical integration. One is the design aspect and one is the manufacturing aspect. So we've been vertically integrated on many of our key technologies for multiple decades. And I was running the development for multiple of those decades. And the key is if you're a transceiver maker, and you don't make those bits yourself, you wait around till someone else develops all the things you need, and then you assemble them and test them. If you can do it yourself, you can decide yourself like Matthias decided I want to make a very small DSP and enable this new product that nobody else can make. And if you don't have that capability, you have to go around and try to convince somebody else to do that work for you, which they probably don't want to do just for you. They want to do it only once everybody wants there's this ability to differentiate yourself is very, very, very important. The other thing is when you pick and pull pieces from other people and try to integrate them, generally, the system is not optimized, right? If you know a lot about your lasers, you can put a lot of features in your laser drivers to make up for different things that might happen in your lasers. And so you can end up with a better overall product by engineering all of it together. So you can think about your road map in that way. So that's kind of the development in the technology pieces, but then there's also the vertical integration, which is the manufacturing piece. And that has a similar piece to it, which is you can invest in CapEx when you want to invest in CapEx. In cases like ours, we develop these testers ourselves, and we bring in-house process automation ourselves. And so that has its own value to us and to our customers. So I think there's 2 aspects. It's been a critical part of our technology. We do have very key suppliers. We can't do everything ourselves, nor do we want to do everything ourselves. There's a certain R&D budget that we have. And we choose to use it on the things that we think are the most impactful. And on other things, we rely on a very rich ecosystem of suppliers, and so we kind of make that decision. But I think it has been a key part of our strategy. And I think as you see these devices getting harder and harder, I think it becomes only either the same or more important going forward. Yes. Yes.

Right, for sure, time to market and cost to market are really enabled by the vertical integration. So that's a -- for sure an advantage.

I'll read the second question perhaps. So the second question is, how dependent is the growth of your telecom business this year on government incentives currently targeting 5G broadband access? And given the various opportunities that you're working on, do you think that business can continue to see double-digit growth going forward?

Okay. And in terms of broadband initiatives and spending, et cetera, ultimately, more 5G and more access -- broadband access. If that grows, it's going to grow the entire optical network from access metro submarine, everywhere. So -- but I don't think we are in a position where we could say exactly this percent of our telecom business was driven by broadband. It's just impossible to do it that way. But we did see a lot of growth in our access part of our portfolio over the last couple of quarters.

And there was an [indiscernible], Mark?

No, that's it.

That's it? Okay. Other questions? Any more questions on the phone -- from the website, Mark?

No more questions.

Okay. Thank you. Any last comments we want to leave all our good investors with here? Okay. Well, first of all, let us thank you all very much for coming. We hope that you enjoyed this. I really want to give Matthias and Julie a great shoutout for being our speakers today. That was as everyone has said, very, very illuminating and exciting as we look forward to the possibilities for our company in the long term. It's exciting to be part of a company that is fueled by megatrends that are irreversible. So thank you very much for coming. If there's anything else you need, you know where we all are. All right. See you soon. Bye-bye.

Thank you.