Fujitsu Limited (6702.T) Earnings Call Transcript & Summary

[Interpreted] Thank you very much for your patience. It is now the scheduled time, so we will begin our Fujitsu technology strategy update. Thank you very much for taking the time to join us today. I would like to explain the flow for today's presentation. Today's presenters are Fujitsu Limited Corporate Executive Officer, Corporate Vice President, CTO in charge of System Platform, Vivek Mahajan; and also Fujitsu Limited Corporate Executive Officer, EVP, Head of Fujitsu Research, Seishi Okamoto. First, Vivek Mahajan will explain Fujitsu's technology strategy, followed by Okamoto's explanation of Fujitsu's research strategy. After the total 35-minute presentations, there will be about 25 minutes for Q&A. And the update is scheduled to end around 11 a.m., followed by a demo tour at a separate venue. Regarding our investors and analysts, you will proceed to the demo tour venue after the Q&A session has concluded around 11:00 a.m. For the media participants, you will move to the demo venue after the photo session post briefing. The event will be moderated by myself, [ Saito ], from the Public Relations and IR office. Vivek will now first present on Fujitsu's technology strategy. Vivek, the floor is yours.

[Interpreted] Good morning, everybody. This is Vivek Mahajan, CTO at Fujitsu, and welcome to our technology update. For myself, what I would like to do is to explain the overview of our technology strategy and provide updates. And after that, I will pass it over to Okamoto to explain the overall Fujitsu research strategy and its details. So first of all, at the IR Day in September, I provided some brief updates. But there is one thing where I'd like to ask for your understanding where, as part of our technological strategy, basically speaking, similar for this year and also for the past 5 years, we've been focused on AI. And the reason why is the software and AI platform and the sovereign infrastructure supporting this. The reason why we say sovereign is because we are in the B2B business, working with enterprise companies and targeting these clients and their customers and to address their needs, sovereign is especially important. And for this, for the platform overall, from network to the software stack, to be responsible for this end-to-end, this has been our core strategy. And now to the question of what is sovereign. There are many different definitions depending on the person. But what's most important for us is security. When it comes to enterprise companies, security and data security, as I'm sure you're all aware, the corporate data as well as the AI models and agents, security becomes very important. For example, taking defense businesses, 99% of data is not on the cloud. And the other thing is flexibility, corresponding to the customers' needs, having the platform and also infrastructure and also on-premise, cloud, hybrid environments, infrastructure, software stack that can support all of this as needed. And the other point is domain specific, which aligns with the customers' needs and demands specific to their domains. A platform that supports this becomes important. And for this, on the right-hand side, we have our Fujitsu sovereign platform that we've defined here. At the very top, we have the AI software stack, Takane AI model, which we have been sharing with you, and the Kozuchi AI platform. We have these two. And Okamoto will be explaining the details later on. And the security stack is very important to support this. Fujitsu has been investing in this primarily, and we have many innovations thus far. So AI LLMs as well as the Kozuchi and Kozuchi security, we have our security stack that is supporting all of this for our software stack. And supporting this too is computing. I'm often questioned, Fujitsu, are you in software or hardware or network? We see this as a comprehensive solution. So it doesn't matter, hardware or software. In this era and time, we must cover everything, which means, for us, in Japan, we're very unique in the CPU, MONAKA, MONAKA-X as well as quantum and the hybrid environment as well with HPC, which we will touch on later, which is made in Japan. These made technologies in Japan is really important. And also the network stack. In order to realize this AI world, software, computing, network, we need these three. As you are very well aware, NVIDIA has invested in Nokia. And as we heard in today's announcement as well and their investments, we see all of this happening. Therefore, we have been investing in this strategy for a while. Now regarding Fujitsu's AI strategy. I'm often questioned, what are we going to do in AI. Obviously, we have the AI Kozuchi platform, which we have explained from before. But there are two points. One, the enterprise generative AI framework. Within this, we have the generative AI reconstruction. The other day, the AI single-bit quantization, we've announced this, first time in the world. And this is really important to us where the LLM model, in a very small computing power, it's still operable; and being domain specific, where it's specific distillation to create a specific AI model, which means on edges and also in a private environment too, we will be able to address the customers' needs and create AI models very easily and we don't need large computing power and a low cost. And Fujitsu's customers who are enterprise companies, they are in high demand of this. And the other thing is the knowledge graph. The unstructured data, we need to be able to use this. Within the enterprise companies, 80% is unstructured data, which means we need to be able to have the LLMs read the data which is unstructured. And to enable this is the knowledge graph technology. Therefore, the sovereign AI platform for us, it is important to be the leader not just in Japan but in global as well. So we want the world in defense, health care, finance, manufacturing, government to become known for our sovereign AI platform. So this is our Fujitsu road map. It's unique for companies to have a 5-year road map. But we have announced this with solid steps and with details that will be coming up every single year. And we have the larger milestones here. And for the AI platform, for Kozuchi, which includes AI and security, also Takane as well as the AI agent, the yearly evolution as well as the AI security stack evolution is noted here. And the other point here is on the bottom, we have physical AI. In October, Jensen, Tokita and myself, we've announced this. Fujitsu, NVIDIA, Yaskawa as the first stage. We have signed the partnership for robotics. And Fujitsu physical AI, we see this as the robots' brain as well as the sensing and the robot motions. On this, the software stack, network and computing, we will provide to support all of this. So we will be moving forward with the various partnerships in this area. And computing, supporting the AI stack is -- this is just a reminder where this will be launched next year, which is completely a 2-nanometer CPU, Japan-made, 144 cores and dual sockets in the CPU. And on this, in the future, the CPU lineup, this will become the foundation, the A64FX and Fugaku. Since then, we've been incorporating many different technologies, but this is going to be very versatile and open in the chiplet. So I hope you will look forward to this launch next year. And for the MONAKA road map, I mentioned the 2-nanometer. But in August, with RIKEN and NVIDIA and ourselves, Professor Matsuoka and Gonokami, we talked about the FugakuNEXT basis, MONAKA-X 1.4. And the biggest difference here is -- MONAKA A64FX, the biggest difference here is the versatile chip. Even in general companies as well as servers, this is usable. And the 1.4 nanometers will be 256 and 128 and NPU will be added in this new version. And in that world, we will aim to become market leaders. In August, we mentioned this as well. In the world of AI, the strongest CPU is what we're trying to achieve. And AI GPU with CPU combined together, we want to be able to provide the overall media platform. And so the AI supercomputer will be based upon this. And of course, quantum. Quantum, for Fujitsu, we want to aim to become a global leader. This is made in Japan. It's a completely made in Japan technology, [ electronic control ] and everything. In a concluded approach, we're going to make this all entirely made in Japan. And this will be launched in March. This 250 logical qubits, this is the most important. There's still a lot of technology that we need to work on, but we want to make sure that software stack and control electronics and keys all are included in our approach to become a global leader. This would be the road map. And for quantum, we have 256 qubits that we will be launching in March. And in 2030, 10,000. In 2035, we are targeting for 1,000 logical qubits. And what I would like all of you to understand is HPC and quantum, if we add these two together, there will be many things that we will be able to do. And HPC and quantum, I think Fujitsu is the only company that is working on both. And so I hope that you can be noted at that point. And last but not least, what I want to talk about is network, Fujitsu's network technology. I would like all of you to understand that without network, the world of AI cannot be realized. This is an area that we want to be ahead of the curve around, photonics, mobile system, network orchestration and data-centric infrastructure, the software stack. What we're developing today is photonics and optic, and we want to have a single software stack that has all that in DCI, data-centric infrastructure. We want to be able to connect that through our software and hardware technology. We also will have a road map for all of these initiatives. And 1.6 terabyte, this is an area that we definitely would like to acquire global leadership position around. And AI-RAN is also a very important technology. We have been announcing this with SoftBank and NVIDIA. This is a technology that we jointly developed. We are working to deploy this with many customers. If this is achieved, then even at edge, we will be able to realize AI. So for Fujitsu, software stack, computing and network, with all of these combined, we want to be able to establish these three pillars. In summary, this strategy is centered around AI. And I think from about 1.5 or 2 years ago, a lot of the companies have this as a strategy. But we have had this for 5 years. It has not changed. Our initiatives are centered around AI, network, computing, security and converging technology. These five technologies are areas that we will continue to work on, and I hope you can look forward to our efforts. Thank you very much. That's it from me.

[Interpreted] Mr. Mahajan, thank you very much for your presentation. Next, I would like to have Mr. Okamoto talk about the Fujitsu research strategy. Okamoto-san, the floor is yours.

[Interpreted] Good morning, everybody. My name is Okamoto from Fujitsu. I would like to explain our research strategy. Starting with our five key technology areas that Fujitsu is involved in. We want to reinforce each of the technology areas. In addition to that, we want to fuse on the technology centered around AI. This is our big approach. And as you can see on the bottom of the slide, we are forecasting the medium- to long-term technology trends to develop technologies that pave the way for next-generation: physical AI, space, defense and next-generation communications. Today, I would like to explain the technology update in each of these fields. Starting with the AI research strategy. We have our R&D activities based on two areas. First is Takane. This would be GenAI for enterprise. And the other would be Kozuchi, Fujitsu's AI platform. With these two axes, we are working on research so that sovereign AI, a strong version of this, can be established. This would be the essentials of our research and development activity. Today, I would like to talk about our involvement around Takane and how we have reinforced Kozuchi. Starting with Takane. The large-sized language model, I think there are three key challenges. One is that the development and operation cost is increasing, and the power consumption is increasing, and the third is the need to address edge AI. To solve these challenges, it will be a requirement for us to achieve a very strong sovereign AI. We have been working on technology to be able to address these challenges. Today, I would like to talk about the generative AI reconstruction technology. One is quantization and the other specialized AI distillation. The GenAI that is generally used is expressed with 16 bits. That's what we can say in general. And the 16 bits, if we lower this to 1 bit, we have been able to achieve this 1-bit quantization. This is what we call the single-bit quantization technology. The GenAI model, when we compare with the one using 16 bits, the accuracy retention rate for the single-bit quantization is 89%, the processing speed is 3x faster and 98% of power consumption can be reduced as well as the GPU cost. The other aspect of this AI reconstruction technology is the upstream technology. It's not just about making the parameter smaller. But this upstream technology, when we compare it with that, the speed of inference is 11x higher and the accuracy has improved by 43% and memory usage is reduced by 70%. So the accuracy is elevated through the reduction of parameter size of 1/100. And this quantum technology can be used not only in a limited area, but it is very versatile. And so we would like to release the single-bit quantization technology as OSS. Next, I would like to talk about the knowledge graph enhanced RAG. This is what we have created. This is not Internet data. This is using work data or enterprise data or domain data. And it's really about embedding this into generative AI. This is becoming a key issue. As you can see here, we are #1 in document search with charts and tables. And as a graph RAG, Microsoft's graph RAG, if we compare it to them, the speed is 50x higher. And the knowledge graph itself is operated, and we're also making self-improvements through context and execution results. We are trying to deploy this in various areas. For example, in the manufacturing sector, we have maintenance and incident documents or components documents that we can capture. For the finance sector, we have a source code and we can also confirm the specification. And for health care, we can create a knowledge graph based on electronic health records. So these are deployments that we are working on at the moment. Next, I would like to talk about the enhancement of our Kozuchi AI platform. What I would like to talk about are two points. One is by working with NVIDIA, we will be enhancing Kozuchi. As you can see here, NVIDIA NeMo or NIM, these are the software stacks that NVIDIA has. We're going to integrate this into Kozuchi so that we can enhance the accuracy as well as enhance automation. FUJITSU-MONAKA can also be combined so that NVIDIA GPU and NVLink Fusion can provide global standard secure AI platform. Regarding Kozuchi, there's one more point, and that it is strengthening Fujitsu's IP. The first is on the AI security area, addressing vulnerabilities in the LLM, 7,700 vulnerabilities. This is the strongest in the industry using scanner technology. And also the attacks to RAG, risks around information leakage, to address this at 98% accuracy, we have developed risk countermeasures. And this year, in October, the Fujitsu cloud generative AI platform, we have onboarded this commercially as well as with Cohere, the AI services security enhancement. We have already started a partnership with them. Regarding Fujitsu's strong IP, the next point is on causal AI. And regarding causal AI, not correlation but accurate measures and initiatives. To give a more accurate cause-and-effect relationship, this is a hot topic in AI. And Fujitsu's causal AI unique points are the following. The first is scalability. And this is 1,000x faster processing than conventional methods. And the other is estimating causal effects. So using the technology that is able to give causal effects from a large amount of data, we are able to apply this to many different areas. So Uvance and Fujitsu's data-driven technology, as part of the core for this, we are planning for commercialization for next year. Lastly, this is agent. And Fujitsu's multi-AI agent, we have announced this as of December of last year. And within the respective companies, it's not just creating the agent systems. But our target for R&D is cross-industry, which means between different industries, how to operate these multi-AI agents. So this is a technology to support that. Largely speaking, there are two here. Agent communication between different companies is what we are talking about. So incomplete data, using this type of data, how are we really going to be able to operate these AI agents. And the second is within the AI communication, the vulnerabilities and using the secure inter-agent gateway to prohibit sharing of sensitive information, we are introducing guardrails for this. And we are going to start the supply chain verification at Rohto. And also the COCN organization, we are implementing this to promote policy recommendations. So we are starting on these initiatives. And also, we are planning to offer this through Uvance's Dynamic Supply Chain business from next year. Regarding security, I would like to provide some input on digital fakes. In the explosive evolution of GenAI, there are obviously many benefits. But on the back end, the content generated by AI, we've seen a lot of misinformation and disinformation which is posing many risks both socially and economically. So at Fujitsu Research, as you can see here on the bottom, the technologies for detecting disinformation with high accuracy as well as the LLMs that are specific for this for detecting the misinformation as well. So we have been developing many different applications. At Fujitsu, what we are promoting primarily, the NEDO K Program. We've applied some technologies to this program already. And the K program, what this is able to do is -- primarily, this is Japanese innovative technology. But we are announcing today the international consortium, Frontria, which is not just in Japan but globally. The AI governance and misinformation/disinformation, to counter many different risks, we have created this international consortium. And the unique characteristics of this is we're able to create new markets by globally combining various IPs through this platform. We have 57 members who are participating in this consortium, and you can see the starting members here on the slide. And today, we've received many different endorsements thus far. But especially, one of Italy's largest banking groups, Intesa Sanpaolo; as well as Comprehensive Financial Services Group, the core business of non-life insurance, MS&AD, we received endorsement videos from the two. So we would like you to take a look.

As a financial institution, we operate in an environment shaped by evolving risks, complex regulatory demands and the growing impact of news information. We view this consortium as a valuable opportunity to advance shared standards and foster responsible innovation. Through our participation, we aim to continue with our expertise, strengthen collective safeguards, and then we shape solutions which promote trust, resilience and long-term societal benefit.

[Interpreted] At MS&AD Insurance Group Holdings, I am the General Manager of the Innovation Planning Department, and my name is Komatsuzaki. I am involved in the testing of new technologies such as AI and business development in our company. Congratulations to Fujitsu on the establishment of the international consortium, Frontria. In the non-life insurance business, accurate information, particularly images and videos of the damage, is crucial for making appropriate liability determinations in the event of an accident. And measures to prevent fraudulent claims are becoming increasingly important. Through collaborative technology development with cutting-edge companies from around the world, we will strive to maintain and grow a healthy non-life insurance system, and we appreciate your support.

[Interpreted] Regarding Frontria, at the exhibit corner, we have a specific booth which we will be providing demos and showing our technology. So the next area is quantum. As you're already aware, 64- and 256-qubit superconducting computers, together with Professor Nakamura's team at RIKEN, we've jointly developed this. And also, the G-QuAT of AIST, we've introduced this to them already. And the 1,024-qubit superconducting quantum computer, which is the world's largest class, we will be announcing this more formally next year. And FTP, Fujitsu Technology Park, which you are visiting here today, within this property next door, we have the quantum building, where within the headquarters, the 1,024-qubit superconducting quantum computer will be operational. And also with FUJITSU-MONAKA, the hybrid HPC environment, by creating this, we will be able to use this as a test bed for quantum. Regarding the qubits, making this more large scale, what is really important in the commercialization is, as you're aware, the quantum error correction technology. So between the logical qubits and the operations between the qubits, through a unique gateway, becomes very efficient, using the evolution of the STAR Architecture, which we have announced. And as one of the biggest updates recently, we have two. One, which is with comparison with more standard technology, we're able to succeed in the computation speed by more than double and reduce error occurrence probability to 1/6 or less. And also in the quantum area, breakthroughs in quantum technology, to pursue these breakthroughs, the crucial part is developing use cases. And this is a very big theme for us. And as you can see here, we have materials, drug discovery, finance, manufacturing and robotics-included machines. And we are aiming for many different use cases to seek breakthroughs. So at the demo booth, we have in the materials area as well as on machines related to robotics, we have a demo that we can provide to you. So we encourage you to take a look. And FUJITSU-MONAKA, we are currently in development. And FUJITSU-MONAKA-X 1.4 nanometers using this semiconductor, CPU and NPU, as you can see here, FUJITSU-MONAKA-X CPU version, what this is, is the MONAKA that we have developed so far. The confidential computing technology as well as the super low energy version, especially on edge computing and uses on those, energy efficiency as well as response times, it's very strong on these points in the CPU that we have developed. And then furthermore, in addition to this, we have the MONAKA-X CPU + NPU. The MONAKA-X inference speed has further been elevated, and it also contributes to low power usage. Furthermore, for MONAKA-X CPU, by combining this with GPU, we will be able to have a large-sized scaled LLM training. As you can see on the right, we have a real-time multi-model AI, edge cloud integrated inference and confidential data AI analysis. All of these AI workloads, out of this, we can offer the best option for the customers. The FugakuNEXT that we announced the other day, the MONAKA-X only CPU and the NVIDIA so-called GPU are combined together so that FugakuNEXT basic specification has been awarded. So we've been able to acquire the order from this. The CPU technology will not only accelerate Japan's technology, but the next-generation AI-HPC platform can be realized. And also around 2030, we should be able to operate with the computational problem-solving technology. Next, I would like to talk about the converging technologies. Digital technology and policy twin can be combined together in this. As you can see, we have the urban transportation, regional transportation gap, environmental measures, EV promotion and rising medical costs that we face as challenges. And we can create the so-called social digital twin and do a digital rehearsal. As you can see here, there are various regions that have already implemented this. Social digital twin and digital rehearsal. At Uvance, the trusted offering in society, has already embedded this technology. And as a new technology, we have what we call a policy twin. This automatically creates policies. And through the development of the technology, we have some demonstrations that I hope you can take a look at afterwards. Next is the ocean digital twin. For ocean digital twin, as you can see, there are key social challenges. We first want to update the technology, especially the automatic control of underwater drone. In the conventional technology, the positioning accuracy was 1 meters of a gap, but we've been able to narrow that down to 50 centimeters. Last year, when we made this announcement we were only able to apply this to approximately 50% of Japan's maritime area, but now it has been raised to 80%. The certification rate is 95%, and so we have been able to win the top-class certification of J-Blue Credit. Furthermore, not only do we want to see and observe, we want to be able to figure out the seagrass bed creation so that we can help reduce CO2 emission. As you can see here, we are deploying this in various areas to fully make use of this technology. Next, I would like to talk about the physical AI area, spatial robotics. As all of you know, not only in the industrial sector, there are human robotic co-existence and scientific experiments where robotics are hoping to be adopted. We're not going to target just single robots, but we want to control multiple robots. The conventional world model will now turn into a spatial world model, and so we are working on developing this as well. We have sent out a press release today. And I think seeing is believing, and so we would like to encourage you to take a look at this at the exhibition corner. In the area of space, on the left, we hear that there are a lot of private companies that are launching these satellites. And we would like to contribute in the spatial area as well satellite data platform technology. So it's about processing data with satellite edge computing. This is very important to have the data processing speed. And we have been able to reduce the power load to 1/3 and data transmission can be done within 10 minutes. This was the targeted time in the industry, and the near real-time has been achieved through our technology. Another point that I would like to mention is the resolution issue of the images. We have an AI technology of high precision and we have a technology to sharpen the image by 225-fold. Not only satellite data, but we also want to integrate that to the industrial data to generate new value. This becomes key. We have the large-scale geographic information processing infrastructure to realize this in the future. This also is available for you to see at the demonstration corner, so please take a look at this. The last field that I would like to talk about is the defense and next-generation communications. The so-called material science and device technology are to be combined so that we can aim to realize a safe, secure and sustainable society. One is continuous wide-area monitoring system for defense and disaster prevention. Fujitsu's so-called two-wavelength infrared sensor is to be developed for early social implementation. We would like to provide the world's top-class high-sensitivity, high-resolution infrared sensor to ATLA. The other area is power-saving wireless communication technology. 50% of the powers in base stations are amp. And so we would like to use our own technology to be able to reduce the environmental impact. In the demonstration corner, we have these 12 booths for you to take a look at. As I mentioned in my presentation, the international consortium, Frontria, we have a special booth aside from these five key technology areas. We also have new fields for exploration. One is the spatial world model as well as the satellite data utilization. So please make use of the time to be able to take a look at our demonstration. And that is the end of my presentation. [Portions of this transcript that are marked [Interpreted] were spoken by an interpreter present on the live call.]