Keynote by Vivek Mahajan CTO Fujitsu India AI Impact Summit
20 Feb 2026 16:00h - 17:00h
Keynote by Vivek Mahajan CTO Fujitsu India AI Impact Summit
Summary
Speaker 1 opened the session by framing the talk around AI sovereignty, stressing its strategic importance for countries such as India that aim to lead in artificial intelligence [1-3]. He defined sovereignty as the capacity to own, control, and flexibly manage data and AI models while maintaining security, without excessive dependence on third-party providers [4-10]. Fujitsu leveraged its 90-year heritage and recent milestones-including the world’s first two-nanometre ARM-based servers and a quantum roadmap-to position itself as a provider of sovereign AI solutions [14-22][20-21]. The speaker identified three essential pillars for any AI platform-software, compute, and networking-and argued that true sovereignty requires independence across all three areas [26-29]. Fujitsu highlighted its Japanese-made hardware, notably the Monaca two-nanometre chip powering a planned 20-exaflop AI supercomputer and an upcoming 1.4-nanometre processor with integrated NPU for inferencing, both featuring hardware-level confidential computing [35-41][42-48][55-58]. He emphasized that the accompanying software stack is fully open source, preventing vendor lock-in and allowing customers to fine-tune models for specific domains [44-52][84-86]. In the quantum domain, Fujitsu claimed a top-three global position, targeting a 250-logical-qubit system by 2030 and a 10,000-qubit machine within three years, to be combined with high-performance computing for mission-critical AI workloads [62-71][69-73]. Its networking strategy includes a 1.6-terabyte photonic switch with low-latency, long-reach capabilities and an open-RAN orchestration layer to move AI workloads efficiently across data centres and edge sites [75-76]. The Takane large-language-model platform and Kozuchi AI-agent framework were presented as tools for building domain-specific, secure, and customizable AI applications in sectors such as defense, healthcare, and finance [77-84]. These technologies are intended to converge on edge devices-including robots, drones, and medical equipment-enabling autonomous operation while preserving data sovereignty [87-90][91-93]. Fujitsu’s go-to-market approach relies on integrated solutions and partnerships with companies like AMD, Lockheed Martin, and Supermicro rather than selling isolated components [95-99]. The session concluded with Speaker 2 announcing the next fireside chat featuring executives from CDAT and Intel and requesting speakers and the audience to clear the stage [100-104]. Overall, the discussion underscored Fujitsu’s strategy to combine proprietary Japanese hardware, open software, quantum advances, and network innovations to deliver sovereign AI capabilities for nations seeking secure, flexible, and independent AI infrastructures.
Keypoints
Major discussion points
– AI sovereignty as a strategic priority – The speaker frames sovereignty as “being flexible and secure,” stressing the need for countries (e.g., India) to own and control their data and AI models without over-reliance on third parties [1-9][10-11].
– Fujitsu’s sovereign-focused hardware portfolio – Highlights the upcoming two-nanometer ARM-based “Monaco” servers, a future 1.4 nm chip with 256-core + 128-core CPUs and an NPU for inferencing, and a planned 20-exaflop AI supercomputer built on the Fujitsu Monaca chip with confidential-computing features [20-23][36-43][44-48][55-58][60-62].
– Open software stack and AI platform – Emphasizes that the software stack is completely open with no lock-in, incorporating the Takane large-language-model platform and the Kozuchi AI-agent technology to let customers fine-tune domain-specific, secure models for sectors such as defense, healthcare, and finance [44-49][78-84][85-86].
– Quantum-HPC integration for mission-critical AI – Announces a roadmap to 250 logical qubits by 2030, a 1,000-qubit machine going live next month, and a 10,000-qubit system in three years, positioning quantum together with high-performance computing as a driver for advanced AI workloads [62-70][71-73].
– Advanced networking and photonics – Describes a 1.6 TB (future 3.2 TB) low-power optical switch for long-distance, low-latency transmission, and the use of open-RAN orchestration to move AI workloads efficiently across data-center and edge environments [76].
Overall purpose / goal
The discussion is a promotional briefing by Fujitsu aimed at positioning the company as a one-stop provider of “sovereign AI” solutions. By outlining its end-to-end capabilities-secure, open-source software, cutting-edge compute (including AI-optimized CPUs/NPUs and exascale supercomputers), quantum accelerators, and high-performance networking-the speaker seeks to convince governments and enterprises (particularly in India and Europe) that Fujitsu can deliver independent, privacy-preserving AI infrastructures without reliance on foreign cloud vendors.
Overall tone
The tone is confident, forward-looking, and technically detailed, consistently emphasizing innovation, security, and independence. It remains upbeat throughout the technical sections, shifting only at the very end when Speaker 2 takes over to announce the next session, where the tone becomes procedural and courteous. No major emotional or argumentative shift occurs within the main presentation.
Speakers
– Speaker 1
– Area of expertise: Artificial intelligence, quantum computing, high-performance computing, networking, sovereign AI solutions
– Role: Presenter / speaker representing Fujitsu
– Title:
– Speaker 2
– Area of expertise: Moderation / event facilitation
– Role: Moderator for the fireside-chat session [S1][S2][S3]
– Title:
Additional speakers:
– None
The presentation opened by defining AI sovereignty as the ability to be both flexible and secure-to own and control data, to fine-tune AI models, and to do so without excessive reliance on third-party providers [1-9][10-11]. This framing linked sovereignty directly to national security and economic independence.
Fujitsu background – The speaker highlighted Fujitsu’s 90-year legacy, from early mainframes through pioneering DRAM to the world’s first two-nanometre ARM-based servers, and noted the recent launch of a U.S. brand that aggregates all Fujitsu solutions for customers [14-23][20-22][15-19].
Three-pillar framework – Sovereign AI was positioned as requiring three independent layers: software, compute, and networking [26-29]. Fujitsu, as a Japanese-made technology provider, was presented as an alternative to dominant American vendors, giving governments and enterprises a genuine choice for sensitive sectors such as defence, healthcare and finance [30-33].
Compute pillar
Fujitsu announced the imminent shipment of “Monaco” servers powered by its proprietary Monaca two-nanometre ARM chip, which embeds hardware-level confidential computing to protect data [35-41][60-62]. Building on this, the company disclosed a plan to deliver a 20-exaflop AI supercomputer within two years, also based on the Monaca chip [36-38]. The next-generation processor, a 1.4-nanometre device featuring a 256-core CPU, a 128-core CPU and an integrated NPU for AI inferencing, was described as the world’s first of its kind [42-43][55-58]. These chips are optimized for data-centre efficiency and include confidential-computing capabilities, reinforcing the security aspect of sovereignty [40-41].
For inference, the speaker distinguished between workloads: smaller- and medium-sized models can run on-premise using the NPU, whereas very large models may still require GPU-based or hybrid architectures [??].
Quantum pillar
Fujitsu outlined an ambitious quantum roadmap that the speaker described as a leading effort: a 250-logical-qubit system by 2030, a 1 000-qubit machine scheduled to go live next month in Kawasaki, and a 10 000-qubit machine expected within three years [62-71][65-68][70-71]. He also noted that Fujitsu designs its own quantum control electronics and is investing in advanced cryogenic cooling technology to support these systems [??]. The convergence of quantum computing with HPC was presented as a way to enable mission-critical AI workloads, shifting quantum from a standalone technology to an integral component of a hybrid compute ecosystem [69-73][S25][S39].
Networking pillar
Fujitsu unveiled a 1.6-terabit photonic switch (with a future 3.2-terabit version) that delivers low-power, long-reach (up to a thousand kilometres) optical transmission with low latency [75-76]. The switch is coupled with an open-RAN orchestration layer that can dynamically move AI workloads across data-centre and edge sites [75-76]. Fujitsu is one of only a few companies that offers both high-performance photonic switching and wireless solutions, enabling end-to-end AI-ready connectivity [??].
Software pillar
The software stack was portrayed as completely open source, eliminating vendor lock-in and allowing customers to fine-tune models to their specific needs [44-52]. It is optimised for AI, data-centre workloads and high-performance computing [44-48][49-51]. Domain-specific platforms sit atop Fujitsu’s proprietary security layer: Takane, a large-language-model platform, and Kozuchi, an AI-agent framework [77-84][85-86].
Edge and physical AI vision
Fujitsu described the “Kozuchi physical OS” as an operating system that embeds intelligence directly into robots, drones, medical devices and other edge equipment, allowing them to retain memory and operate autonomously [86-93][87-90][91-93]. This edge-centric approach unifies compute, networking and AI software into a single consumable platform, extending sovereign AI capabilities to the network periphery.
Services and ecosystem
The speaker emphasized that Fujitsu’s large services organization integrates hardware, software, networking and consulting to deliver a turnkey sovereign AI platform for customers [??]. Partnerships with AMD, Lockheed Martin, Supermicro and various robotics manufacturers were cited as ways to broaden the ecosystem, ensure interoperability and accelerate delivery of “physical AI” solutions across multiple industries [95-99][97-99].
Transition
The session concluded with the speaker handing the stage to Speaker 2, who announced an upcoming fireside chat featuring executives from CDAT and Intel and asked the audience to clear the stage [100-104].
Overall, the talk positioned Fujitsu as a one-stop provider of sovereign AI infrastructure, combining Japanese-made, security-focused hardware, an open software stack, a quantum roadmap, high-capacity photonic networking and a robust services ecosystem to deliver flexible, independent AI solutions for critical national and industrial applications.
AI commerce. What I’m going to talk about is something that was discussed in the plenary session yesterday as well about sovereignty. And I believe something like sovereignty is very, very important for countries like India, which are trying to eke out a path in leading AI and being dominant in AI. Now, what is sovereignty, first of all? For us, it is being flexible. And being secure, right? So you want ownership of your data. You want to control that data. But you also want to have flexibility to manage that data, create models that meet your needs, that doesn’t have to be reliant on third party overwhelmingly. And you can modify and tune that data, right? Modify and tune those models.
So Fujitsu is on a path to – we’ve always been an innovative company, and we have a long history, and I’ll talk briefly. But how do we make that sovereign? And that’s what I’m going to talk about today. So Fujitsu, some of you might not know it. I mean, we have a 90 -year -old history, right? So we are a pretty old company. We have our roots all the way in technology. And if you look at some of the things that are demonstrated here, one megabit DRAM, for example, right? Of course, we were one of the pioneers of mainframe business along with IBM. In recent past, we’ve announced, which we will be shipping very shortly, the world’s first two nanometer servers, ARM -based servers.
We announced for quantum, if you are not aware, which I will be talking about shortly as well, we have the world’s leading quantum roadmap here that we are going to deliver. Same on networks. And our U .S. brand that we created in 2021, that effectively brings all of Fujitsu’s solutions. to be consumed by our customers. Now, how does this work in the context of AI? And why is this relevant in the context of AI? And that’s what I’m going to talk about. So, to effectively drive artificial intelligence, you need three key components, right? You obviously need software, you need compute, and you need three networks, right? If you don’t have those three, you can’t really build an AI platform that will suit your enterprise needs.
And our focus on sovereignty here is really being independent in all of these three areas and give customers a choice. We are a Japanese company. Our technology is made in Japan and that’s where we find ourselves at a very interesting point because we are a choice to a lot of American companies as an alternative. So if you’re looking for leading edge computing technology, leading edge quantum technology, leading edge network technology, leading edge AI software technology, agentic technology, and there’s an end user application on which you can build an AI platform in the area such as defense, government, healthcare, manufacturing, finance, where you do care about privacy, this becomes very, very important. Now how do we actually drive that?
Some of the speakers talked about commerce, which are big, but at the end of the day, if you don’t have a platform that helps you deliver that, you’re never going to be sovereign, you’re never going to control the AI business. Fujitsu has a couple of areas that we are focused on, as I mentioned, computing. If you think about CPUs, you think about AMD, you think about Intel, we were, until two years ago, we had the fastest supercomputer in the world for five years running. And we announced that we will be building a 20 AI exascale AI supercomputer in about two years from now, which will be driving pretty much AI application, AI workloads. This will be powered by our Fujitsu Monaca chip, which is a two nanometer chip.
It’s built in Japan, and it is completely ARM -based, highly power efficient, focused on data centers to reduce power efficiency. Okay, and it has confidential computing built at the hardware level to drive security. Now, this comes out, the servers come out in about two months from now, the test servers. It’s ARM -based. The follow -up of this is a 1 .4 nanometer, and that will also be the world’s first 1 .4 nanometer, which has two versions, 256 -core CPU plus 128 -core CPU plus an NPU to drive exactly what India needs, sovereign AI models focused on inferencing. And this is something that I believe we will drive a lot of value in countries like India as well as Europe. I’m not going to go into this in detail, but this stack is a completely open software stack.
I just want you to remember, it’s a completely open software stack. There’s nothing locked in. There’s nothing. You don’t get locked into a Fujitsu stack. All the software that you see here, it’s completely open. It’s focused on AI. It’s focused on data centers, and it’s focused on HPC. This is what you need for AI, right? All the key areas are a lot of open source software that we have fine -tuned to work on this process. This can help you drive your AI workloads today on the Monaco servers. As I mentioned, what’s coming? There are two versions, the 256 -core CPU plus 128 -core CPU with the NPU on it. The NPU is focused on AI inferencing. You will see a lot of work going into inferencing moving forward.
And especially when you talk about sovereign, this will become extremely important, especially with small language models and medium language models. So you can contain that in a private or a semi -private environment that you can choose. Obviously, if you want large language models, you can choose what is going on on GPUs. And you can obviously choose the Monaco GPU hybrid architecture as well. Now, for those of you who might not be aware, we are extremely highly invested in quantum. We need quantum in Japan. I would say we are probably one of the top three players in quantum worldwide. We have announced a 250 logical qubit roadmap by the end of 2030, which is ahead of any other company in the world that I know of.
We make our own control systems. We are going to focus on driving the cooling systems as well. And this is going to become extremely important as you go ahead. Quantum plus HPC together driving mission -critical AI workloads. The 10 ,000 -qubit machine will go live in about three years from now. Next month, the 1 ,000 -qubit machine goes live in Kawasaki, Japan. As I mentioned, you would have HPC and Quantum working together to drive AI workloads. This is how computing will be consumed moving forward. And the software stack that we are working on, it will make transparent to you and users to use to consume compute and the workload can be optimized to whichever computer you want. Now, I’ll briefly talk about the networks because that’s the other part of the puzzle.
And finally, I’ll talk about the software for AI. photonics and wireless we’re one of the probably want two companies that does both no cares another one right and we are doing a 1 .6 terabyte switch that travels that is highly power efficient that drives about a thousand kilometer this distance distance on this long range transmission low latency low power consumption that’s the beauty of the switch right and and we are we will go on to 3 .2 tera this is very strong implicate implications and data centers that are being built in India as that would be highly highly power hungry and you would need to connect that through optical fibers and same with the wireless mobile systems okay now what we do is we also connect with open RAN and the network orchestration stack to bring the AI workloads move them in a highly efficient manner and we’re going to do that by using the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the RAN and the This is the third part that really brings everything together, the AI software stack.
Fujitsu, as I mentioned at the very beginning, we are focused on sovereign. When we talk about sovereign, it’s got to be domain specific, something for defense. If you’re making nuclear plants or submarines or healthcare, this is not the data you want to put on public cloud. You want to define and build these domain specific models. Second, you need to have flexibility. You, as a company, should be able to fine tune these models to your own benefit, to your own needs. They need to be highly secure. These are the three key areas that we are focused on using what we call our Takane, the large language model platform, as well as our AI agent tech model from Kozuchi, powered by the security platform that we have built within our own research teams.
It’s a complete platform that you can use to build your own applications. Thank you. Again, I won’t go into details on this. but this is a platform that also uses third -party tools and you see on the extreme right where we have issue with their government manufacturing health care finance applications so Fujitsu has a fairly large business and services which brings all of this together so we are not just selling you pieces of technology we are selling your total solution or we are asking you to use a total solution here from compute all the way to networks and the application stack together and this is our vision that we want to continue to build on this continue to bring this to other to the end customers as well as users now where are we headed right we see all this converge in the physical AI platform space and what we are building is Kozuchi physical OS which will have the intelligence based on the brain intelligence for the robot and what that means is robots tend to forget And what we are working on, some intelligence work and research so that robots can continue to remember.
But then this technology, the compute networks, as well as the AI platform stack, comes together in edge devices. Robots are one example, but even drones or medical devices or your healthcare on your iPhones. That’s where it will all come together. And that’s the world we are aiming for. That will bring together the AI agentic platform together. That will bring the security platform together in the complete platform that could be consumed for our end users, our companies. And you can choose to play in a part that is comfortable for you. And we are obviously going to partner with a lot of different companies on this. So, as I mentioned, the software, compute, network, the three pillars.
And we are going to be able to do that. We announced in October last year, our CEO Tokita and Jensen were on stage together announcing a huge partnership on physical AI, where we’re partnering with different robotics manufacturers. So it’s working with AMD, working in defense with Lockheed Martin, Supermicro. So this is something
Thank you. Thank you so much. For the next session, we have a fireside chat between Mr. Vivek Kaneja, Executive Director, CDAT, Mr. Nitin Bajaj, Director, Sales and Marketing, Intel, and the session will be moderated by Mr. Aman Khanna, Vice President of the Asia Group. May I request all the speakers to join us on the stage, please? I also request everybody to please clear the pathway. May I request the audience to please clear the pathway?
“AI sovereignty is defined as the ability to be flexible and secure—own and control data, fine‑tune AI models, and reduce reliance on third‑party providers.”
The knowledge base explicitly describes AI sovereignty in terms of flexibility, security and reduced third-party dependence, matching the report’s definition [S8] and also notes the focus on control over data, models and security measures [S45].
“Sovereign AI requires three independent layers – software, compute, and networking – and Fujitsu, as a Japanese‑made provider, offers a genuine alternative to dominant American vendors.”
A Fujitsu representative states that sovereignty means independence in three areas and highlights the company’s Japanese origin as a choice against American suppliers [S26].
“Linking AI sovereignty to national security and economic independence.”
The broader discussion connects technological sovereignty with economic independence and strategic security, emphasizing the risk of “renting intelligence” from foreign providers [S46].
“Emphasis on reducing third‑party dependence as a core aspect of AI sovereignty.”
The knowledge base highlights a nuanced framework that balances selective control with collaboration, underscoring the importance of limiting reliance on external vendors [S21].
“Incorporation of hardware‑level confidential computing to protect data in Fujitsu’s compute solutions.”
Fujitsu’s security-by-design approach, integrating security into software and hardware development, provides additional context for the confidential-computing claim [S17].
The transcript contains a single substantive contribution (Speaker 1) that outlines Fujitsu’s vision for AI sovereignty, emphasizing data ownership, open hardware/software, quantum‑HPC integration, high‑capacity photonic networking, and end‑to‑end AI solutions. Speaker 2 only provides a procedural hand‑over to the next session and does not present any substantive policy or technical arguments. Consequently, there is no demonstrable overlap or shared stance between the two speakers on any of the listed arguments.
Very low – the only points of convergence are procedural (both participants are part of the same event). The lack of substantive agreement limits any immediate implications for policy or technical coordination on AI sovereignty.
The transcript contains a detailed presentation by Speaker 1 on Fujitsu’s sovereign AI strategy and a brief logistical hand‑over by Speaker 2. No substantive policy or technical positions are offered by Speaker 2 that could be compared with Speaker 1, resulting in an absence of identifiable disagreement or partial agreement between the speakers.
Very low – the two speakers do not present conflicting viewpoints, so the discussion proceeds without contention, implying smooth transition but offering no debate on the AI sovereignty topics.
The discussion was driven by a series of strategically placed, thought‑provoking statements that each expanded the notion of AI sovereignty from a high‑level definition to concrete, technology‑specific commitments. By first framing sovereignty as flexibility plus security, the speaker set a lens through which every subsequent claim—whether about open software, exascale compute, quantum roadmaps, ultra‑high‑bandwidth networking, domain‑specific model requirements, or an integrated edge OS—was interpreted as a step toward independent, secure AI capability. Each turning point introduced a new pillar or application layer, progressively deepening the conversation and reinforcing Fujitsu’s narrative of offering a complete, sovereign AI stack. This layered approach kept the audience engaged, shifted focus smoothly across topics, and culminated in a holistic vision that ties all components together, thereby shaping the overall discourse around a unified, sovereign AI ecosystem.
Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.
Related event
