SemiWiki.com - Podcast EP339: Unique Scalable, Power-Efficient AI Technology from EdgeCoritx with Dr. Sakya Dasgupta

Episode Date: April 10, 2026

Daniel is joined by Dr. Sakya Dasgupta, CEO and founder of EdgeCortix. He is a seasoned technologist who has been at the forefront of artificial intelligence systems for the past two decades. His experience spans public research institutes such as The Max Planck Society and RIKEN, as well as leading corporations like Microsoft… Read More

Hello, my name is Daniel Neni, founder of Semaiwiki, the Open Forum for Semiconductor professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Sakiya Dasgupta, CEO and founder of Edge Cortix. He's a seasoned technologist who has been at the front line of artificial intelligence systems for the past two decades. His experience spans public research institutes such as Max Planck Society and Riken, as well as leading corporations like Microsoft and IBM Research.

This journey led to the founding of Edge Cortex, a company committed to optimizing AI processing at the edge, while minimizing power consumption. Sake observed that many powerful AI solutions were ineffective in edge scenarios due to excessive power consumption. Unlike most AI companies that traditionally prioritize silicon devices and adapt software afterwards, Saki believed that there had to be a better approach. Thank you for joining us today, Suckin. Thanks for having me, Daniel. To start our conversation, can you tell us what brought you to edge cortex?

You know, I'd love to hear a bit of a background story on you and the company. Sure. I think you've summarized it fairly well in the introduction itself. Then the better part of the last two decades, building frontier AI models from research all the way to deploying them in production systems. In that journey, I've not only worked on building these AI models itself, but also kind of witnessing firsthand the challenges we face when we try to deploy them at scale,

especially across systems that are constrained by power, size, and many cases require real time per se. Especially my own experience with the robotics domain, what we now have come to call physical AI, convinced me that fundamentally in the last 67 years where the industry was, it required some first principles or ground at thinking, where one needs to build a software and the AI semiconductor or AI processor together from the ground up, really co-designing them, so that we can head on tackle the challenges of making AI processing at this edge landscape, a lot more energy efficient than what it is. And really, without sacrificing the solution aspect that a customer would be accustomed to. That means not making it more painful for the customer

to deploy the solution on a new type of hardware, but really building hardware and software together, where deployment becomes seamless. So that's really kind of the, you know, my own journey and what led us to founding edge cortex. Interesting. So can you share a little bit more about edge cortex and your core product, Secura 2? Secura is cherry blossom? That's absolutely right. So Sakura or Sakura 2 is, you know, it's named after the cherry blossoms in Japan, where we are headquartered as well. And just like, you know, cherry blossoms, you know, every year they bring a renewed energy and joy to the overall society, we imagined

Sakrit 2 to effectively do that to the industry, especially across the overall edge of landscape. It is one of our flagship products today in the market on the hardware side, but what we really offer is an integrated platform from software all the way to the silicon, while Sakrit 2 is designed as an accelerator or what we call in an industrial co-processor, effectively enabling customers to take the device and augment an existing system rather than replace it. Think of it as 60 trillion operations per second capable device, which is able to operate in under 8 watts of power adoption. As a result, it can bring the value of the latest multi-modal applications like large language models, language vision models, or what we now call Genitive AI, including traditional vision applications directly to device. devices and systems that are constrained by power and size.

Your robotic systems, aerospace systems could be as constrained as a satellite or a space rover operating in a harsh environment like space. And across that gamut, Sakura 2 really was designed to provide that efficient AI processing without requiring the customer to entirely change the system. They can take our M.2 modules or PCI cards, a small form, to solution, augment the existing system, essentially added onto the existing system, and then our software does the rest of the heavy lifting, enabling our hardware to easily talk to whatever is the host processor, be it arm x86 or even risk five.

Interesting. So SACRO2 is small, highly energy efficient and high performance. I imagine there's significant interest in leveraging small compute boxes or creating combination solutions for or Sakura 2 and other processors to deliver on the customer needs. Can you tell us a little bit more about that? Absolutely. And that's exactly what we've been witnessing in the last few years. We essentially announced Sakura 2 a little bit over a year, and then we have been steadily working with customers across the globe,

deploying that across many customers from early phase to late stages. And what we have witnessed is the common theme across almost all our customers, at this time spans over hundreds of customers across Japan, US, Europe and some parts of Asia as well, is the theme of requiring low power consumption, typically under 20 watts of power consumption, most of the edge applications that we are targeting at the system level, while also requiring performance that is able to run the latest large language model of vision models. Think of essentially a model like Lama 3 with 8 billion parameters. It's quite a sizable model.

We have to now run that not just in isolation, but in parallel to other additional models. Directly in a small form factor box that may sit on a factory floor or on a robot itself, that requires considerable performance, but also more importantly efficiency to be extracted from that system. That's what Sarker 2 is delivering on. And of course, in order for that to work, it has to work in conjunction with a software stack. And that's really where we think that soccer with an existing system when paired with our Mera really enables a solution that is no longer edited, but really exponential, where we are able to deliver additional value to the system that would originally not be captured

through our competing solutions or traditional GPU centric processing, which is a lot more power heavy. Right. You know, I was reviewing your website and it looks like you and the Edge Cortex team has been very busy the past year with a lot of news. Can you talk a little bit about it? Absolutely. It's been a fantastic year overall, and I think there have been a number of different things that we've been able to achieve and announce. If I look back, I think some of the prominent things that has really stood out over the year last year, we have been working with

NASA on space applications for a while now. Really kind of the our core, I would say, motivation has been to design for the hardest of constraints, both for our software and hardware, because that effectively enables it even the terrestrial applications or commercial applications to be more robust. And last year, we essentially completed the testing with NOSA very successfully with our latest Sakura 2 product, which is in mass production today, demonstrating that just like Sakara 1, which was previously tested by NASA, we came out with flying colors. What this means is that our commercial product as a cost product, what we call commercially off the shelf device, without requiring any special packaging, no additional cost, is able to withstand radiation

impacts from low Earth orbit, geosynchronous orbit, all the way to lunar missions at 10 times higher radiation tolerance compared to competing solutions in the market, which may also require special packaging. We think this is really a game changer. So that's been a fundamentally new discovery on our front at least which is now enabling us to work with partners and companies in the space area and today we are collaborating on from prototype to a system that could be deployed as soon as 2027 from Leo applications all the way to lunar mission solutions last year we also saw a tremendous amount of customer engagement overall our customer pipeline

grew significantly. We have seen a significant growth in our overall revenue numbers as well, while I'm not sharing the exact numbers, but overall, we grew our company's top line by more than two times. What we also saw last year that stood out has been a significant amount of non-dalliative funding that goes a long way for companies like us. I think one of the things that stood on last year was an award that we received from Japanese government through their Ministry of Economy, Trade and Industry, where as of today, collectively we have been able to secure almost $50 million in active projects, and around $20 million of that was awarded to us last year, which paves the way for our next generation chiplet-based solutions going beyond the

current generation Sakura 2. And last but not the least, last year was also significant for us for all the work that we are doing on dual-use applications, especially in defense aerospace space areas where we became the first Japanese company to receive a direct contract with the US Defense Innovation Unit. And today we are working very successfully with them across various stages, some of which are very close to seeing deployment with different departments across the overall US Air Force. Right. You know, I did see the report from NASA.

Very interesting. What is the long-term impact of that level of radiation resiliency testing on SpaceX Overall, you know, it really opens up the landscape of not necessarily requiring to spend millions, in some cases, billions of dollar to build a completely esoteric space-grade hardware, which may find difficulty in working with current generation software ecosystem. And that's been the concentrated effort that NASA, through a program called NASA Net, NASA Electronic Paths and Energy, packaging program essentially started working with our devices where the objective has been, can we take the commercially off the shell devices? That means devices that were originally designed for commercial applications and ensure that they can work and withstand the harsh environment on space.

Now that we have proven that with our security devices, this really showcases that even open source software ecosystem, that means the broadest amount of software today is really open source like the machine learning frameworks, whether that's PITO, TAN, NX, as well as AI models that are now being deployed across the commercial sector can, in combination with hardware's like SACRA2, which is now tested from a radiation tolerance perspective to work in space, we can bring all those models and the benefit of those AI applications directly to space environment across satellites all the way to lunar missions. And that really becomes a game changer, where we don't have to necessarily spend millions of dollars to essentially build a space-grade product which may find a lot more

challenging and difficult to work easily or seamlessly with existing software ecosystem. I also had a fantastic opportunity last year to be on a panel together with one of our customers, partners, I-Space, their chief engineer, Jeremy Fix and I, we really talked about how we are working together to find the best ways to integrate our, the SACRAT II hardware which has gone through these NASA testing across a number of different projects which in the more near term can span from supporting satellites all the way to landers and rovers enabling really AI enabled and efficiencies that we can achieve by bringing frontier models across autonomy, navigation and advanced processing with such compute platforms.

So I think that's the overall impact of the NASA results and it really opens the door for many different applications in space in a manner, I would call that the final frontier. That's amazing. So can you share more on what's next for edge cortex? You know, what's going to happen next meeting, you know, one to five years? Well, we are not stopping. We are accelerating. We are growing at a tremendous pace. And what we are now working on in terms of our next generation is what we call Saker X. And I'll give you a little bit more insight onto that. So today, in the industry, with Sakerat 2, which is already delivering value as a co-processor, it's really designed for what we call the thick edge, effectively applications that require high performance

in the tens of tops of compute, but needs to operate in few tens of watts of power consumption or lower. And that's an area where Sakritu delivers great value. But when we scale up, and today we are seeing that with humanoid robotics platforms, physical AI's growth, as well as things that are moving at the intersection between traditional edge and maybe even the data center, that's an area where you require tremendous amount of compute growth, maybe also combination with next generation communication systems

without increasing overall power consumption. That's what we are now working on delivering with our next generation Sakura X platform. It's a chiplet-based platform which enables significantly greater performance scale up compared to Sakura II without increasing the power consumption significantly. So we maintain our overall efficiency focus while delivering on the latest AI models and that's also supported by the Japan's

Ministry of Economy, Trade and industry and what is known as the new energy development organization, NEDAW's project funding that we've been able to secure over the last few years and I'm really excited to bring that to market very soon. So we'll have a lot more things to share about where you know how we are kind of bring that to market. We have all already working with lead customers even before we have, you know, started bringing the production chip to market. And I think that really showcased both the appetite and the excitement that we are witnessing from customers across industrial robotics as well as aerospace sectors who are seeing

value in higher performance, but at low power in a manner we see Sakura 2 and our next generation Sakura X to enable us to increase our footprint across the overall market. And that's really the next generation that we're working on. That's impressive. I look forward to having you back to talk about it. So final question. How do customers normally engage with your company? Fantastic question.

So we work with both our direct sales channels as well as we work with our partner ecosystem. We work with a number of different, you know, kind of distributors and resellers as well, prominently in the public domain. We work with companies like WPG, in fact, in North America. Haktou, for example, in Japan, amongst others. I'm not, of course, naming everybody. We also work with companies, for example, in Saudi Arabia. We have been partnering with a company called Eptiqar

to essentially enable our solutions to enter all of these markets. And then, of course, our own sales team and sales channels work together with our direct enterprise customers, OEMs and ODMs. We are also, you know, for folks who are interested in our hardware, as well as software, all the information is available on our website. As a result, it's really quite seamless to be able to even request access to any of our hardware. If somebody wants to test out from an evaluation platform, it's very, very simple to be able to reach out to us. Or they can also, of course, work with our distributors and channel partners when they're looking for a scale of deployment.

Thank you, Sakiya. Great conversation. It's a pleasure to meet you. And yeah, do come back and visit again so we can hear about your progress. Like once, looking forward to it. That concludes our podcast. Thank you all for listening and have a great day.