Kazu Gomi has an enormous view of the technology world from his perch in Silicon Valley. And as president and CEO of NTT Research, a division of the massive Japanese telecommunications firm NTT, Gomi can control the R&D budget for a large chunk of the fundamental research that is completed in Silicon Valley.
And perhaps it’s no surprise that Gomi is pouring quite a lot of money into AI for the enterprise to find latest opportunities to reap the benefits of the AI explosion. Last week, Gomi unveiled a brand new research effort to deal with the physics of AI and well as a chip design for an AI inference chip that may process 4K video faster. This comes on the heels of research projects announced last 12 months that might pave the best way for higher AI and more energy efficient data centers.
I spoke with Gomi about this effort within the context of other things big firms like Nvidia are doing. Physical AI has turn out to be an enormous deal in 2025, with Nvidia leading the charge to create synthetic data to pretest self-driving cars and humanoid robotics in order that they can get to market faster.
And constructing on a story that I first did in my first tech reporting job, Gomi said the corporate is doing research on photonic computing as a approach to make AI computing so much more energy efficient.
Decades ago, I toured Bell Labs and listened to the ambitions of Alan Huang as he sought to make an optical computer. Gomi’s team is attempting to do something similar many years later. If they’ll pull it off, it could make data centers operate on so much less power, as light doesn’t collide with other particles or generate friction the best way that electrical signals do.
During the event last week, I enjoyed talking to slightly table robot called Jibo that swiveled and “danced” and told me my vital signs, like my heart rate, blood oxygen level, blood pressure, and even my cholesterol — all by scanning my skin to see the tiny palpitations and color change because the blood moved through my cheeks. It also held a conversation with me via its AI chat capability.
NTT has greater than 330,000 employees and $97 billion in annual revenue. NTT Research is a component of NTT, a worldwide technology and business solutions provider with an annual R&D budget of $3.6 billion. About six years ago it created an R&D division in Silicon Valley.
Here’s an edited transcript of our interview.
VentureBeat: Do you are feeling like there’s a theme, a prevailing theme this 12 months for what you’re talking about in comparison with last 12 months?
Kazu Gomi: There’s no secret. We’re more AI-heavy. AI is front and center. We talked about AI last 12 months as well, nevertheless it’s more vivid today.
VentureBeat: I wanted to listen to your opinion on what I absorbed out of CES, when Jensen Huang gave his keynote speech. He talked so much about synthetic data and the way this was going to speed up physical AI. Because you possibly can test your self-driving cars with synthetic data, or test humanoid robots, so rather more testing could be done reliably within the virtual domain. They get to market much faster. Do you are feeling like this is smart, that synthetic data can result in this acceleration?
Gomi: For the robots, yes, 100%. The robots and all of the physical things, it makes a ton of sense. AI is influencing so many other things as well. Probably not all the pieces. Synthetic data can’t change all the pieces. But AI is impacting the best way corporations run themselves. The legal department could be replaced by AI. The HR department is replaced by AI. Those sorts of things. In those scenarios, I’m unsure how synthetic data makes a difference. It’s not making as big an impact as it will for things like self-driving cars.
VentureBeat: It made me think that things are going to come back so fast, things like humanoid robots and self-driving cars, that we’ve to choose whether we actually need them, and what we would like them for.
Gomi: That’s an enormous query. How do you cope with them? We’ve definitely began talking about it. How do you’re employed with them?
VentureBeat: How do you employ them to enrich human employees, but in addition–I feel considered one of your people talked about raising the way of life (for humans, not for robots).
Gomi: Right. If you do it right, absolutely. There are many good ways to work with them. There are definitely bad scenarios which are possible as well.
VentureBeat: If we saw this much acceleration within the last 12 months or so, and we will expect synthetic data will speed up it much more, what do you expect to occur two years from now?
Gomi: Not a lot on the synthetic data per se, but today, considered one of the press releases my team released is about our latest research group, called Physics of AI. I’m looking forward to the outcomes coming from this team, in so many alternative ways. One of the interesting ones is that–this humanoid thing comes near to it. But at once we don’t know–we take AI as a black box. We don’t know exactly what’s happening contained in the box. That’s an issue. This team is looking contained in the black box.
There are many potential advantages, but considered one of the intuitive ones is that if AI starts saying something improper, something biased, obviously that you must make corrections. Right now we don’t have a excellent, effective approach to correct it, except to only keep saying, “This is improper, it’s best to say this as a substitute of that.” There is research saying that data alone won’t save us.
VentureBeat: Does it feel such as you’re attempting to teach a baby something?
Gomi: Yeah, exactly. The interesting ideal scenario–with this Physics of AI, effectively what we will do, there’s a mapping of data. In the top AI is a pc program. It’s made up of neural connections, billions of neurons connected together. If there’s bias, it’s coming from a selected connection between neurons. If we will find that, we will ultimately reduce bias by cutting those connections. That’s the best-case scenario. We all know that things aren’t that easy. But the team may have the opportunity to inform that should you cut these neurons, you would possibly have the opportunity to scale back bias 80% of the time, or 60%. I hope that this team can reach something like that. Even 10% continues to be good.
VentureBeat: There was the AI inference chip. Are you attempting to outdo Nvidia? It looks like that may be very hard to do.
Gomi: With that specific project, no, that’s not what we’re doing. And yes, it’s very hard to do. Comparing that chip to Nvidia, it’s apples and oranges. Nvidia’s GPU is more of a general-purpose AI chip. It can power chat bots or autonomous cars. You can do all types of AI with it. This one which we released yesterday is barely good for video and pictures, object detection and so forth. You’re not going to create a chat bot with it.
VentureBeat: Did it look like there was a possibility to go after? Was something not likely working in that area?
Gomi: The short answer is yes. Again, this chip is certainly customized for video and image processing. The secret’s that without reducing the resolution of the bottom image, we will do inference. High resolution, 4K images, you should utilize that for inference. The profit is that–take the case of a surveillance camera. Maybe it’s 500 meters away from the article you desire to have a look at. With 4K video you possibly can see that object pretty much. But with conventional technology, due to processing power, you’ve to scale back the resolution. Maybe you would tell this was a bottle, but you couldn’t read anything on it. Maybe you would zoom in, but you then lose other information from the realm around it. You can do more with that surveillance camera using this technology. Higher resolution is the profit.
VentureBeat: This could be unrelated, but I used to be enthusiastic about Nvidia’s graphics chips, where they were using DLSS, using AI to predict the subsequent pixel that you must draw. That prediction works so well that it got eight times faster on this generation. The overall performance is now something like–out of 30 frames, AI might accurately predict 29 of them. Are you doing something similar here?
Gomi: Something related to that–the explanation we’re working on this, we had a project that’s the precursor to this technology. We spent quite a lot of energy and resources up to now on video codec technologies. We sold an early MPEG decoder for professionals, for TV station-grade cameras and things like that. We had that base technology. Within this base technology, something just like what you’re talking about–there’s a little bit of object recognition happening in the present MPEG. Between the frames, it predicts that an object is moving from one frame to the subsequent by a lot. That’s a part of the codec technology. Object recognition makes that occur, those predictions. That algorithm, to some extent, is utilized in this inference chip.
VentureBeat: Something else Jensen was saying that was interesting–we had an architecture for computing, retrieval-based computing, where you go right into a database, fetch a solution, and are available back. Whereas with AI we now have the chance for reason-based computing. AI figures out the reply without having to leaf through all this data. It can say, “I do know what the reply is,” as a substitute of retrieving the reply. It might be a distinct type of computing than what we’re used to. Do you’re thinking that that will probably be an enormous change?
Gomi: I feel so. Numerous AI research is occurring. What you said is feasible because AI has “knowledge.” Because you’ve that knowledge, you don’t should go retrieve data.
VentureBeat: Because I do know something, I don’t should go to the library and look it up in a book.
Gomi: Exactly. I do know that such and such event happened in 1868, because I memorized that. You could look it up in a book or a database, but should you know that, you’ve that knowledge. It’s an interesting a part of AI. As it becomes more intelligent and acquires more knowledge, it doesn’t should return to the database every time.
VentureBeat: Do you’ve any particular favorite projects happening at once?
Gomi: A pair. One thing I need to spotlight, perhaps, if I could pick one–you’re looking closely at Nvidia and people players. We’re putting quite a lot of deal with photonics technology. We’re enthusiastic about photonics in a few other ways. When you have a look at AI infrastructure–you already know all of the stories. We’ve created so many GPU clusters. They’re all interconnected. The platform is large. It requires a lot energy. We’re running out of electricity. We’re overheating the planet. This isn’t good.
We want to handle this issue with some different tricks. One of them is using photonics technology. There are a few other ways. First off, where is the bottleneck in the present AI platform? During the panel today, considered one of the panelists talked about this. When you have a look at GPUs, on average, 50% of the time a GPU is idle. There’s a lot data transport happening between processors and memory. The memory and that communication line is a bottleneck. The GPU is waiting for the info to be fetched and waiting to write down results to memory. This happens so over and over.
One idea is using optics to make those communication lines much faster. That’s one thing. By using optics, making it faster is one profit. Another profit is that in relation to faster clock speeds, optics is rather more energy-efficient. Third, this involves quite a lot of engineering detail, but with optics you possibly can go further. You can go this far, or perhaps a couple of feet away. Rack configuration could be so much more flexible and fewer dense. The cooling requirements are eased.
VentureBeat: Right now you’re more like data center to data center. Here, are we talking about processor to memory?
Gomi: Yeah, exactly. This is the evolution. Right now it’s between data centers. The next phase is between the racks, between the servers. After that’s inside the server, between the boards. And then inside the board, between the chips. Eventually inside the chip, between a pair of various processing units within the core, the memory cache. That’s the evolution. Nvidia has also released some packaging that’s along the lines of this phased approach.
VentureBeat: I began covering technology around 1988, out in Dallas. I went to go to Bell Labs. At the time they were doing photonic computing research. They made quite a lot of progress, nevertheless it’s still not quite here, even now. It’s spanned my whole profession covering technology. What is the challenge, or the issue?
Gomi: The scenario I just talked about hasn’t touched the processing unit itself, or the memory itself. Only the connection between the 2 components, making that faster. Obviously the subsequent step is we’ve to do something with the processing unit and the memory itself.
VentureBeat: More like an optical computer?
Gomi: Yes, an actual optical computer. We’re trying to try this. The thing is–it feels like you’ve followed this topic for some time. But here’s a little bit of the evolution, so to talk. Back within the day, when Bell Labs or whoever tried to create an optical-based computer, it was mainly replacing the silicon-based computer one to at least one, exactly. All the logic circuits and all the pieces would run on optics. That’s hard, and it continues to be hard. I don’t think we will get there. Silicon photonics won’t address the difficulty either.
The interesting piece is, again, AI. For AI you don’t need very fancy computations. AI computation, the core of it is comparatively easy. Everything is a thing called matrix-vector multiplication. Information is available in, there’s a result, and it comes out. That’s all you do. But you’ve to do it a billion times. That’s why it gets complicated and requires quite a lot of energy and so forth. Now, the great thing about photonics is that it could possibly do that matrix-vector multiplication by its nature.
VentureBeat: Does it involve quite a lot of mirrors and redirection?
Gomi: Yeah, mirroring after which interference and all that stuff. To make it occur more efficiently and all the pieces–in my researchers’ opinion, silicon photonics may have the opportunity to do it, nevertheless it’s hard. You should involve different materials. That’s something we’re working on. I don’t know should you’ve heard of this, nevertheless it’s lithium niobate. We use lithium niobate as a substitute of silicon. There’s a technology to make it right into a thin film. You can do those computations and multiplications on the chip. It doesn’t require any digital components. It’s just about all done by analog. It’s super fast, super energy-efficient. To some extent it mimics what’s happening contained in the human brain.
These hardware researchers, their goal–a human brain works with possibly around 20 watts. ChatGPT requires 30 or 40 megawatts. We can use photonics technology to have the opportunity to drastically upend the present AI infrastructure, if we will get all the best way there to an optical computer.
VentureBeat: How are you doing with the digital twin of the human heart?
Gomi: We’ve made pretty good progress over the past 12 months. We created a system called the autonomous closed-loop intervention system, ACIS. Assume you’ve a patient with heart failure. With this technique applied–it’s like autonomous driving. Theoretically, without human intervention, you possibly can prescribe the best drugs and treatment to this heart and produce it back to a traditional state. It sounds a bit fanciful, but there’s a bio-digital twin behind it. The bio-digital twin can precisely predict the state of the guts and what an injection of a given drug might do to it. It can quickly predict cause and effect, settle on a treatment, and move forward. Simulation-wise, the system works. We have some good proof that it would work.
VentureBeat: Jibo, the robot within the health booth, how close is that to being accurate? I feel it got my cholesterol improper, nevertheless it got all the pieces else right. Cholesterol appears to be a tough one. They were saying that was a brand new a part of what they were doing, while all the pieces else was more established. If you possibly can get that to high accuracy, it might be transformative for a way often people should see a health care provider.
Gomi: I don’t know an excessive amount of about that specific subject. The conventional way of testing that, after all, they should draw blood and analyze it. I’m sure someone is working on it. It’s a matter of what type of sensor you possibly can create. With non-invasive devices we will already read things like glucose levels. That’s interesting technology. If someone did it for something like cholesterol, we could bring it into Jibo and go from there.
