In this interview from the theCUBE + NYSE Wired: AI Factories - Data Centers of the Future event, Honghao Deng, chief executive officer and co-founder of Butlr.com, joins theCUBE’s John Furrier to discuss the explosive convergence of the physical and digital worlds through "Physical AI." Deng details how Butlr’s privacy-first technology uses thermal sensors – described as a "humble sense of touch" – to instrument the built environment without relying on intrusive video surveillance. The conversation highlights the critical shift from data scarcity to energy scarcity, with Deng arguing that the most efficient path to scaling AI compute lies in intelligently retrofitting the 80 gigawatts of existing legacy data center inventory to accommodate modern GPU densities.
The discussion explores how Butlr’s predictive thermal data enables operators to optimize cooling systems and prevent throttling, allowing them to manage high-density AI factories without the "fear-based" over-cooling that drives up energy costs. Deng showcases Butlr’s latest innovations, including wireless sensors with multi-year battery life and edge compute units that run inference locally, acting as a nervous system for mission-critical infrastructure. From enhancing safety in healthcare to maximizing thermal efficiency in enterprise data centers, Deng outlines a strategy for deploying smart infrastructure that delivers granular insights while strictly maintaining privacy and dignity.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: The AI Factory - Data Center of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AI Factories - Data Centers of the Future
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AI Factories - Data Centers of the Future.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: The AI Factory - Data Center of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: The AI Factory - Data Center of the Future
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: The AI Factory - Data Center of the Future. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Honghao Deng, Butlr.com
In this interview from theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event, Glean co-founder and CEO Arvind Jain joins theCUBE’s John Furrier to unpack what’s really working in enterprise AI today and what comes next. Jain explains why knowledge access remains the first successful AI use case at scale and how Glean’s enterprise search brings AI into everyday work. He details the past year’s lessons with AI agents – from the need for guardrails, security, evaluation and monitoring to democratizing agent building so business owners (not just data scientists) can create production-grade agents.
The conversation dives into Glean’s vision of the enterprise brain powered by an enterprise graph, highlighting the importance of deep context, human workflows and behavior to reduce “noise” and drive outcomes. Jain outlines core building blocks – hundreds of enterprise integrations and a growing actions library – that let agents securely read company knowledge and take actions across systems (e.g., CRM updates, HR tasks, calendar checks). He discusses how organizations are standing up AI Centers of Excellence, prioritizing “top 10–20” agents across functions like engineering, support and sales, and why a horizontal AI data platform that unifies structured and unstructured data – accessed conversationally and stitched together via standards like MCP – sets the foundation for AI factory-scale operations. Looking ahead, Jain says Glean’s upgraded assistant is evolving from reactive tool to proactive companion that anticipates tasks and accelerates productivity.
In this interview from the theCUBE + NYSE Wired: AI Factories - Data Centers of the Future event, Honghao Deng, chief executive officer and co-founder of Butlr.com, joins theCUBE’s John Furrier to discuss the explosive convergence of the physical and digital worlds through "Physical AI." Deng details how Butlr’s privacy-first technology uses thermal sensors – described as a "humble sense of touch" – to instrument the built environment without relying on intrusive video surveillance. The conversation highlights the critical shift from data scarcity to energy s...Read more
exploreKeep Exploring
What is the next major wave in AI beyond agents—specifically, how will the convergence of the physical and digital (physical AI) and constraints like data or energy shape AI's future?add
Can you show the Generation 2 product and explain how its wireless thermal sensor detects people and what its key features are (sensor resolution, battery life, frame rate, and the role of software)?add
What are these other devices and what are their uses and features?add
What does a data center's power consumption and heat output look like, and how do smaller edge compute devices compare and get used for real‑time AI monitoring in factories?add
>> Welcome back, everyone. I'm John Furrier, host of The Cube here in our NYSE studio. Of course, we have our Palo Alto studio connecting Silicon Valley and Wall Street. Technology's the market, and this is our AI factory series. And our next guest is a Cube alumni here on the NYC Wired Program was in Palo Alto. Now, he's here in New York, Honghao Deng, CEO, and co-founder of Butlr.com. B-U-T-L-R.com. Was IO, now they got a B round. They get the dotcom. Thanks for coming back on. Good to see you.
Honghao Deng
>> Thank you, John. Good to be back.
John Furrier
>> You were in Palo Alto when we just started covering the AI leaders. Now it's a full-blown series, AI factories. Fast-forward about a year and a half since we last talked. NVIDIA rises up, continues to dominate, thunder away on the innovation, supply chain, ecosystem, application native, AI-native applications are coming. More and more AI is coming. So it's been quite a market and the physical AI is now here. We've been talking about it for a year now. Agents are hot, okay, in the enterprise. I love agents. Agents work with data. We're seeing great value. But the real story in AI is the convergence of our physical and digital world. Physical AI is and will be the biggest wave we've ever seen.
Honghao Deng
>> A hundred percent, John. And then I think the community really looks through the corner. I think last time when we talked, the topic is really, do we have enough data to train all this large AI models? Now it's really, do we even have enough energy? Indeed, we live in the physical infrastructure and then building the roof above us, it's actually the biggest manmade product on earth. It's not really the screens. This is the largest interface we interact with every single day.
John Furrier
>> Yeah. And you guys have been building a cool product. Last time you were on, you had the sensor. It's almost, like, to me, my takeaway was instrumenting the physical world the way you did it. You guys had milestones since then. You got a series B financing. Keeping the team lean and mean. You got a hardware product. You give it away for free. You have a data driven backend, data centric business model. Give us the update. I see some new additions to the family here on the desk. Show us the new products and tell us what's happening.
Honghao Deng
>> We innovate fast, right? Speed is looking now. I do have to say that if AI is going to have a way to interact with already physical life, it's going to through the interface around us. And I see it really, not just through the swings, but through this interface, built environment. And then you don't... And it has to understand individual's needs. It really couldn't be... We really wish it shouldn't be that AI behind the cameras everywhere in your bedroom, in your bathroom. That's why we created Butlr. We really think of as a humble sense of touch to understand the warms, the colds, the safety, and also risk in the space. So very naturally, as a thermal sensor player-
John Furrier
>> Hold on, just before you go any further, explain... That was good that you kicked that. You guys are doing a lot with the environment. Touch, sensing, video. Explain the multi-modality of what's going on.
Honghao Deng
>> We don't use video at all. So that's a principle because we all don't want to live in a world that we are always surveillance. No matter if the people behind or AI behind, even AI behind the response.
John Furrier
>> No one wants a camera in the bathroom. Nobody wants a camera in the hospital.
Honghao Deng
>> No one wants that. No one wants that. So when we spin out from MIT, the reason why initially we go into healthcare use cases in senior living is because of my grandpa fell. And then, as a nurse in the family, I put cameras around him.
John Furrier
>> So he fell at home.
Honghao Deng
>> At home.
John Furrier
>> And you're like, okay, got to keep an eye on my grandpa. Yep.
Honghao Deng
>> It's like that.
John Furrier
>> What did he react to? What was his reaction?
Honghao Deng
>> He was like... He didn't say anything, but it's the family's homework for me. So he keep it quiet and then secretly just unplug everything.
John Furrier
>> So he didn't want it.
Honghao Deng
>> He obviously don't want it.
John Furrier
>> Because it was privacy for him, right? He didn't want you staring at him all the day.
Honghao Deng
>> And then, it's a quick lesson learned for me. I realized when I get to 80, I don't want my grandkids watching me over CCTV. And we all want to live with dignity. So that's why we build a sense of touch. Just like taking temperature, surface temperature data point. So under the sensor, we put the sensor in the ceiling. We can easily even implement the studio now. When I put it up, it will just see two of us sitting down and then we are all thermal blobs, just temperature. And then for obvious reason, because we have... Our raw data is, like, high accuracy temperature data point. It becomes also very natural to use that to monitor the thermal performance in data center.
John Furrier
>> So you get all the imagery you need through non-video, thermal, sense, feel. Is that right?
Honghao Deng
>> Exactly.
John Furrier
>> So that's how hospitals would want it. So if you could detect someone falling in a hospital, say they decide go for a stroll from the room or it could be someone working, you would see that in the thermal image.
Honghao Deng
>> Correct. It's not just the after fact. So that's the powerfulness of AI. It's really prediction models. So we develop a lot of specialized model into fall prevention and similar ideas into, like, data centers. So when you retrofit that data center, it's really... And then, the problem there is really like people do static control loop based on fear, essentially. They don't want the struggling of a GPU. They don't want... It's just like going to the automatic shutdown loop when it go beyond like 95 degrees C. So that's why when people set the temperature of the airflow, it's actually even 10 degree lower than the actuary standard. So just to keep it safe. So it's like really-
John Furrier
>> So they're not really hopefully optimized. Explain the data center because this is going back with your markets. That's a hot market for you. It's a mission-critical, like hospital. What actually are you providing for value? Is it the difference between highly optimized management of what's going on in the environment? Is it for detecting failures? And what's the alternative? Why are you better than what they're doing?
Honghao Deng
>> It's actually all the above you just mentioned. So you were spot-on. And then like, if you look at the efficiency when we build out all this infrastructure, obvious feedback we can get today is from all this, like, data center project being banned by the cities or being paused because when we think about it, the efficiency is not that great. It needs actual infrastructure for water, for electricity. And then, with AI taking people's job, now taking away our water and also electricity, it's just hard to swallow, right? And then, it's not the most efficient way to do it. We have an inventory, human society, we have an inventory of like 80 gigawatts of existing legacy data center. And then, to be honest, the fastest way to get GPU being turned on is actually to retrofit them. And then they-
John Furrier
>> Retrofit the data centers.
Honghao Deng
>> Exactly. The existing legacy one. Those are the inventory. And then each year we try to build another 10 gigawatts of new data centers, but like those need additional infrastructure. So the right thinking is, can we actually retrofit the existing CPU based ones into the ones that can take L40s, even H100? But the problematic part is that they control it based on the results. Just like I mentioned before in hospital, you only detect a spot of fall. It's aftermath. It's just like the way they control the air code ventilation feedback loop. It's really like, you don't know the end result, you don't have the feedback loop. That's why you just crank it up to the max. And then to be honest, there is, like, huge room to optimize to improve the efficiency and also most importantly, understand the thermal limitation. Otherwise, they don't even know.
John Furrier
>> So it's like insurance, too. You give them a predictive view of what the risks are. So if I'm retrofitting a data, just playing this out loud, okay, yeah, it looks good on paper, but there could be a blind spot relative to say some sort of leakage in a building with energy, air, or some flaw. Is that right?
Honghao Deng
>> Exactly. It's really about a predictive model to know the problem beforehand.
John Furrier
>> Give an example. Give an example.
Honghao Deng
>> Give an example is that like when the GPU is already throttling, it's already end result. There might be lack of circulation. It's just like the cooling is not efficient enough and get it to that end result. So you want to match the demand curve. You want to match how it got built up.
John Furrier
>> So you guys have the ability... So I'm just thinking if I'm the data center owner or operator or whatever, I get your Butlr product and I say, okay, it's going to measure a lot of things, but I don't need custom probes and instrumentation on say, the GPU. You say, "Hey, I'm going to monitor the whole area." And by the way, I use my data algorithms to saying, "Yeah, you care about the GPUs and, hey, it's throttling, but the air condition's on, the airflow, the power and cooling's working, but it could be something else." That would be almost impossible to detect in advance.
Honghao Deng
>> Correct. It wasn't possible before, especially if you think about . Sometimes they don't even have the data points.
John Furrier
>> It's a no-brainer for them. What were some of the reactions on those data center deployments? What's been some of the feedback from the people?
Honghao Deng
>> Oh my gosh, they desperately need it because there's no way they know whether they can upgrade it or not because they're already having the fear that I don't want the GPU, the CPU right now, the legacy one to , and then they don't know whether they can fit the newer generation ones. Now we have the full data point for them to know that what's your thermal limitation, whether you can do a retrofit on the existing rack or not, what's the recommendation? And then in this way, essentially you can upgrade the existing rack to fit in-
John Furrier
>> You're essentially like the smart data center fabric layer of what's going on in the environment of the data center, kind of, like, outside of the physical plant itself and the gear. I mean, these AI factories are basically large scale systems. The whole data center is the computer. Everything about the buildings is going to be optimized as much as 100% to support energy usage, efficiency. How can I get more out of the GPUs, which are, by the way, huge cost.
Honghao Deng
>> It's enormous cost. So it doesn't stop there after we recommend a new layout based on the thermal limitation by monitoring the historical performance. And then, because of now you introduce additional data to be preventive, to prevent the temperature and energy build up, then you want to have constant monitoring to make sure it's beforehand. You don't want to find it later on. It's already struggling. It's all about that. When you find something that's already swaddling, it becomes vastly energy inefficient because you will then crank up the ventilation, the airflow, like, turn the temperature 10, 15 degree below the standard to just cool down.
John Furrier
>> Yeah, you're basically creating an environmental change that could be managed differently. So people just go to what they see in the training manual. Turn up the AC, get more airflow or cooling, water cooling or whatever, liquid cooling.
Honghao Deng
>> Yeah. It's just like-
John Furrier
>> So data center is a big market for you. So you're doing well there. I can see that working for you. Other areas that you guys have momentum in? What else?
Honghao Deng
>> In general, it's just really about energy, right? So it's really about, again, data center is definitely hot to trot right now. There's a lot of inbound for us. Besides as really general speaking, if you look at energy as a whole, buildings itself actually consume 40% of the total energy of the whole mine plant. And then data centers actually now is only 2%. So if you look at building, even a lot of strategy in building physics can be applied to data center, too, but it also applied to all buildings. For example, you can essentially store energy into the building mass. So pre-cool it. So when the energy are cheaper, so when the vast majority of the world or all the city is sleeping, it's cheaper. Then you store the cooling, you store the heating energy into the building mass, into the air, and then release it later when you need it. Just by that, there's like 5% to 7% energy efficiency gain towards all type of different buildings, including data centers. And then if you look at that, if you gain like roughly 10%, let's say, off the cooling and heating efficiency, and then the building's already consuming 40%, the whole world's energy. And then, ventilation is actually 70%, then you can gain potentially up to 10%, which is like 3%. You're creating 3% energy for the world. And data center's only 2% and we're figuring out all the nuclear plants and also solar.
John Furrier
>> Yeah. I mean, the world's like, oh my God, we have a crisis of energy, but like it's laying around in plain sight. You guys are making that happen. Explain the products now because I get the office piece, too. I think that's going to come later after the data center. Data center is a critical path, healthcare, these are critical services. I mean, basically critical infrastructure is what you're seeing demand in, right?
Honghao Deng
>> Correct.
John Furrier
>> But office is kind of mainstream. I've heard stories of post COVID, no one's coming back to work and people are turning their campuses into data centers. Why? They have all the facilities. They have all the electricity. Everything was already kind of plumbed into the system. They just turned it into data centers, put some pods in the parking lot. Next thing you know, they're running hyper scale level data centers.
Honghao Deng
>> Exactly.
John Furrier
>> Stay at home. We'll use it for data center.
Honghao Deng
>> Yeah. I mean, the data center is actually the new workplace if you think of it.
John Furrier
>> Yeah.
Honghao Deng
>> So we did-
John Furrier
>> If agents are continuing to do their thing, yeah. I mean, you sell the CloudBot, Multbook, they're calling it. I mean, that's in the data center. Okay, show the products. I want to get these in if you'd hold them up. That was the original one. Hold it up to the camera.
Honghao Deng
>> Yes. We showed this last time, too. So this is the Generation 2 platform. So does sensor take roughly 10 by 10 surface temperature and then we translate it into location of people and also by-
John Furrier
>> It's all wireless, right?
Honghao Deng
>> All wireless.
John Furrier
>> No real... No provisioning align, none of that.
Honghao Deng
>> It is a world first and only sensor that can allow multiple years of battery life and then stream at 10 frame per second. So that's why we can do mission-critical fall detection and fall prevention.
John Furrier
>> So secret sauce is software.
Honghao Deng
>> Both software-
John Furrier
>> Pretty much and the integration of tying the software to the physical-
Honghao Deng
>> Exactly....
John Furrier
>> thermal instrumentation. It's not so much the hardware. It's like basically, it's the key sensors, right?
Honghao Deng
>> That's why it's, like, physical AI.
John Furrier
>> Yeah. That's why I get it. Love physical. You're way ahead of the curve, which I love why you came in, again and again. So now, take us to the other ones.
Honghao Deng
>> This is something we're going to announce very, very soon. So this is more into the use cases that are mission-critical. That's like for health and safety and also preventing risk, fire risk, and also understand detailed temperature performance of the equipment. So beyond just low resolution, thermal temperature data point, it got a little bit higher resolution and also higher sensitivity.
John Furrier
>> No camera.
Honghao Deng
>> No camera at all.
John Furrier
>> So that's the key.
Honghao Deng
>> That's the key.
John Furrier
>> You get the privacy, but you get all the data you need. That's what you want. And what's this bigger box here?
Honghao Deng
>> This one is really... I want to give an example of the compute.
John Furrier
>> Hold it up.
Honghao Deng
>> Yeah. This one?
John Furrier
>> Yeah.
Honghao Deng
>> So inside is just a nano, just an ordinary nano. So people always forget how much power consumption it is or in reality, what does a data center's consumption look like? Some of us still have the impression that it might just be a couple of racks down in the basement or in the office. I think the AI factory key, the word is, like, heat perfectly because of it's indeed having industry. And then that's why the size scale heat transfer, which we have a great product potentially that can help there is critical because if you think of it, so this thing, this is a small GPU server and then a normal rack, roughly like a legacy one is, like, 2000 watts, that's roughly 50 of them.
And then if you think about the heat generated by them, it's going to be a funny analogy. If you have like two of this, it's a slow cooker. And then, if you have a legacy rack, that's an oven. And then, a black wall is actually 50 of those regular legacy racks, 42 used. You need 50 of them to have the equivalent output of a black wheel, like 42U.
John Furrier
>> Right.
Honghao Deng
>> And that essentially you have like 50 ovens behind you. That's amount of-
John Furrier
>> It's a lot of compute. Tons of power. But the smaller box is for smaller use cases, right?
Honghao Deng
>> Yeah. This is essentially like we have them... We run inference at the edge. So essentially it can take care of dozens of our essential real time understanding whether somebody fall or whether equipment go over heat, whether there's a risk of fire or even leaking in the future.
John Furrier
>> So you can run models locally, do inference. You can talk to the factory, use the internet to pull down the models you need in real time. So you're really more edge specific.
Honghao Deng
>> It's really about building the nervous system, right? I wasn't saying that we're trying to build up the whole brain, but Butlr always try to be a humble nerve cell that's connected to contribute the overall large intelligence, nervous network. So we are part of it being the sense of touch. And the future will be, the way we see the future, it will be multimodal just like people, right? You have eyes somewhere and you have sense of hearing and you have sense of touch, which is Butlr.
John Furrier
>> Honghao, it's great to have you back. I love what you're working on. I think the edge with physical AI is going to be the biggest story no one's talking about. We've been talking about for a year. It's going to get more explosive. I'm expecting GTC next month to be another barn burner of an event. I think it's going to be fantastic. We'll see more physical AI and robotics, which again, is a tell to where the market's going. This is a practical example. I love how you're using the wireless, very clean. Now you just got to connect to the internet via ethernet and you talk to the factory. So it's really kind of like a transceiver layer for that. It's like almost... And this is the future. I mean, I think these kinds of boxes will be kind of AI at the edge, AI factories at the edge. Just a note in the network.
Honghao Deng
>> Totally. And the new one is like you can wire it in and also it also can be wireless. That's the beauty of it. It's maximum flexibility. And then-
John Furrier
>> And why would they do wire for, like, what? Just security or...
Honghao Deng
>> A lot of times it's just, like, for example, retrofitting existing data center, people want the speed, right? You want to put them on collecting data and know what to do. And then, when we do start to do monitoring and also having an insight, a part of your feedback loop, people want it permanently. So that's the beauty of it. You start with the wireless-
John Furrier
>> So I go for the connectivity, security, speed. So that's speed issue.
Honghao Deng
>> Best of the both world, essentially.
John Furrier
>> And wireless isn't slow either. You were saying, what's the stats on the wireless throughput that you have?
Honghao Deng
>> Oh, we have the state of the art. We're the only one that can do 10 frame per second, maintaining more than three years or better. Nobody else can do it.
John Furrier
>> On wireless.
Honghao Deng
>> On wireless.
John Furrier
>> Yeah. Great. So that one is for people who want the wire. They want this high speed, the mission-critical, no big deal, wire it up, just plug an ethernet cable into it.
Honghao Deng
>> If you do one at the beginning, we see a lot of customer Fortune 10s, they start with wireless and the wiring become part of the system.
John Furrier
>> Well, we'll keep in track. I'm sure you're going to be very relevant as the edge continues to get smarter, more intelligent. This is a great example. Again, love the no camera, humble nature. I think that's really a great decency, but it's also like nobody wants to be surveilled in these critician critical, but they want the data.
Honghao Deng
>> Correct.
John Furrier
>> You got it. Thanks for coming on. Great to see you again. Physical AI is coming. This is The Cube and the NYSC Wired program, Cube Original connecting Silicon Valley and Wall Street. I'm John Furrier, your host of The Cube. Thanks for watching.
John Furrier