Join us for an engaging conversation with Bill Tai, Chairman of ACTAI Global, as we explore the transformative world of artificial intelligence and data center infrastructure. This discussion, hosted by John Furrier of theCUBE, examines the dynamic shifts in computing systems, particularly focusing on the emergence of AI factories and their relationship with large-scale data centers.
In this insightful session, Tai shares expertise on the evolution and integration of AI, photonics, and optical interconnects. They discuss the pivotal role of NVIDIA in advancing matrix math and graphics processing unit (GPU) technologies, making the company essential in this new computing era. Additionally, Furrier of SiliconANGLE Media highlights these groundbreaking developments.
According to Tai, the future of AI is rooted in multidimensional computing systems resembling neural networks. They emphasize the structural changes required to support these AI factories, highlighting aspects such as NVIDIA's strategic role and the integral relationship with energy consumption. The conversation also addresses the economic impact of these shifts, including potential market opportunities for companies such as Dell and Oracle.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers 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: AI Factories - Data Centers 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: AI Factories - Data Centers of the Future
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: AI Factories - Data Centers 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
Bill Tai, Hut8
Join us for an engaging conversation with Bill Tai, Chairman of ACTAI Global, as we explore the transformative world of artificial intelligence and data center infrastructure. This discussion, hosted by John Furrier of theCUBE, examines the dynamic shifts in computing systems, particularly focusing on the emergence of AI factories and their relationship with large-scale data centers.
In this insightful session, Tai shares expertise on the evolution and integration of AI, photonics, and optical interconnects. They discuss the pivotal role of NVIDIA in advancing matrix math and graphics processing unit (GPU) technologies, making the company essential in this new computing era. Additionally, Furrier of SiliconANGLE Media highlights these groundbreaking developments.
According to Tai, the future of AI is rooted in multidimensional computing systems resembling neural networks. They emphasize the structural changes required to support these AI factories, highlighting aspects such as NVIDIA's strategic role and the integral relationship with energy consumption. The conversation also addresses the economic impact of these shifts, including potential market opportunities for companies such as Dell and Oracle.
In this theCUBE + NYSE Wired exclusive conversation from AI Factories – Data Centers of the Future, Hut8 chairman Bill Tai joins theCUBE’s John Furrier to explain why AI factories mark a structural shift from linear to multidimensional computing. Tai breaks down CPU vs. GPU architectures (matrix math at scale), how clusters of tens of thousands of GPUs form “giant brains” and why the data center is evolving into a single working computer connected to others across a core-to-edge fabric. He analyzes the announced NVIDIA–OpenAI buildout and the ripple effects t...Read more
exploreKeep Exploring
What are the current trends in the evolution of computing infrastructure among large enterprises?add
What are the implications of AI factories on traditional computing and enterprise data centers?add
What was the impact of the announced deal between NVIDIA and OpenAI on the stock market?add
What are the challenges associated with setting up and running large-scale data centers and GPU clusters in Texas?add
What is Oracle's approach to data centers and AI services in comparison to Amazon AWS?add
>> Hello, I'm John Furrier with theCUBE. We are here at the NYSE CUBE Studios here on the East Coast, of course, our Palo Alto CUBE Studios there, and again, part of the CUBE and the NYSE's partnership and the NYSE Wired program and community Cube alumni, and always great commentator on big news and also big trends that are happening. And breaking it down, Bill Tai, chairman of Hut 8, among other things, the great ACTAI, global organization. Bill, great to have you on, a lot of news to break down. But more importantly, we're entering the era of AI factories where large-scale systems of computing in the world, you got NVIDIA planning to do a hundred billion in OpenAI as part of a 10-gigawatt infrastructure project. You got large enterprises looking at how to go from IT stack and rack servers and storage to large-scale connected systems, photonics, optical interconnects, the data center as a working computer connected to another data center as a large cluster. Again, the world is moving to this era. We've talked about it in Maui at your event, we've talked about it on theCUBE. I mean, go back to 2018 at a crypto event, we talked about this decentralization. So it's all playing out.
Bill Tai
>> It is all playing out. And I think one way to summarize what's happening is that we've moved from a world where computing used to be kind of linear to computing is now multidimensional. One of the reasons NVIDIA is the center of the universe now is they got a very big head start processing what's called matrix math. So all of you in high school took matrix math, you may not remember it, but instead of multiplying one number by another number, you might remember little tables where you might've had three by three or four by four, whatever, and you would learn methods to multiply all of those together at once. That is in effect what a GPU does. So some, of course, scientists ended up adapting that type of architecture to where each of the little points in the matrix might represent something like a single neuron in a brain. In a small brain, but these brains are getting larger and larger as people are able to now connect tens of thousands in some cases of GPUs into one big cluster that then really does resemble something like a brain. And I think it's leading to structural change in computing, structural change in the software and applications that can run on them, and also insane amounts of power required to run all of this stuff. These AI factories are redefining computing.>> Yeah, they are redefining computing. And again, it's all levels. I mean, the NVIDIA deal with OpenAI, the 10-gigawatt project, Jensen Huang is saying it's equivalent of four to 5 million GPUs with the first phase expected to come online in the second half of next year on the Vera Rubin platform. Okay, just put that in perspective. That's a large-scale data center. But if I'm an enterprise, I've still got my classic data centers. I have a lot of them. I might have co-location facil- ... I might want to tap into these supercomputer centers, these AI factories. But every company's rethinking, "What's my relationship with the AI factories?" You can't have an AI factory without having a data factory. You can't have an AI factory without having a software factory. You can't have an AI factory without an energy factory. Bill, this is your world. You've been living in that confluence of large-scale data, power, software, because AI factory, you can't buy the product. There's no product to buy. I mean, it's a system.
Bill Tai
>> It is. It's the integration of all of those things you mentioned into something that looks like a gigantic neural fabric of computation elements with some very, very dense areas that might resemble parts of your brain that do certain things and other parts that are the interconnect between those dense parts and other dense parts that do other functions. I think in the end, it's going to resemble a little bit of what the internet backbone looks like. On the internet, which is mostly connectivity, not computation, you have these centers, these kind of network centers, may East, may West, where all the telco lines converge. Now you're going to have basically gigantic computation AI factories placed throughout this nation and the world where those connection points of telco lines connect and interconnect those brain elements into a giant, giant pattern with both central stuff and inference edge. So we're going to see something that is just a vibrant fabric where information and intelligence, so we're moving from kind of a world of connected information to connected abstracted intelligence, which will be available at any of the edge points that can come from anywhere.>> Yeah, that really summarizes the architecture. You mentioned matrix math earlier about NVIDIA, and you look at the roots of NVIDIA, Bill, that was early days. I'm not saying they got lucky, but matrix math translates directly to neural networks, the same kind of function. Explain that nuance, because it's not like they just got lucky. They kept building it. And I saw a video of Jensen Huang years and years ago, he was still talking about CUDA software at that time. So this was a long game for them, but the world spun beautifully in the direction. Turns out the generative AI, the transformer, paper revolution, that tipping point really favored that architecture. So good vision, good execution, and perfect math connection there.
Bill Tai
>> Yeah. So I think it's important for people to understand the basic difference between a CPU and a GPU. Why do these different architectures even exist? So everyone is probably pretty familiar with a CPU, and when you use a calculator, you're basically the interface exercising numbers that go into a calculator in a serial fashion. And those are very, very efficient for things like the math problems you do on a calculator, but they are not that efficient for moving pictures around on a screen. So if you think about what it's like if you're using a video game, so let's say there's a car racing around your screen in a street corner where you've got light shining on it that has to understand the dimensions of the car and the angle at which the car is moving, a CPU would try to create that picture or render that picture one pixel at a time. So it would have to do a very intense series of calculations with the angle and the light source and the motion and do a set of calculations for every single dot on your screen. What a GPU does is it takes it in blocks. So there are these blocks that are applied to that matrix math that we talked about, and they multiply it by another set of blocks or another matrix that might have the angles and other things embedded in that. So it's a quick transformation of, relatively speaking, huge amounts of data. And so I think because the architecture allowed multiple little points to be calculated all at once against many multiple other points, it became a much better architecture to kind of resemble the neural patterns of a brain where you have many, many, many neurons connected, interconnected by many synapses. So GPUs then, they're performing incredible amounts of calculation at rates that are just thousands and thousands of times faster than a CPU could do. And without that kind of architecture, you literally just could not do the kinds of things ... I guess you could do them with a CPU, but it might take a thousand years to do something that you can do in minutes.>> So I have to ask you about the computer revolution. You talk about the internet revolution. Companies like Dell Technologies, they're a partner with NVIDIA's AI factory. They make servers, so they love it. They're selling more gear. A lot more hardware doesn't necessarily go away, it's just reconfigured differently. Talk about the role NVIDIA plays because they're if not the largest customer of TSMC, they have the allocation, they have the demand. Now NVIDIA's got their production rolling first. Of course, everyone's clamoring, "I need to get more NVIDIA." Well, now they're coming off the line, but not like they're letting up on the supply chain. So you're starting to see that shift. Now, this deal with OpenAI that's been announced and other deals are going down too, similar way. The factories are being built.
Bill Tai
>> The stock market sees it instantly. On the announcement of the deal today with NVIDIA and OpenAI, and just for those that did not catch it this morning, it's an up to $100 billion funding by NVIDIA to fuel OpenAI's data centers, which might be deploying up to four to 5 million GPUs. And it's a 10 gigawatt amount of electricity cumulatively. It's massive. On that announcement, TSMC went up $10. So I think everybody knows whatever NVIDIA's selling, it's all made at TSMC.>> What's that, Ronald Reagan, the trickle-down effect. Trickle-down economics, what was it?
Bill Tai
>> Yes.>> Yeah, I mean it goes-
Bill Tai
>> But in this case, it's not trickling down to the masses. It's trickling down to one company in Taiwan, which is a massive company also that's growing at 30 to 40% year over year, mostly because of this AI demand, not just from NVIDIA, but also from much smaller players like AMD.>> Talk about what that means and it's good to lay that out. The origination is TSMC, obviously. You got Silicon players, you got the fab. They go up and they make NVIDIA great. They're doing a lot more and they're doing stuff with other folks too. I shouldn't count out Broadcoms of the world. But then now you have a second, another layer of enablement. That's where we're starting to see these factories, large-scale data centers in Texas, monster OpenAI tied in with NVIDIA, Dell will win big on this. But there's still yet more work to do. You got to nail the power equation and you got to get the software right, because it's not like you can just load Linux on these things. Now, you can load Linux on little, many parts of it, but it's not like a server in the old days, "Hey, here's a server. Bill, load Linux on, then we'll run it and then load apps and it runs stuff." No, no, no. The software piece is huge.
Bill Tai
>> Both the interconnect, so the networking element and the software are obviously very critical. And when you're running these big GPU clusters, oftentimes the point of failure is not the GPU, it's the moving of the data around between the GPUs. So it might be something that didn't, kind of glitched a little bit or just didn't work right in the interconnect. So I think companies that provide that are increasingly valuable at this stage. And they are companies. Broadcom is making some things that are alternatives to GPUs, custom silicon that does some of that function, but they're also, they and Marvell are kind of kings of the interconnect layer, which is the technology that connects GPUs to each other. is in that game too. All those companies are growing like crazy. And I think the equipment part of that, that would be companies like Arista Networks and Cisco, of course. And so all of those things are important. It's the chips, the servers, the interconnect, and the software that runs on top of them. And then on top of that, there's the data feed. There's so much that if the intelligence that comes out of the AI factories can't happen unless there's information to feed it. Your brain, if you lived in a dark cave your whole life but could never leave, you had no new input, you wouldn't learn a thing. So I think now we've got this flood of data, and the question is how do you pull information out of that data? And then once you've got that information, how do you feed that into the AI factory to get intelligence out as a result, both in scale and also in granular form for the individual users or customers of the enterprises that are deploying it for their customers?>> Yeah, it's the data factory. It's the apps that are running it. It brings up a great point. I think you're starting to lay out the lines of distinction between infrastructure and then disruptive enablement for software. For example, if I'm going to run a software app, say OpenAI ChatGPT, that's their stuff. Now, it runs on NVIDIA and CUDA, but let's just say I have a SaaS application or I have the Cube app or whatever I have, I got to make that AI native and then run it on the infrastructure. It's not like these big data centers that can also be used for Amazon for offloading stuff. So, I mean, people have been slinging gear around, more CPUs over here, more GPUs over there, for decades. That's just making things faster that's not supercharging the apps and turning it into a unique asset like a brain, a neural network. Talk about that distinction, because a lot of investors I talk to here in Wall Street are like, "Oh yeah, these clouds are great." They're just stacking, they're just creating factories of gear. Now that's a start.
Bill Tai
>> Well, I think a lot of the future value to be had is going to show up in companies that basically increase productivity a lot. And you're starting to see whispers of that. Companies like a Cursor, or what Windsurf was before it got acquired, or Lovable or even Canva, there's a great application, right? Because Canva, Canva has CanvaGPT. I think earlier this year, like February or so, it was the second most used AI application next to ChatGPT ahead of Google, Gemini, and DeepSeek, and all the other companies you know of, because they really speed up your work for something that is pretty common, high volume, and straightforward, which is creating designs. So if you want to do a restaurant menu or a new flyer for a show or an invitation or any of your branding for your company, that historically has been a manual process where you get on to Canva or a few other companies and provide things like that, and you would basically configure your design and change the colors and change the font and kind of tweak and tweak and tweak. And now you can literally, with a prompt, just describe, "I want an Italian restaurant menu with three appetizers, five desserts, 20 hors d'oeuvres," boom, bam, and it's done for you. And you can iterate. So I think the AI behind Canva and other things like that are taking what might be 40 hours of work and turning it into four minutes. So I think that productivity is, I think, what's going to accelerate everything. And if you have your own data, which Canva does, you're just in a sweet spot.>> Bill, that's exactly my point.
Bill Tai
>> companies that don't have their own data, you might've seen the Anthropic lawsuit. So Anthropic agreed to settle some kind of data stuff for quite a big number.>> Copyright thing.
Bill Tai
>> Because, yep, there's a question out there. Well, the big LLMs, OpenAI, Anthropic, whose data are they really using? And are there rights assigned to that data that they didn't pay attention to previously? So I think companies that have their own data or rights to their customers' data, they have a huge advantage in this window.>> That data moat's huge. Your point about Canva though is really right on, because that's an example of them sitting on top of the infrastructure and they could leverage all that benefit.
Bill Tai
>> Yes, yes.>> So take that to the mainstream. First of all, Canva's a use case. First of all, side story, not to get sidetracked, Canva, I did a tweet and an Instagram post and a LinkedIn post on the eve of Figma's IPO. I think I might've mentioned, because I saw a text that you wrote about, man, this is going to be big. I predicted Figma was going to be massive, which it was. But I said, "If you think this is going to be big, wait for Canva to go public." Well, I got kind of stink-eye a little bit inside the exchange that next day, maybe they were thinking I was trying to promote Canva on their Figma IPO. I got some weird looks, but I think they had the big day. But I was trying to point out that Figma was big, but Canva's bigger and it's not yet public.
Bill Tai
>> Canva is roughly two and a half to three times the size of Figma, and I think it's been growing faster as well. It's been profitable for eight years, where Figma I think has been profitable for a year. So they occupy slightly different layers of the design stack. Figma is a fantastic product for engineers doing wireframes for what their apps or products look like. Some people do use it for trading pitch decks and things like that too. But I think Canva started more grassroots and came up and occupies this gigantic space of designers all over the world doing everything from logos to pitch decks to->> Yeah, more ....
Bill Tai
>> presentations to whatever. It's like the main product for cloud-based graphical design. And I think there's an economic industry structure change in that now that the internet is pervasive and available in many countries where labor costs are lower, Canva has become an economic empowerment engine for hundreds of millions, if not a billion people, to basically be productive, just individually connected to the internet to sell their work in digital design.>> I mean, different markets, but the point was is that the numbers are bigger and Canva's still private.
Bill Tai
>> Yes.>> Any word on their IPO.
Bill Tai
>> Well, you, the CFO of Zoom moved over to become the CFO of Canva, and I think I've seen it publicly stated that her role is to really, really build a world-class finance and accounting team. ->> That sounds like an IPO to me, Bill. It sounds like readiness....
Bill Tai
>> its role as a public company if and when it's ready. So she's there for a reason.>> Yeah, readiness, sounds good. Bill, thank you so much. Comments on just in general, what you're seeing in the market, the energy piece you've been active on. For the folks that want to see more of Bill's narrative around this, watch theCUBE videos with him, he's really nailing the economic side of the energy. It still is bounding the capabilities on AI and crypto specifically. So any latest thoughts on the market?
Bill Tai
>> Yeah, I'd say years and years ago ... So I've been involved with some of the very early, very large Bitcoin mining companies, namely a company called Bitfury. Over in the early days between say 2012 and 2015, Bitfury at times had 40% of the world's Bitcoin mining infrastructure out there. And because of that, a lot of the techniques that that company and others in the Bitcoin mining space pioneered to build massive computational data centers at very big scale. Those techniques are becoming kind of standard today for what is needed for AI. So Bitfury had immersion cooling going on eight years ago, liquid cooling years before that. Of course, liquid cooling was always around, but Bitfury was doing it at scale. So a lot of the Bitcoin mining companies of old have migrated into this space because they naturally already had the skill set and the knowledge of how to build ultra dense compute, massive scale data centers. CoreWeave literally started as a Bitcoin mining company and made a pivot. We also did that at Hut 8. Years ago, we bought a bunch of NVIDIA GPUs, I think 2020 or so, and then now we have a subsidiary that does a GPU data center. It's called Highrise.ai. And Hut 8, which was a subsidiary of Bitfury that went public on its own about seven years ago, it's now becoming an energy infrastructure company, and you're going to have the CEO, Asher Genoot, on your show, I think, in a couple of days.>> Yeah.
Bill Tai
>> But I think all of those techniques are now things that have paved the path for these AI factories of the future that are just going to get bigger and bigger and utilize more and more of the pioneering techniques of the Bitcoin mining companies of old.>> And final advice just for enterprises, what's your take on the enterprise? Because they'll have many factories compared to the big factories. This is basically now data center as one unit of a computer, super computer. Thoughts on enterprise? Their mindsets clearly slow on the uptake, 80-20 relationship between hyperscalers, neoclouds, and 20% enterprises, roughly our guesstimation on theCUBE Research. That we think is going to increase radically in the next two years.
Bill Tai
>> Oh, absolutely. I think as time passes and people find applications that can help the productivity of their customers, as Canva is doing with their customers. Zoom, for example, just had their annual Zoomtopia a few days ago, and the AI companion in Zoom Workplace is just fantastic. So there's so many companies, ServiceNow is one of the leaders in this space, that are basically taking applications that they are delivering to their customers, making those lower cost, more scalable, more powerful, using AI. And I think they themselves, if they get really big, might have to build out data centers, but for the most part, I think in the earlier years, they can basically just use some of the GPUs, they'll have to pay from, of course, but they'll use the GPUs that are being deployed at companies that are growing rapidly like , and you're starting to see a bunch of these Bitcoin miners. If you look at the stocks, the shares of Iris Energy or Cipher Mining or Hut 8, over the last four or five months, all of them have been building out very large data centers with a lot of power that all of the hyperscalers are looking at, and it's pretty clear from today's announcement, if they got to find another 10 gigawatts, it's going to have to come from somewhere. So all of those companies are candidates.>> Final question for you. I heard someone say to me, "Oracle's involved in OpenAI and NVIDIA, but they're the GPU. What is Oracle's role? Dell's involved with AI Factory. So what's the role of Oracle?" NVIDIA makes their own GPU. Oracle makes no GPUs. They have a database. Dell makes servers. They license GPUs, they put it in their systems. What is the relationship between the ecosystem when you talk about the hardware guys and the databases, because they're partnering?
Bill Tai
>> Well, Oracle, yeah, Oracle is becoming kind of like, it's a very different architecture, but it's a little bit like the role of AWS. Now, Amazon AWS does make their own compute cards, but they have to buy a lot of the silicon from outside vendors. They buy some servers at scale from other vendors. So Oracle basically is buying these GPUs in racks and installing them in big data center facilities that they acquire power for, and then they sell those services. The access to that integrated system, the AI factories that they're building, they lease out that capacity to either OpenAI or companies like Zoom.>> And the role Dell plays, the server, they still need compute? It's not always the GPU, is it? Or is it?
Bill Tai
>> There's still a lot of CPUs in those too. Every single GPU card from NVIDIA has a CPU on it that is sort of like the symphony conductor with the GPUs being the instruments. The CPU does have to instruct all of the other cards together what to do and when to do it.>> Bill, thanks so much for taking the time to break down the news, of course, share your perspective on the AI factories, of course. This is changing from large scale factories with OpenAI building up massive centers to small and the enterprise, just normal data centers, or data center classic, and, of course, the energy's key. Thanks for spending the time.
Bill Tai
>> Thank you, John. Take care.>> All right. Bill Tai, Chairman of Hut 8, among other things. Really has the finger on the pulse as the allocation of TSMC shifts to the allocation of NVIDIA, which shifts to the allocation of energy, all integrated into these AI factories. Of course, just kicking off our series here on theCUBE of AI Factories, the future of the data centers and data clouds and neoclouds. For John Furrier, Dave Vellante. Thanks for watching.
