Gaurav Sharma - The Future of AI Won’t Be Centralized - TOKEN2049 Dubai 2025

Hi everyone. My name is Gaurav. I'm CEO at io.net. I've been a builder for the last 20 plus years.

I've worked in startups. I've worked in big organizations like Amazon, eBay, HPRLD. And for past 10 plus year, I've been working in booking.com group and Binance. And over there, I've been building platforms for hotel and flights for more than 150 million users.

And from Binance, KYC compliance and risk platform for more than 200 million users. So, I know a bit how to build a scalable platform. And in the last few years, I've realized how powerful AI is in the hand of somebody who's building, scaling, and automating. And in the next 15 to 20 minutes, I'll give you examples and data. And we'll prove it to you that why the future of AI will not be centralized.

In the early 2000s, if you want to scale up your business, opening a new branch in a new city or a country wasn't good enough. You had to adopt a technology. You had to use internet. You had to go online. And there were businesses who did it, became billion-dollar businesses like Netflix and Amazon.

And there were businesses who resisted this, like Blockbuster. And they had a slow death. We are at the same inflection point now. But today, it's about AI. There'll be companies who will adopt this change, will flourish, will become big businesses.

And there'll be companies who will resist again, and will die. Let me make you go through a a journey of scale from last 100 plus years. In the 1900s, if you want to make a billion-dollar business, you had to have a big team of 100,000 developers or workers. You had to add good supply chain, big factories, and then you can make a billion-dollar business. In the 2000s, you need 2,000 odd workforce, adoption to internet, adoption to cloud, and you're there.

But in today's world with AI, all you need is less than 100 developers, maybe less than 10, and you can make a billion-dollar business within a year. Midjourney and few other companies are best example of it. Let's take a look at some numbers, innovation often starts the research. The best of the minds in the world are sitting in one room thinking about problems, creating concepts, proving them. And then when they graduate to the real world, they build you products.

And that's exactly what we have seen in AI. Look at this graph. In the last 5-6 years, you have seen just the papers on LLM and AI increasing from 42 to 28,400. I was reading a research just couple of days back that 90% of the papers in the software side of things today are either directly or indirectly working on AI. That's the kind of adoption is happening in the research side of things.

How does it look in the real world? In the real world, this is where we are. Almost every company, be it Google, be it Amazon, be it your OpenAI of the world, everybody is working on LLM solutions. I'm very sure five or 10 people in this room itself would be also working on LLMs. That's a kind of a growth in actual work which is happening on the AI side.

Yeah. Now, let's talk about some numbers in terms of adoption which is fundamental to the AI industry which is compute. If somebody doesn't know why and what compute does, I'll give a 30-second brief about it. When you say that you're creating a model, what you're basically doing is you're throwing a bunch of data to a hardware and telling it to learn with the help of a software module. Now, that learning for that software historically used to take months and sometimes years also.

With the invention of GPU, the same learning now happens within a day. So, then learning has become exponential. When you're using a chat GPT, you're basically using a model deployed on a GPU. That's why you're getting a good response. So, without GPU, there is no AI business.

And these numbers show the adoption of that. So, there are 22% growth in just the compute industry right now year by year. But, there's a 39% adoption of AI. What does it mean? It means there's a big delta between how the adoption is happening and the rate at which the compute itself is created.

So, today practically the rate at which we are adopting and the rate at which we are creating compute has a delta of 2.5x. And this gap is just increasing. Why? Because the number of companies who can create these GPUs are only a handful.

And there are a bunch of problems. There's a hardware problem, the raw material problem, there is tariffs issue, there is talent issue. So, in short, in the coming time, this delta of 2.5 x is not going to reduce. Now, let's look at the numbers which everybody likes.

They want to know that the money. Can we make money in this business? So, this graph shows you that the GPU as a service business, along with the cloud, by the year 2032, will become a 2.5 trillion-dollar business. Just 1% of this business is 25 billion dollars.

That's the kind of scale we're working at. The market is huge. The adoption is real. The business is there. If my word is not good enough, let's hear somebody whom you might trust.

The scaling law. Well, this last year, this is where almost the entire world got it wrong. The computation requirement, the scaling law of AI is more resilient, and in fact, hyper-accelerated. The amount of computation we need at this point, as a result of agentic AI, as a result of reasoning, is easily a hundred times more than we thought we needed this time last year. So, these are the people who basically live this life.

They have the best of researchers in the world around this problem statement. And they also undermine the usage and adoption. The best guy in the industry is saying that what they thought and what the actual adoption is, there's a 100 x difference. That's the kind of market we are working in. But yet, there's a fallacy going around.

And what is that? The newer, cheaper, faster, more efficient ML models like Deep Sea will kill the market. And I'm here to tell you, it's on the contrary, exactly the thing the industry is looking for. This is exactly the thing which will exponentially increase the adoption. Why?

Let's take a practical example because we all have lived through it. So, the mobile in my pocket today is 10 times or more stronger than the desktop I had 20 years back in my college. The software which we write today is more efficient than the software we wrote in my college days. But yet the CPU demand has a 100x in last 20 years. Businesses like Amazon, Azure, and GCP have come in have made 100 billion dollar businesses in this area.

So, what's the secret sauce? What's happening? It's a very simple concept. There's a flywheel effect coming in here. And what's a flywheel effect?

That when the cost of experimentation goes down meaning the cost of trying out an idea goes down people try out more ideas. When people try out more ideas, more successful businesses are built. When more successful businesses are built, they need more compute. That's what exactly happened in the case of CPUs. And that's exactly what we're going through in the GPU market also.

We're just in the very early phase of this hockey stick growth. Mhm. Now, I started with the statement that the centralized cloud are tracking and the future of AI will not be centralized. Why do I say that? Let me give you a practical example.

Let's say you're a builder. And you're a blessed builder, not like a majority. So, you have the money. You have the vision. You have the team.

And you want to move fast. Let's go. But very soon the reality will set in. When you will To for these GPUs you need you have no KYC on these platforms. You have to complete long processes.

You have to wait for weeks, if not months, to get these GPUs, even though you have money. And you're not just competing with the market forces, you're also competing against the deep-pocketed enterprises like OpenAI, which are always given the priority for the hardware which you need. Meaning what? Meaning that the top-tier hardware which you asked for, when your turn will come, you'll not get that. So, basically, you're competing in the most technological area with the deep-pocketed enterprises with the lower-grade hardware.

You're basically being set up for a failure right from day one. What about the money? Oh, you'll pay. You'll pay a huge amount of dollars to get those GPUs. And if you'll ask why, the reasoning which you learn here is that these hyperscalers of the world, to give you the GPUs, they'll have to buy billion-dollar worth of inventory.

They have to create these big buildings which they call data centers, and they'll spend million dollars there. They have to hire hundreds of people. They have to pay for rent, cooling, compliance, and so on and so forth. And for all this upfront cost, who will pay for it? The what?

Yes, you guessed it right. When you're paying for this compute, you're not just paying for compute, you're paying for this upfront cost which is baked into the business model of these centralized players. So, in short, what's happening? You are basically paying a huge amount of money. You're not getting the hardware you want.

You're not getting it timely, and in our world of innovation, time lost is business lost. Opportunity lost money lost. For lose-lose. And that's exactly why I'm saying that the centralized players of the last decade are not built for the future builders like you. Don't get me wrong.

These centralized players have played their part. But now it's time to break free. After if you still don't trust it, just talk to people around you, the fellow builders. These centralized players give some compute credits to people. The success rate of the crypto builders in these computer credits is abysmally low.

And even when they get it, it's far lower than the competitive web two players which are in the same market. There's a straight away alienation they have done to the crypto industry and that's what centralized space can do and we're living through it. So, we see what the growth of AI is, we see what is important which is compute, we see how much money we can make and we see what are the problems in the centralized space. So, what's the solution? The solution is very simple.

We don't need these trillion dollar data centers. The only thing we need is a deep end. And to be honest, it's not a very noble problem. This is a problem which is existing in a lot of other businesses also and people have already solved it. But in this industry, the money is so huge that there are players who are not allowing it to happen.

Let me give you a practical example because I have worked in such a company for seven eight years at leadership level. So, let's say you come to Token Dubai. You want to book a hotel, right? There are popular chains which you are low of. So, you go to Marriott, you try to book a room there.

Token is happening, rooms are totally booked. You go to Hilton, rooms are totally booked. Now you go to Kempinski. You've already been said no twice. In your mind, you think the inventory is gone and they charge you twice the price and you're ready to pay for it because you think the inventory is not existing and you can pay whatever you want, right?

To get the inventory at that time not to lose the business. But in the reality, there are a lot of hotels in Dubai which give same or better experience to you. And that's what was happening 10 years back. And what happened? Players like Airbnb, booking.

com came in, created this platform where all the supply of hotels got aggregated and given to the customers. So, suppliers and hotels got profited and users also get better prices. And that's why you will see practically on ground that the prices of these chains haven't changed in last 10 years. Right? There's And the same thing exists in the GPU business also.

The 90% of the world computing power other than these other than these centralized player is still unused. There are data centers and miners who have GPUs but nobody has reached out out to them. There are software companies who have bought in the GPUs. They've created the model, but nobody is going back to them to rent those GPUs. And what are we doing in Deeper Network?

We're aggregating all this inventory like these Airbnbs and Bookings have done and giving it back to the users. So, simple idea has already been succeeded in the other players and other markets. We're doing the same, right? So, what do you get as a part of it? Let's take another example what benefit as a builder you get practically, right?

So, you can understand. Let's say you're doing a KYC and you're trying to click a good image of your passport. Right? You're on the passport. You're about to click an image.

Now, the model have to tell you whether the image is blurry or clear. So, the software clicks the image, goes to a server from Dubai to US, comes back and says your image is good. Take it. But the amount of time it takes to go there and come back is 1 or 2 seconds. By the time your camera is not even on the passport, it's somewhere outside.

The model worked, but the location of your GPU was so important that it's not useful at all. Like it's unusable. Meaning the location of the GPUs also very important. And that's where the Deepen also help. So if you'll go and research in this AWS, the biggest player in the market, they only have their GPUs in 37 places in the world.

At present right now, in our platform, we have GPUs in more than 55 countries and ready to be added in other 150 countries if you need. That's the kind of scale we have. And when you would have been hearing these terms of edge computing and why it's important, it's exactly because of this example which I gave you. AWS just can't compete because each time they want to spin up a new data center, they have to hire a building, the team and so on and so forth. It's not scalable.

But for us in Deepen, it's just few more boxes and add inventory and good good to go. So the scale is there. What else? No waiting for the long processes. You want the GPUs?

In 2 minutes you have it. Cost. You pay 70% lesser cost. 70% lesser. What about security and compliance?

All our GPUs on our platform for the enterprises is SOC 2 compliant. So what does it mean? For a builder. You're basically today as a builder spending 50% of AI funding on this compute. Let me give you an example what it relates to, right?

Let's say you're again a blessed builder. You've been given $20 million funding. It's a big number, right? But the $10 million out of that, you're just spending in compute. Now imagine out of that 10 million spending, you save $7 million because the compute was 70% cost cheaper.

You're saving $7 million. Just imagine what all you can do in the $7 million as a builder. You You basically can hire more engineers, more data scientists. You can do more experiments, meaning try out more ideas. You can have a longer runway as a company.

And this is exactly what a builder of today and tomorrow wants to have. And that's why I'm saying that the last decade centralized players were not built for the future we are going in. And this is exactly why you need a deep end. Now, in this crypto world, normally people talk about decentralization. In io.

net, we basically build it. We live it. And when we talk about traction in io, we don't talk about retweets and all these things, right? We actually talk about in our KPIs real adoption, real number, real users. And we built our company in an old-fashioned way in this industry, meaning we were not a devnet company for 2 years and then launching a product.

We are less than 1 and 1/2 year old project. In the first 6 months, we built the product. And like a good startup, we gave it to the customers. We got a good feedback. We got a bad feedback.

The bad feedback, we took it on the chin, improved the product. And on the good feedback, we doubled down. And we are very proud to tell you the numbers we have achieved in just the last 6 months of adoption. We have given away 12 plus million computers. We have more than 27,000 clusters.

Our cluster stability is 95% plus. We have 8 and 1/2 thousand users. We have more than 300k enterprise GP GPU and CPU live today on the platform which you can use. And these upcoming features, there will be some builders out here. For them it will look like boring.

But honestly, this is what we're excited about. Why? Because just in past one month, we have three web two customers coming to us under the adoption coming to us and adopting the platform we have. And two of these are web two projects which have more than 100 million users. So it's a real adoption.

And these are the kind of players who are asking for these features which exist in the centralized players so that they can bring in more million dollars of businesses to IO. And this is what we're excited about. We want to build these features which hypothetically will not bring businesses, but we bring business right from day one. And that's what we're excited about. I'll give you one more example which we recently tried and we were awesome surprised.

We were blessed as a project. We got a good traction. We were the first day Binance launch project. We did well. Right?

But we realized that anytime you want to launch a project, you should launch a stable project. So we started doing my Amazon philosophy which I learned in past. Meaning we build a product, we use it in house for 5-6 months, and then we launch it to the people so that it's more stable. And recently we launched our influencing layer where we deployed all the popular models of the world like deep seek, llama, and so on and so forth. And gave it back to the users.

We didn't even do too much marketing around it, right? Because we wanted to learn from users and then make it better. And within 7 weeks of not doing too much marketing, these are the numbers we actually achieved. We have more than 500,000 prompts. We have more than 4,000 devs working on it.

And as a builder there's nothing more like feeling of proud when you actually see real adoption. Yesterday literally two people from Solana super team came and telling us that they built on IO intelligence. Last to last week when I went to India, there were three developers who came to our event and told they have built projects on top of IO intelligence. This is what the real adoption looks like. And that's what we're doing in the crypto industry.

Now, again, everybody likes to look love numbers and especially about revenue numbers. Because that's what something which a lot of the crypto companies are not used to. We are proud to tell you that in this industry, we have just not achieved the real adoption of web two people coming to web three, but we have made money also. We have made more than 11 million dollars total network earning in the last five months itself since the time we started focusing on the revenue. Because the first one year was just building the product.

And in terms of the ARR revenue, we're already 35 40 million over. The adoption is real. Now, a lot of the projects talk about what they will build in five x five years, six years, or what the future is. And we believe there is no future without an architecture. And we've already built it.

All these things are already there. I'll not bore you because there's less time left, so I'll just skip the slide. But all this architecture is already built. People are using it. The only thing I'll motivate you to do is please after this talk go to ionet and use it.

You'll yourself see what we are. Real adoption. We have players like Preia which got acquired by Canva very recently, one of the biggest players in the market. We have Wandera AI which is a voice generation AI in the web two side who's using us. We have Leonardo, an image generative company, they are using us.

And we have our internal crypto players also we work with. So this adoption is actually real. We have more than 20 enterprise customers already with us. Big sugar. Paid customers.

Now, this plan, which we are showing out here, is something which we've already built in last 4 months. There projects who will show these features as their year-long road map, and they make billion-dollar businesses just on the feature which we have developed in just the last 4 months. What is that? In-fencing projects. You will see a lot of projects here calling them as in-fencing project and getting uh long valuation.

We just built it within 3 months. The platform is there. It's already having users. We built it core staking. We are one of the very rare, if not the only project which has a rag.

What does it mean? Your LLM models which you create, it basically trains on a huge amount of data. And if you have a business which is having a data being generated every day, your LLM cannot train on it fast enough. Let's imagine if you're having data being produced just now. You can't have your LLM like learning on that in next 1 hour.

What do you use? They use a particular concept called a rag. We built it in-house. All these projects are not white-label projects. These are projects built in-house with adoption.

This is our road map for 2025. So, I'll make it a bit brief cuz time is less. But logically, you can imagine there two streams in IO. One stream is compute GPU and creating products to consume these GPUs in easy way. Meaning, boring stuff like Kubernetes, VMs, and all these products which make it easy to consume GPUs.

And we internally a company consider this consider this as a solved problem. The tech is there. Consumers are using it. Money is coming in just from word of mouth. Business is increasing.

This is a solved problem. Now, we are focusing on the other cloud side of thing, which is creating intelligence and AI layer above it. Like IO intelligence. We are creating model deployment layer. We're creating your BYOD, your intelligence next level layer, which is anytime a builder wants to come to our platform, they can create a model, deploy a model, scale a model.

And very soon we'll create a economy around it where if a web 2 player wants to come to web 3, we will give them some compute for free for particular certain time, and then after that they'll start getting charged. Unlike in the web 2 industry, where the builders have to pay right from time T0. Just imagine when this economy will be there where builders will be getting some resources for free, how many web 2 people will start adopting web 3? Three. The future is tremendous, to be honest.

Now, we're a tech company, but in actuality, our real tech is our people. Um I've seen in the crypto industry a lot of times there are people who can build tech, can build platforms, but they don't know how to build businesses. I've seen teams who basically have good builders, people knowing good crypto, but again, they can't build platforms. We have one of the very rare projects which are the great mix of builders who know tokenomics, crypto, who knows how to build platform because we have people from Google, Amazon, VMware where they've done this thing in past. And we have people who have done this thing of creating businesses in Uber and Binance in past also.

We have a right mix, and that's why we've succeeded in past, and that's why we'll succeed in future as well. Our mission is simple. We make it easy for anyone to access the tools to develop AI through reliable, affordable, and performing platform. Guys, there's one thing which I want to tell you. When Satoshi was building Bitcoin, it wasn't about money.

It was about the concept, and the concept was that the power which resides in the hands of few, it can be given back to the masses. But today, we're going through the same problem, and today it's about AI. I mean, what? Innovation should never ask for a permission. And decentralization makes sure it never has to.

If we don't own our intelligence today, we will not own our future tomorrow. Please, decentralize intelligence and own your future. Thank you.