DePIN for AI: Building the Internet of GPUs - TOKEN2049 Dubai 2024
TOKEN2049·Mon, Oct 21, 2024, 12:00 AM
Transcript
hello everyone my name is Ahmed I'm the CEO and founder of iret today we are going to show you what we have been working on we think the world needs to see see this we've been working on this for a very long while um Ion at um working so hard into democratizing access to AI compute power and democratizing earning from AI compute power Services uh in a way it's a it's a network that have two sides those who provide the GPU compute power and those who use the GPU compute power um IET have started um almost a year ago after six years of research in Quant systems engineering and building low latency trading system something really requires a lot of compute power from there our story started we were using gpus from cloud providers that are extremely expensive and we had to we had to find a way to reduce our expenses in 2022 the market was really bad um so we were not breaking even from the earning from Market making and all the trading we are doing in the stock and us and a crypto Market liquidity was really bad so we I needed to reduce my capex and and I was like ethereum at that time was suffering the miners were like moving from proof of fork to proof of stake so there's a lot of GPU Compu power that is in the market or basically in facilities that people don't know what to do with it there's no mining anymore and if you mine any coin at that time in 2022 like you would make I don't know 60 Cent a day for the best card in the market like RTX 3090 at that time so I was like what if I could just you know call my friends who are mining crypto and they were like can I use your GPU for my AI power at Quant systems so I had a friend in Saudi Arabia who had a facility of around 300 GPU something really small and I called him I was like how much you earning he said almost 60 Cent a day without electricity cost and I'm like what if I give you $3 a day for this card will you give it to me of course so I took the cards from him the same card is sold in Cloud providers like genesis Cloud AWS and so on for $24 a day so I was paying hundreds of thousands of dollars on gpus and now I need to reduce that expense so I called this this facility was like I would pay $4 he was so happy because that's four times what he's more than four times what he earned from crypto so I started you know preing his facility for machine learning workload basically our Quant fund was Trading you know 1,000 stock and 150 cryptocurrency and for each one there's a model and we needed multiple gpus for each one of the pairs to trade fast and to like learn uh the machine learning has to learn from the tech data that's generated in the market so we need a lot of computer power now that's why we need the compu power so I started using facility in um uh Saudi Arabia then another friend in Kuwait who has also cheap electricity there and then I wanted to expand more like I it's so cheap computer power so I can do more with this so I called friends in Slovenia they have Mining facility there more than 1,000 GPU connected to it and then I had to figure out how to make this like gpus between multiple locations communicate with each other uh fast forward the year we found out that for like it it came like this moment is like wait a second did you just build a compute Network that potentially could be used by others and that's actually what happened so started working on the project 26th of February 2023 um repurposed the architecture for the Quant fund the the one that's using the gpus in other providers and started to give an example to like clients Uber openi instacart we pitched to like the best in the market and they're like when you have more than 200,000 gpus let's talk now we have many GPU but so we like from the day one when we built this systems that allow people to use this compute power it was already configured for like really high bar like low latency trading is extremely complex requires like really good internet the best connectivity speed the best security etc etc so it took us a year of work before we do I we start even IET as IET to do while we're doing the fund and configuring this architecture and building what we now call the internet of gpus so the concept of the internet of gpus is like imagine if every chip in the world gpus across you know hundreds of facilities what if all of them are like a one Hive like one mind what if what if I could have GPU just like I have electricity so and the question is like what if it's a really simple but yeah profound question what if accessing GPU compute power was as easy and as seamless as accessing the internet um right now I can challenge everyone in the audience I will give you like $500,000 if you can in one hour from this time rent 100 GPU anywhere you want any cloud provider any searge do whatever you want if in within one hour you can rent 100 gpus I'm going to give you $500,000 it's impossible you wouldn't be able so the GPU availability is really really critical problem that is facing large scale AI startups like if you want to rent two three four five gpus you will find it like rum pod vast AI you will find a lot of providers who give you this number but when you want to really cluster for like large scale or like really heavy workloads you wouldn't find so we ended up IO as building the compute layer that is crowdsourcing this uh GPU Compu power and allowing mining facilities that used to mine crypto or they have the the budget to to buy gpus to earn from the AI compute industry and also allow the lar scale AI startups to have access and availability to GPU Supply that is really critical for ai ai that have changed the world um it's reshaped Industries and that AI requires GPU so we see IO as the fuel the digital oil um and that's why I introduced the internet of gpus now as I said we want to democratize access to the AI we started by partnering up with you know fcoin Solana Aptos using multiple layers of Technologies so we can serve large scale AI startups at the latency we need uh you know proof of compute ult latency payment system um we needed chain to record all the Clusters all the payments for workers across the planet we have gpus in 140 or 139 countries how can we pay these gpus when we use their compute power using stripe or using other payment providers is really expensive so the best thing is pay them in usdc pay them using salana pay them using Aptos um that's how we started now what the thing is like 90% of the GPU Supply in the world is underutilized it's at home data centers that they don't have a good sales team so sell it some corporates who bought the gpus that they don't know what to do with it there's massive computer power out there but still if you you go to AWS Google Google Cloud Microsoft aure you wouldn't find this GP Supply you it will take you weeks to get a cluster of 100 GPU they will ask you a lot of questions why you need it what's your pitch send us your pitch deck your LinkedIn profiles like what is this like why it's not that easy and simple and the challenge is live I'm recording this do it and you have $500,000 if you manage to do it in the next one hour now so what we want is like want to create this economy that is powered by the currency of compute like we believe the real currency in the world is compute compute power is the digital oil I mean if you look at the GPU power in like if you look at China for example China have imported $233 billion of chips last year compared to $185 billion of crude oil so China's Imports of chips is more than Imports of oil it's really important that's why we believe like if this asset is somehow tokenized in a way where like we have this million gpus from across the globe high-end Enterprise level high security and then we have a coin in the middle that allows that facilitate payment system facilitate payout to people all across the globe we can actually build the currency of compute how much you have that currency how much compute power you have and based and the reason why we say how much compute power we have because we're going to live in a future looks like skying it in a way like you have a lot of AI agents across the globe that talk to each other and execute AI tasks all that will need compute power and each time you run an inference on any model that is deployed anywhere in the world you have to pay inference like inference fee for each time you inference a model you get a result out of a model when two agents AI talking to each other and they want to get a result from one model that makes image the owner of that model also should earn royalty fee what is that currency right that is the AI the io currency the iio is going to be launched next month hopefully we we planning to launch 28th of April um probably the lunch within first week of May now the supply shortage is the compute power demand is doubling every three and a half months so if we have a million GPU now after three and a half months we need another 2 million gpus that has been like this for the last 10 years so and right now the compute demand is two and a half times more than the capacity that's available in the market and the manufacturer like TMC in in in Taiwan is unable to manufacture the chips enough to match the demand now and all these new AI platforms like you know Chad gbt or or or or mid Journey or so on they all need this GPU Compu power and they're paying hundreds of thousands a month like uh they're paying around like for example openi is paying $700,000 a month just on gpus just on the gpt3 model and now you know they have other models that they are Sora and other models the amount of computer paying every day is like23 million a day on just the cost of GPU imagine if that money is flowing in into a cryptocurrency that is fueling that AI e that that sharing economy of people providing these gpus all across the globe forming what we call the internet of gpus now let me introduce you quickly to the team we are 100 people we are Engineers from all around the world um from or tree AWS Intel Ripple binance crypto.com our our vpf engine our CTO is VP xvp of engineering of binance he just quit and joined us um from firor from aptus digital ocean Yandex any scale Nvidia like name it neom we have really crazy amazing team that's working so hard we're now across 110 people since we made this slide we hired 10 people um we're growing really fast um and we have amazing Partners in the crypto industry that are building with us we have 560 Partners in the pipeline working with them providing them AI compute power and our investors we have Anatoli from Sana Mo Avery from Aptos yatsu from animoca Brand Sebastian from the sandbox we have amazing um VC we have raised around 4 $40 million uh reaching $1 billion valuation just on this round when we had only 80,000 gpus and only 20,000 user now we have 400,000 user uh more than 100,000 of them are Engineers uh we have uh 800 and actually today so it's a million GPU it's growing really fast um we got Engineers from sorry we got investors from animoca otoy U sandbox Aptos San Adventures of course okx Ventures Dy hack VC multi capital name it more than 40 investor 40 great VCS helping us and supporting us on building the right economic system building the right engineering the proper way to launch in the market engineering the proper to economics system that could handle massive like rewards uh H how can people who don't have GPU compute power how can they join this like how can how can you earn from some like someone who have $100 million who builds a facility in Taiwan or in Singapore or Malaysia whatsoever how can you also earn like him so we built a what we call Sharia compliant staking system in a way where basically each Data Center and facility they have to provide let's say I have have $100 million facility 10,000 gpus or 3,000 GPU 5,000 gpus for me to even be accepted to join the network I have to provide an uh uh some kind like a security Insurance like I have to put $50 million of usdc or IO coin sorry in lock I have to someone has to se for me $50 million to ensure the security that I me as a provider of data center I'm not going to manipulate the network I'm not going to be offline I will keep up my ab ability my my devices are going to always be online now why would I do that because every hour there's an earning reward that is minted to every provider of computing the network but this provider of computer network he cannot even join the network without putting the insurance now it's a mutual relationship so this who provides compute let's say hour one I have you know I'm earning $5,000 of ioin from emission reward now the Staker will earn will take part of that will take 10 20% of that now if you go offline the Staker is going to be slashed 3040 $50 he will lose from his money so now you as a ster without like he has as a provider of compute he has to maintain his availability and be um reliable he maintains relability score because if that score basically your device goes offline you will lose your Staker will lose money so and if your Staker decided to leave the staking you are entirely out from the network you're not allowed to join the network so as a provider of compute he's incentivized that he makes sure that his Supply is the best standards otherwise his Staker will lose money meaning that he will go out and find some other data center to stake to and you will lose your ticket to be part of the network now that system we've been testing we've been working on this with all our VCS we believe this is going to be something really revolutionary and a breakthrough we believe the icoin is going to be the the compute currency and I think that's all what I have for today thank you very much for listening and um yeah
Automatic transcript — names and jargon may be misspelled.