Aethir provides decentralized cloud infrastructure for AI

0:01I'm Ashen Addison from the cryptocoin show and today on blockchain interviews we have Mark Ryden co-founder at Aether Mark welcome to the show and thank you for taking the time H thanks for having me yeah great to be here Ashton yeah I know uh aether's been uh crazy busy with uh the launch of of the platform and the token and uh it's a pretty hot area um

0:26decentralized Cloud uh graphic processing Ai and gaming which I'd love to dive into um I I would love to start off our conversation for those who aren't familiar just a high level on what is everything that Aether is working on and then we can dive into all the latest details yeah for sure uh we're an infrastructure project so sometimes it's better for us to start with the problem

0:48we're solving give people a bit of context and then I can jump into explaining exactly what we do so taking kind of a problem first approach I think a lot of people in your audience have probably heard the phrase compute scarcity or GPU scarcity right uh people like Elon Musk have come out and said things like gpus are harder to buy than drugs right you've got people like Mark

1:12Zuckerberg and meta purchasing you know just under 11 billion dollars worth of uh nvidia's pop GPU chip for for their uh kind of llama 3 AI Ambitions right it's get increasingly more difficult for startups and small to mediumsized Enterprise Rises to access this uh critical uh GPU resource right critical for for AI companies critical for gaming companies

1:40at Aether we kind of studied this problem quite deeply right uh and and and we' broken it down really into four distinct pieces the first is obviously by pressure right by pressure is driving uh scarcity in the market to the biggest tech companies in the world open AI Google Microsoft Amazon anthropic they all essentially have an insatiable appetite for gpus right they know that

2:08for them to win the AI race they need to purchase more gpus essentially more gpus means training AI quicker and they can't train FAS right so you have this sustained ongoing buy pressure that makes it really difficult to get access to these resources second we have uh the inability of of Nvidia to produce chips fast enough the market right they can only produce a certain number of chips

2:33per year they're doing their best and you know they're they are the biggest company in the world or they swap places with Microsoft every couple of days they doing a dance um but they can only produce so many chips and as I mentioned uh it's not yet enough to kind of meet Global demand thirdly we have what we call kind of increasing llm complexity right so llm stands for large language

2:58model right it is the name name for uh the technical term for chat key3 or chat key4 or llama 3 these are all versions of large language models right and what we're seeing is that later generation models are exponentially more expensive to train and when I say expensive I mean from a compute perspective from a data perspective and then also from a cost

3:27perspective right so what what's happening is as these companies are trying to make their models more intelligent right so that they become you know more effective tools for us to use they're finding that it's also costing them exponentially more dpus than it did just a year before right so we're seeing this exponential ramp up of compute but we're not seeing a similar

3:50ramp up in Supply or a similar ramp up in the performance of chips so we're getting this like growing gap between what's required of models and and the tip is coming online and then finally and this is where we get into I guess really The Sweet Spot for ather finally we also have the fact that these resources are really poorly managed right as a as a resource that's out

4:14there they're incredibly inefficiently managed and and we did a lot of research on this and we found out that when you take a look at Enterprise GPU resources sitting in data centers in the US for example they're sitting at anywhere from 15 to 18% utilization rate right so they're underutilized often nearly 80% of the time right so we thought to ourselves if we could create this

4:43infrastructure layer right that would allow us to onboard compute from a variety of different owners and then surface it really really simply to the market we could connect a new very meaningful amount of Supply to This Global Global compute Global demand compute demand ecosystem uh so that's effectively what we built right we built this Enterprise focused GPU compute

5:11marketplace where data centers tech companies Enterprises um you know even crypto mining organizations that have a large amount of latent gpus they can connect their defuse Cloud Network surface the surface them within our infrastructure earn rewards for doing work within the ecosystem uh and provide their supply uh to to This Global compute demand um

5:37market and obviously that demand Market can then access compute within our infrastructure and um and uh and access it in a way that's cheaper and and and also more available than than current big solution no it's a great overview Mark I appreciate that explanation and it definitely seems like there's a disequilibrium in the Demand versus Supply uh that needs to be solved um

6:02with the amount of underutilization from all these Enterprises and it's just not clicking properly I'm curious on you know I know Bitcoin Miners and and crypto miners have a lot of GPU power as well but what about individual people or that want to contribute to a decentralized network um maybe they don't have as much significant GPU power as others but is is anybody able to

6:27contribute to GPU power and then earn something through decentralized Market yeah that's a that's a great question um and it it's it leads to to kind of a big differentiator that Aether has in the market now when you look at deepin um the decentralized physical infrastructure network uh which is kind of the broader classification for for projects that build infrastructure

6:55that's contributed by others right so Bitcoin is effectively a deepin people contributing their compute to uh the deepin algorithm helium people buying these boxes contributing it to the helium Network all a deepin uh within the deepend sector there's compute deepend right and this is where uh ather lives uh it's where we have multi multiple different compute providers

7:18contributing their compute to a broader Network objective now when you look at how these deepend projects have generally been built right what the traditional model is uh is effectively taking your token and incentivizing uh consumers to contribute their dpu right contribute their gaming PC GPU contribute their idle you know MacBook GPU when they're when they're

7:46not using it because there a huge amount of like latent compute sitting at that sitting out there in people's homes um and traditionally compute deepins have focused on this form of latent compute right because you can take your and you can pretty easily incentivize users to join your network right because that compute that GPU it's essentially a stunk cost to them and anything that

8:11they can uh can kind of do to earn um from that compute is is beneficial to them the challenge here however right is even though you can build a network of gpus really really quickly right and even though you can build a really strong community because you have a bunch of people that are very interested in your token doing well because they're providing the resources they're mining

8:37that Network it's a really difficult if not impossible um compute Network to sell to Enterprise customers right because let's say you take a bunch of gaming PCs you know a th 5,000 gaming PC and you take that to an Enterprise customer the first thing they're going to do is ask you well Ashton how can you guarantee to me that uh those gpus aren't just going to

9:06be switched off at night right how do you know that the game is not going to you know just switch his PC off when he goes to sleep or how can you guarantee me that the bandwidth of that user's home is not just going to be throttled because they have a zoom full or because uh they're watching netlix right you really struggle to meet the service uh and quality and performance and uptime

9:31requirement of Enterprise compute buyers when you're building a a a compute Network on the back of consumer equipment right so I guess your your your question was can consumers contribute to this type of network yes they can traditionally can they contribute to aether's network no they can't not today and the reason is because ather has built a 100%

10:00Enterprise Focus Enterprise grade deepin right and and that's a massive differentiator for us right we have 42 43,000 gpus within our Network 100% of them are Enterprise grades sitting inside a data center or connected to Enterprise grade Network infrastructure we have a global compute Network that's perfectly suited to meet the requirements of the biggest

10:28Enterprise customers in the world right and that's something that no other deepin can say um it's put us in a situation where we can on board the biggest AI companies in the world we have the larest uh llm training contract putting on the back of the dpin right this is uh fed ml or or as they rebranded recently to to tenta Opera they're training their

10:52own open- source model on the back of ather h100 uh we have the biggest telecommunication company in the world out on our infrastructure and you couldn't do any of that if you had a network built on the back of consumer gpus uh so to be honest it was a harder thing to do right I just spent the first half of this telling you how difficult it was to these types of gpus and it is but if you

11:20can aggregate them right if you can create a meaningful amount of supply and convince and show to your customers that that thisly is available at the highest level of quality and performance uh you have a you have a massive Advantage because obviously the unit economics of compute within our Network are very very different to a network where you have to

11:43purchase every GPU yourself you have to own those servers yourself right so we can we can out compete compute traditional compute providers uh in in almost every way uh because we're we're built in this in this decentralized way now that being said and sorry for rambling but that being said uh do we intend to uh open the door for Consumer compute in the future

12:09princi all yes right uh will it be within the next maybe six months we'll see um if we do open that door it it has to be for Meaningful product that we can sell right ather is very much about real business that's why we have you know probably the strongest revenues uh in definitely in deepen if not in the broader sort of crypto Community um so we're very very focused on making sure

12:37that the supply and the compu within the network has utility and that we're able to sell it and we actually have customers for it um so we do intend to bring consumer computer into our Network in the future uh what form that'll be I can't tell you today however I can just and I know we'll touch on this in a second but I can mention that we specifically did a node sale right

13:03because the specifications for running a node were such that anyone with a computer anyone with a consumer grade device could run a checker node and and that Checker node is is functioning doing work within our ecosystem so for us we were like do we downgrade the the requirements of our Enterprise compute so that we can let retail in or do we give them this give them access to to a

13:33different role within our EOS system a very very important role right the Checker and and let them in there and we chose to keep the standards for our GP compute extremely high but open the door for the Checker role and decentralize that um to to kind of onboard consumers into our Network it's a lot of very great points Mark and I agree with you that you know

13:58this is business and you want to go for the highest quality and make sure that you have 99.99% uptime and uh it's great to have a decentralized network uh where anybody can contribute but who knows you know what old graphics card Jimmy is using as a hobby trying to connect and and if you're trying to train the next chat GPT 5 on that it might just not cut it so

14:21um% I see I see that vision and uh it's it's nice to see that that's possible in the future but also the Checker node because I did see uh on the site there there's a node sale with with with dpin um but that's how people can contribute and not just through uh their their graphics card but they can be a part of the network so that's really good to know um now at the beginning you spoke

14:43about the huge gap between you know there's a lot of demand and the supply and and how you're filling that what it's hard to quantify but are you guys making a significant uh Gap being filled there you know and how much more demand is needed with what you guys have um are you guys able to sort of Supply what everyone needs yeah that's a that that's Alo a

15:13great a great question um short answer short answer no right now we're not we're not making a meaningful Dent right the the the the massive amount of compute that is needed right the gap between Global compute demand and Global compute Supply is just so immense um that really unless you're unless you're in Amazon right unless you're a a Google you're not you're not

15:40making a de right um I mean it's crazy to think about the scale of what's being built in this space to service the compute sector so for example um I you definitely try and find the the exact quote because I'm going to butcher it but Microsoft is deploying at the moment or they're undertaking the

16:11deployment of the biggest infrastructure project in the history of uh Modern Men right they're deploying more Capital to standup compute data centers than has ever been deployed on any infrastructure project ever right it is massive like where we're sitting at the point now where the world is looking down the barrel of so much compute demand that they're estimating that the world can't

16:41produce enough power right we literally won't have enough power supply at the planet right with current infrastructure to satisfy the compute demand just five years from now right so is ather making a meaningful Dent there not yet right but uh I can guarantee you that nothing scales quicker in a compute ecosystem than a decentralized compute Network

17:11right and as more and more compute is coming online right and as Aether refin the the onboarding mechanism as we get better and more visible to traditional compute providers do I think that we have the ability to make a very meaningful debt 100% right uh because you have to remember the companies that are buying compute are not just companies like Amazon and

17:38or or I'll say AWS in this context they're not just companies like AWS or Microsoft aour or or Google Cloud right those companies have the ability to buy and on sell right but the average tech company doesn't right they buy and it sits in their own data center or it sits in a data center they're renting space in and if they stuff up their forecasting if they over bu right or if

18:03they delay production um that expensive compute and this is the most expensive compute of all time right there have never been chipped this expensive right uh it's burning a massive hole in their pocket so you have this resource that's incredibly rare right that companies know they need to compete in this in this market uh and they also know that it's probably be more difficult for get

18:30to get this compute in six months right so if I have the opportunity to buy now i' better do it uh and then they have the flip side of that which is if they stuff up how much they need or if their timelines a little bit off it's so expensive to hold this compute idle and that's where Aether has this really really incredible opportunity um to to deris that compute infrastructure for

18:58all of the companies out there that don't necessarily have the ability or the BD teams or the infrastructure to onell their compute through the cloud that's a great point I'm looking forward to it and you know I've been following web3 Cloud a little bit over the years and it's growing but it seems like the majority of you know Cloud infrastructure you know I don't know the

19:23exact percentage but it seems like Amazon or AWS you know runs like 80% of the internet um um and it's hard for web 3 to make a dent you have to either compete on price or quality um what do you think about that transition of Hosting processing power uh everything data uh into web 3 do you think that web 3 will gain more market share um than the traditional cloud and traditional

19:52Cloud will go down or they'll both grow but you know web 3 will it be able to keep up as much yeah I think I think um anyone that's out there kind of thinking that that web 3 will will overtake kind of web two in in any sort of major sector it's probably a little bit uh delusional uh today right it's particularly when we're talking

20:24about the the core competency of of the biggest companies in the world right um so is is is web 3 going to to to overtake I would say probably not right but like I like I kind of mentioned in the previous question do I think that we can have a very meaningful position within that 100% right uh because if you think about if you think about the ecosystem right just comparing Aether to

20:54uh AWS for example gpus are the same right we both we're selling the same compute it's not like we have special web 3 gpus and they have web 2 GPU no same gpus right infrastructure right physical infrastructure same data center right so this this equipment is sitting in the same places that that Amazon's aws's equipment is sitting in right we're linked up to the web through the same

21:21subc cables through the same network infrastructure right it's all it's all the same right so what what's different well web 3 has kind of two really unique advantages right or two levers that they can pull that I think are very useful lever first of all we have the ability to onboard other people's computes right I think that is a a meaningful um scalability Advantage from

21:50a cost and unit economics perspective but also a speed of reaching Market perspective right and then we have the token right the token is just something that these companies don't have and the ability to take that token use it to uh incentivize certain behaviors within your network use it to to subsidize use it to transmit currency across borders into regions that you don't have a

22:19presence like the there are just great like there are just some really strong Network advantages that uh that a web fre provider in this space has that allow us to to to to compete in some specific areas in really really interesting ways now I think whenever you look at a web 3 company coming in and trying to compete with with web two generally generally we

22:45will try and compete on price right because we'll we'll take the token and we'll subsidize uh and we'll we'll have like a a lower cost product in some sense and if ather going to do that uh yes right I think we'd be silly not to take advantage of the token in ways that can affect the price that we're able to provide to to our customers but I think the the Trap that a lot of projects fall

23:11into within within this space with their token is using the token as just a as a as as effectively only a discount and also using the token to heavily subsidize the scalability of your network right so creating these incentive programs that allow you to onboard a huge amount of users really really quickly right but then put you in the situation where

23:38you've attracted a lot of people that have a short-term interest in your network right and then as a as a an Enterprise service provider right and you're going to these customers you're saying hey I've got 10,000 gpus they're perfect for you right and the compute provider says Okay I want a three-year contract can you guarantee that all of that going to be here in three years if

24:01you've created a network that's really kind of uh volatile because people need rewards to be there then you've done yourself a real disservice because actually the biggest thing you need to be optimizing for as a web three company is stability because that's where we get picked apart by every Web Two company out there right they rip us to shreds on on our credibility they rip us to shreds

24:25on on the state of our ecosystem as a whole right so I fundamentally believe that the most important thing you can do with your token is actually ensure that it's incentivizing the stability of your network right and if you could do that first and then also use your token to to create great price incentives right to create great Compu incentives like all

24:48of those are really important pieces and they give you a great Competitive Edge but unless you're solving for the stability P um you're not going to be around long enough for someone like Amazon to even really care about you yeah no that's a great point mark it's uh it traditional business is so much about you know how can we make it 10 years down the road so that you know

25:09this can be a company that is here for decades and becomes one of the biggest companies in the world uh and on the crypto side you're also sort of fighting against uh the in some sense Bitcoin as a gatekeeper and the market fluctuations of of crypto being such a small industry that um although you have a great traditional service with the web 3 twist

25:32you're you're sort of thrown into the the wild waters of Bitcoin going up and down and a lot of people are there thinking shorter term although we are starting to see a lot more institutional investors especially with the ETF coming in thinking long-term uh with 10year plus investment Horizons which I think is great as that trickles down into web 3 services and infrastructure that

25:56long-term uh investment Horizon is hopefully going to affect these kinds of companies more where where they're thinking long term I think it's a great point about stability and um because it's not going to be uh you know it it will scale very quickly with AI as you said in the next year or two the compute power will be be needed but you need to be thinking 5 10 years down the road

26:22otherwise you you're not going to make it because it's going to continue to scale exponentially faster each year 100 perc yeah totally agree and and infrastructure you know people in crypto love the term infrastructure right but uh infrastructure is something that takes time right infrastructure takes time and uh yeah we're we're definitely in this for the

26:49for the long run right that's that's why we chose to be Enterprise first even though we knew it would take us longer to get to Market uh it's why we have such a strong Focus on real business and onboarding real revenue and real customers into into our network uh because you know five years is going to be a blinket of an eye for us right we're going to be here for for decades

27:12and I think um we'll be a part of building some really significant pieces of the AI economy that we're we're all going to start to to really be uh beinging over the next kind of couple years love I'd love to see it and now I was just looking at you know the launch just happened in in the um in the recent weeks here and also the the at token um I saw the node sales going on is there

27:40any other way that people can contribute say they don't have an Enterprise Warehouse of graphics cards uh at in their backyard um how else can people sort of get involved with this if this is the kind of thing they're interested for sure uh well just to CL ify for for any of the the listeners the node sale itself is now closed right so the node sale closed uh that was a a

28:09fantastic um campaign uh we we ended up selling just just over 72,000 Checker nodes uh raising around 42,000 heat right so depending on uh East price anywhere from 125 to 150 mil right biggest biggest node sale in in web fre history it was it was totally insane and really to be honest um just

28:39really a such a strong indicator of the power of a community right and the conviction that that Community has for what we're building and I think uh having the node sale and giving people the ability to come in and run those nodes and do truly meaningful work within our system effectively the checkers are the police force of the network so it was this really clear

29:06separation of police and state right we were effectively handing over the keys of the police force to our Network through this decentralized Checker ecosystem and by running those Checker nodes by conducting those Checker tasks within a network they're keeping the network safe they're keeping the network um uh free from Bad access and and that's a really powerful role and and it deserves

29:34a lot of reward as well so there's a significant amount of Acres token Supply that's distributed every year to all of those uh those 72,000 node operators so that was super cool unfortunately uh we uh for for those that missed out we closed that uh that sale a couple of weeks ago um there is a really cool staking product that's available so if you're really interested in committing

29:59and and and and earning rewards but you don't necessarily have any physical Hardware to contribute you're more than welcome to go to uh state. a.com and contribute your a to the staing pools there uh and then you'll also earn reward as a result of of that contribution I think for those people out there that are really interested in uh the hardware piece and and being a

30:25part of the physical Network infrastructure we have a really cool product called ather Edge right so you can check out that it's my edge. right and it is a physical mining Box about this big it looks like a a short fat PlayStation 5 um it's super cool it's a very very premium um device right you buy it it

30:55it's boxed incredibly well right it it just feel awesome anyway you buy one of those you plug it in to the power in your home it connects to your network and it will passively operate as a net as an Enterprise grade network node within A's ecosystem we actually partnered with Qualcomm right so it's their latest AI chip in that device uh and it functions as an AI capable node

31:27within a Enterprise node Network right so we will actually be pushing work out to the ather Ed ecosystem it will be doing that work and and receiving you know really really um you know wellth thought out rewards uh for for those that are that are uh that the owners of the device so super cool you just you purchase it you plug it in you forget about it right it's passive but it takes

31:55advantage of the fact that we're putting compute closer to user because big AI companies big tech companies they care about what we call the edge right that is compute that can happen closer and closer to users because the closer you can do some tasks the quicker they can then distribute those tasks to the user who's requested it right and it's traditionally being a very very

32:23difficult um part of the network infrastructure puzzle right if you try and solve it with consumer compute like you mentioned you don't know whether Jimmy has a different PC from someone else and and you've got so much heterogenity and variety on the edge that it's really difficult to to like unify and sell it right but what we've done with the Aether Edge is homogenize

32:49that layer right because if we sell this box the box is the same everywhere and then we've got this unified Edge Network that we can start delivering really cool Enterprise workloads to um and uh and and people can earn rewards from from owning it wow very cool I'm going to check that out for sure uh I can leave a link in the show notes to the to the

33:14edge and also to the staking platform for those who are interested um it it sounds like you guys are on the cusp of something huge because I know the processing power is going to be ridiculous in what is needed uh in Ai and I'm I don't think I I can even fathom how good it's about to get um even just the incremental versions of uh the llm models over one year um is

33:39crazy to see the expansion and I think the next few years are going to be mind-blowing so uh I'm I'm wishing all the best for you and your team on delivering the processing power to the right areas to make all that happen yeah it's not going to be easy but this is what we're gr at right um so we're very excited uh we've got a really strong Community we uh we have fantastic

34:05infrastructure already out there we're in you know 25 different countries and and that number goes up you know almost every couple of days so we're extremely well positioned and and uh I'm very very optimistic amazing thank you so much Mark I would love to have you guys back on in the near future um and I'm wishing you all the best until then thanks very much Ash and thanks for

34:29having me it's been a pleasure