AI SUPERCLOUD with Greg Osuri of Akash Network

Transcript

we're live my guest today is Greg osuri CEO of overclock and founder of the Akash Network today we're doing a deep dive into Akash how it works and why it's an important part of the future of cloud computing we're also going to talk about honestly what I think is the most exciting thing happening on Akash which is that they will soon offer gpus and that will enable all sorts of applications in machine learning and large language models which of course are all the hype right now we'll also spend some time looking at the GPU market and unpacking what's going on there AI plus blockchains all the buzzwords today um we'll also be talking about his views on the broader Cosmos ecosystem I'm also

dying to find out why he thinks decentralized experiences are better than centralized ones so if you enjoy this content please consider sticking with us we're validating on fmos Quicksilver osmosis and Juno just look for interop in the actorset and if you're in Paris this summer I'd love it if you come to nebula Summit it's The Interchange Builders conference that's happening on July 24th and 25th it'll be two days of technical talks about Cosmos IBC and the inner chain no panels no suits just Tech talks the second batch of early bird tickets just went up you can get those at nebular.builders my guest Greg wessiri is coming up next right here on the interop [Music]

hey Greg thanks for joining me today hey Seb great to be here thank you so much for having me yeah it's been a while since you're trying to plan this I'm glad it's finally happening and it feels like uh it was good to wait because like there's a lot going on uh in Akash and um yeah I you know we'd love to kind of deep deep dive into the technical underpinnings of Akash but also all this work you guys are doing in the GPU space so yeah for people who are not familiar with Akash can you give us a high level overview of the vision and what does the product look like today absolutely so um I know it's I think Akash has always something going on but

this is definitely the most important moment for Akash uh so we had a perfect time for those of you that don't know what Akash is essentially what we call a super cloud a super cloud is a um a a cloud uh some could say it's evolution of how we look at the cloud essentially uh the idea behind it is there's so much heterogeneous compute uh right all over the world um and if there's a way to Federate this compute and give a homogeneous experience to uh to a developer a builder we can use these compute resources in a more efficient way so

Akash is the first supercloud and only decentralized super cloud we um the research for Akash has been you know happening for quite a lot really began with uh with Cornell writing the paper on supercal cloud and and uh in the overclock Labs team uh taking the paper extending that and adding the the marketplace component to it and making it permissionless and decentralized right so uh right now Akash uh is a functioning platform and uh you know several Cosmos projects including osmosis Mars and quite a lot of Juno stargaze and all these incredible projects you know are deployed on Akash in some form or the other

and uh we are right now uh you know we've progressed quite a lot and Akash also happens to be the first IBC chain right if you remember uh the first IBC transaction happened between Cosmos and uh and and Akash network uh and uh we're also one of the first few projects to adopt tender mint uh in 2018. so we've been you know in the cosmos ecosystem for a very very long time um also osmosis pool number three and four are Akash right so if you remember those days of three thousand percent APR so we've had quite a lot of uh journey journey through device wow the girl days uh you know we've we've been long uh and

now we are on a pinnacle of the most important uh important upgrade uh for Akash Network introducing something called the gpus we're pretty sure we'll get into that very soon uh and that's uh that's International yeah very cool um yeah I'd love to talk a little bit more later about how uh teams in the cosmos ecosystem are using Akash like how they're leveraging it I I so I use like Akash a little bit like um I guess last year I I was hosting a bunch of websites on there like my personal website and the interop ventures website and and I really enjoyed it I mean it it it felt really good to be able to deploy a website to like a decentralized

compute cloud and um just point your you know your cloudflare to to that link and it just it just worked the issue that I was kept running into is that I would go to my website once in a while and it would just be gone and I didn't know why or for how long it had been gone and I think probably after looking into it now those resources that might have been reclaimed by the uh by the people providing the infrastructure you know that that seems like a ux problem that um you know kind of sucks you know if you have like resources up there and they're and they're they're falling out uh off the network

um what what kinds of what's the biggest challenges that you guys have when uh you know in terms of managing all these resources like how difficult is that uh it's a very good point sakash is in such early stages right it is purely non-custodial one of the most important aspects of our network is uh the non-custodial nature as well as the open source nature of it right uh the challenge with non-custodial interfaces uh being a Pioneer is we don't have engagement infrastructure for the user that you're used to in web 2 like notifications or Auto renewals of things of that nature right Auto

deployments and things of that nature even though the platform works as as advertised the ux has been quite caught up um and the challenge is really doing it in a non-custorial way and um you know some platforms take the shortcuts they actually offer custodial solutions to offset some of the non-custodial capabilities rakash has been very very uh uh strong in holding to the values of non-custodial media and that's reason why you know Europe website probably went down our biggest challenge right now is Char right so we get adoption but we also have quite a lot of people leaving and there are two primary reasons for that biggest reason I would say 90 of the

time is because the uh deployment runs out of funds and the deployer the tenant has no idea that the they need to renew and there's no email addresses to to email them or notifications so I mean perhaps you're you ran out of funds in your in your escrow account and then you don't you don't even realize that because you're probably aren't monitoring that right yeah second Big Challenge is recognition right so providers can reclaim resources in a scenario like a website which doesn't have a state it should automatically deploy to a new uh new uh provider if you set it up that way but most people don't most people just deploy dark container uh and forget about it so yes there are a lot of these like small

small nuances and majority of them sort of like cause is shown right even the platform works as advertised uh and those issues are getting solved very very quickly and they're getting solved in a in an open and distributed manner that means a team that created a cash network is not the only one solving these problems but there are 10 other teams we Fleek we experim with cloud cloud Moss predator and a lot of these applications that are addressing these uh capabilities of pitfalls in their own way some non-custorial some custodian that's the beauty of an open club and um and another uh from a churn standpoint uh you know the you know if

you control the chair and we will see net Positive Growth uh but also from a uh from a um on ramping or like uh you know using the platform or converting the platform converting a non-user to user the Big Challenge there happens to be you know happens to be getting access to tokens right and this increasingly regulatory uh realm which is extremely hostile uh it is impossible for has been so far it'd be possible for us to provide credit card based payments because it's a very regulated uh space and so far the experience using critical on credit card systems has been subpar

but now you have uh companies like stripe which we all use quite a lot in companies in to some degree care of money I've seen a demo it looks very very promising uh are offering credit card on-ramps and Akash soon will enable um uh you know stable settlements right that's also been a challenge especially for long-running workloads because you know you price the workload in ekt indicates volatile and so on a in a good enough period you may actually lose money or gain money depending on on on how you're looking at it but subpart experience right and I think that has to be fixed so these two yeah and you get

you guys are not the only ones sort of dealing with that problem I think like it is an experience problem that needs to get solved across you know the entire crypto ecosystem is how do you get access to to crypto and there's always this onboarding problem um you know what a question I wanted to ask you like sort of as we maybe start diving in here is like you know public clouds are great like at iws and you know digitalocean and all these things are amazing um why do we need Akash like what is really the fundamental uh thing that you guys are trying to do and what does that look like in a more long-term you know like five ten year vision so three main reasons a lot of reasons

why uh in decentralized systems uh you know I'll read but in terms of uh Cloud right the first big challenge is as you scale out on a public Cloud you end up uh uh paying quite a lot more than what you should pay so your expenses are outrageous so if you look at any public filings from companies that were built on a cloud like Netflix's or dropboxes even Asana as you notice 50 over 50 of the margins actually go to a cloud provider and that's not sustainable in most cases and there are indicate there's quite a lot of uh work done by companies like 37 signals campaigns that actually came back from the cloud to our data center

saving quite a lot of money but again in a data center the challenge is you know uh under utilizations because of our Provisions right because it takes quite a long uh you know quite a lot of time to put a server on your data center you tend to over provision and then you just sit on under utilize resources right so uh that's one problem and second problem is uh the stifling of Open Source software we all know what happens when companies like Amazon see a successful open source software and they tend to offer the take away the core business model for these open source companies which tends to be you know hosting right so we see quite a lot of dampening of revenues for this open source companies

data bricks included and quite a lot of fightings and and and you see the quality reports as to they have interest how how do you deal with the Amazon thread right and the third main um the third reason is anti-market forces right so Amazon currently Amazon web services is currently under investigation by U.S government for antitrust uh the monopolistic behavior and this is a natural sort of like cause of hyper concentration of power corruption always comes with power even though the company may not have intentions to operate this way it just nature and organic to have this concentration lead corruption right so and the solution really is to uh claim

sovereignty how do you get back the power that we gave in you know through this anti-market forces also you see censorship right so you see censorship both on political biases also you see censorship on like business biases or business profits companies today uh that are doing machine learning cannot get these gpus the high-end gpus we call a100s and h100s on the cloud unless you have a special deal with someone higher up in one of its Cloud providers right so they selectively choose which business and which uh you know company gets the resources thereby they selectively choose which company should be successful openai got a big leg up because of this reason and that's not fair that's not fair to to a lot of the

innovators which is a foundation for our our economy and our society to get priced out out of these resources that's why we need a if you consider Cloud to be the fabric that's all that holds a society we need this fabric to be open and we need this fabric to be governed in the public and resources allocated in the public uh rather than a single Corporation doing it in a completely opaque manner right yeah yeah this idea of super cloud I think you know which we'll talk a little bit a little bit in the later on I think kind of touches to that and you know how do we best govern this this infrastructure um you know one thing that I've one thing that I've sort of noticed over the

years it covering crypto and being in the ecosystem is like there's been lots of projects that I've attempted to do decentralized compute um and and then to some extent you know projects have tried to do decentralized storage you know those I think you know there are some signs of success there but on the decentralized compute side like we have projects like very early projects like Gollum and and eye exec you know they're based here in France and I I haven't followed up with those projects so much but it feels like you know they're not massive successes or like adoption is slow and you know they're they're not like you're making headlines every day

um why do you think that is or is it just like a timing thing or are we at a point now where it makes more sense or was there something kind of fundamentally wrong with some of those earlier iterations of decentralized compute right there are lots of things that went wrong with the early release projects and the main reason I would say is uh over focus on technology than the market right so a lot of these compute providers be it column column is a prime example the over focused on verifiability uh of compute general purpose try to do a general purpose platform but and then ended up building something that's not usable for us to launch in ethereum we all know how that goes right uh you know high gas fees to deploy a Blog is not okay in general but

you can't even deploy a Blog it's it's basically a machine learning work with us that's what they focus on but a challenge with Gollum was you have to rewrite your entire application to suit their framework which is not the best framework so they had to while fighting the non-custodial element it's in itself is extremely hard but then trying to change uh the source code base to suit a decentralized network for a value that's not very obvious right is very very hard so um so that caused failure I think the The Columns usage is like seven dollars a week they have no new users at all it's just there it's just a zombie

project right uh I exec on the other hand went through a few iterations but they began in pretty similar mode as a column but I think they're you know transitioned into more of a darker based they're doing some work I you know I haven't really seen any latest uh successes but I exit is very batch optimized not necessarily latency sensors that means you cannot deploy a WordPress website with a i sequel backend on it yeah you can do batch workers right and all the workloads are priced in public it's a different model altogether it's not a what you would consider your typical you know your application running on Amazon Deploy on two okay exactly right and Akash was the

first general purpose platform in the sense you can take a container that's literally one running on gcp or or Amazon most likely if you're using something like a Docker compose you can take the docker compose file with some modifications deployed on your cache right and that was a key Innovation Akash like introduced and uh and execution happens off chain we're not the first ones to do that but you know we're the ones to sort of like really Embrace that option execution and the um chain becomes a coordination and a settlement mechanism at that point right so the chain becomes very lightweight component and that goes away where you interact you so the look and feel really

feels like a super cloud right that's really the the idea so you get a very homogeneous very familiar interface the docker compostar interface where you have the resources where you have projects and uh and and Akash also has this powerful um um configuration language called sdl which is like an orchestration language that lets you be extremely complex or extremely simple it's very very flexible language and um and uh and you know with that we were you know hyper focused on open source and Community First approach uh so with that we were able to have a good ecosystem of projects building on top of a cache that helps users adopt Akash and that's that list is growing as well so a

lot of things went wrong with the early set of cloud computing projects that were very ambitious and the difference prakash was we were very Market focused we were very uh focused on what the user needs and what the market is asking for it was very very clear was not a verifiable general purpose compute verifiable compute was for the domestic compute which makes sense uh verifiable compute is where is impossible for general purpose computer right so and and we really took that and we introduced verification based system uh and um and a lot of other projects came after Akash that took their model and you know had success as well cool yeah so earlier we're talking a

little a little bit about so you know the user experience challenges that that you guys are facing and I mentioned at the top of the show that you know you thought that decentralized experience was better than centralized experiences uh why do you think that is that you know I think some people would agree with you others would disagree uh yeah yeah so the non-custodial nature of the system makes for extreme composability right and uh and also from a safety standpoint and and just the experience of logging and authenticating uh in a non-custodial manner where you you know your information is not shared and you

know exactly what kind of information is there right right now we use a browser Plugin or browser which is extremely subpar but imagine if that device in itself can give you the authentication mechanism right whatever device is using and even extend that so uh and and the non-custodial nature also calls for uh that and along with interoperability calls for much better composable applications right so I as a developer do not need permission from someone else to use a protocol but rather build a system that is hyper integrated hyper like converge right uh with with not having to worry as a due

to how as to how I should secure a user's credentials or user uh you know private information gives me a lot more power to create incredible applications so the on-ramp is is a challenge engagement is a challenge and these two problems will get solved and they are getting solved very quickly and uh once this is solved uh the user experience will be using a decentral system is significantly better than using anything centralized right um yeah because uh you know um you know for passwords you need a password manager right and that's it that's a big attack Vector with this you choose where you want and how you want

to secure ID and that gives for a much broader set of options in terms of uh in terms of uh you know control right so I can choose my device as my secure mechanism that I can use a ledger or I can use I have you know infinite possibilities and as we see more more Builders using these possibilities and and innovating on them we're going to see significantly better experience I mean you know if you remove the on-ramp part using uniswap versus using a kraken right like just the experience part is incredible using uniform it's like a couple of clicks I'm done except for the metallask right like the more all that stuff yeah yeah

um cool let's dive into some of the technical Classics here because like I I said I use the cost you know use the interface or at least the application you guys had uh I'm sure it's been updated since then I should probably check it out again but um I was always curious like how it works under the hood and you know who are the network participants and specifically like the the I'm curious like who's running this infrastructure you know is it is it like mostly at home users or or is it people who are deploying Akash on uh on data center infrastructure is there a range of users and what are some of the challenges that come into you know running a network of Computer Resources with different uptime and latency and

connections fees and how do you manage all that great so to understand akash you really understand the actors different players right you know it starts with the tenant which is a user and then you have a provider that you know provides compute then you have an auditor which essentially audits providers and then you have better data switch in a secure the blockchain right uh we all know what validators do they secure manager blockchain the governor blockchain and delegators will get tokens to them and the uh providers provide compute but how do you know who's running right your question who's exactly Providence compute there is a decentralized uh uh auditing mechanism where Auditors actually do off-chain uh verification and post those proofs on

chain for someone like you to go and actually verify so this is a section in your sdl file called audited attributes so you can choose an auditor that you want to trust and get a web of trust mechanism through that Association and it's very very common to see the actual provider uh that's that's deploying you'll see equine access of the role Cherry servers you have a ton of these like tier I would say like tier five tier four tier three data centers and then you have just companies with excess capacity right so we're seeing quite a lot of trend now with 37 signals for example which create a base camp and you

know Ruby and rails and whatnot uh they themselves are bringing data centers in-house and with excess capacity for them is beneficial to deploy and to provide the computer nakash right so companies are becoming their own cloud providers as a technology gets commoditized in terms of containerization and realizing that you know you know you can actually offset some of the costs by putting the computer in a car so solve trying to be Sovereign at the same time scaling is really what we offer for providers right um in the sense that hey you can you know actually use your compute offer that or when you want more you can actually get back get to the network so um with now with gpus we see a little different type of user base uh a big set

of users or providers are are former eat miners that we're doing proof of work mining a lot of institutional miners um that are you know hot days of the world of the clouds I think they're offering on our on our on our on the network they are you know just offloading like high-end gpus that they have and then we're going to see professional miners that were not institutional but have you know smaller color capability or even home capability with not a latency sensitive uh not capacity that's data center great but good for batch jobs right like I could

use so you're saying like you have Cloud providers like like Equinox and and some smaller Cloud providers that are deploying Akash on machines that they're that they're not utilizing is that right like they're leveraging Akash um to essentially offer up their excess capacity correct access capacity is what we're going for and soon you'll be uh we're also working with some of the hyperscalers to enable unused capacity for their pre-committed users so a lot of times when you are a company if you want to get you know access to some of the high-end resources you have a pre-commit a certain amount and more often than not you're not using this

pre-comed uh you know so it's beneficial for these customers to put their computer in Akashi and Cloud providers like that because it's helping their customers and helping themselves their pre-commitments quicker and um and especially in gpus because you know companies like Lambda Lambda Labs which is uh quite popular in GPU cloud their users are buying up their compute just to have access to the compute and they're not using it so you know yeah that hurts everybody right so we're helping those uh essentially creating a secondary market right that's what Akash really is yes and use cloud compute which is so basically you'll get the same exact thing you get on Google or Amazon for much cheaper right so using that okay so I didn't I didn't realize

that this was actually I saw you know I I I knew that there were some higher some sort of actual Cloud providers and data centers deploying a cost I thought I thought these were you know people that were deploying Akash on like buying up some some service space and deploying I didn't realize the data centers themselves these companies themselves were actually doing this so it's the customers of the companies right uh the customers right okay we may or may not have something in the works but really what we have what we are making is making it easy for these people to deploy onto our cash or provide to Akash right yeah because um you know

we will be making some really good announcements very soon you'll notice the work we've been doing in this in this sphere of like enabling you know pre-existing customers on the cloud to deploy Akash but uh but yeah uh uh that's it that's a huge problem I mean it's a huge like pre-commitment problem right that even Banks uh that approached us to a point where they could Finance some of these companies to get pre-commitments and in order to service their debt put that in a car so a lot of creative sort of like ways people are looking at deploying Akash and um and soon I think a real like real vision is machine learning companies should own their own gpus because that's ultimately where the sovereignty stands

you know when they're not using it okay interesting um so so Akash then acts as a market for compute where actors can come in and out of the market and because they're offering resources that they otherwise would I mean it creates more fungibility in the compute Market you as a whole where there uh were you know you don't have this sort of functionability between one provider and another provider say between like Amazon and like Equinox or some other tier like ovh or whatever um and and by leveraging a cost you you sort of create this Market of compute

resources that can come in and out and it's just a lot more fluid and fungible uh right yeah it it yeah I I feel like I've got a whole news review on Akash now after after explaining that um yeah it's funny because like a lot of the cloud providers we actually talk to they acknowledge the problem with the cloud they acknowledge Classics pretty extremely high technology that it's you know this is the market forces are not very uh I mean the company may or may not act in the best of just the market all of them actually acknowledge and they're very produce inflammation surprisingly a lot of employees that work at these large companies uh after talking to them uh we realize even

though our paper written five years ago talks about the industry as a problem and and points out but in order for Akash to be successful we need participation from everybody Akash is a market he's a permissionless decent class Market uh and the mission for Akash is to ensure the tenant the user gets the best deal right no matter what cloud provider you are and we have nothing against like Cloud right and Cloud I think has done a phenomenal you know phenomenal work in enabling giving access to companies that couldn't get access to a data center so cloud

played an incredible role and now we need to think about evolving the club to a super cloud right and in order for the Supercar to be successful you need participation from the big big guys or else you just end up being this Fringe you know dark Network right so I think in a year or two uh you know a time frame you're going to see quite a lot of links participation things are just getting easier and easier and easier in terms of how a provider can come online and organic and use and as that progresses it's not hard to see how Akash becomes a second market for for cloud-grade compute yeah so how does it um like how how do you we talked about this problem earlier where resources can

get reclaimed now of course that's an issue if you're you know running a website or something that's publicly facing I mean any any kind of compute right if you stop that compute while it's happening you know that might that might affect the user of that compute resource quite a bit um you know if we look at say Cloud store or decentralized storage there's often this replica this replication aspect to it so that the resources are spread across different nodes so for example in our weave the resources are spread across the network and so this is replication and that ensures that the resource stays online you know for a long period of time without any downtime for compute that's different you you have like a VM right that sort of has a run time and and it is running some

computation if that resource goes on offline it can't right at least I don't think so it can't just sort of pass that on to another machine without having provisioned it and everything so how do you deal with that is there like a is there plans to create more robust sort of you know mesh of or you know of compute that would um that would abstract away some of these uh some of these lower layers yes a good question so if you think about it any resource in the cloud unit non-cloud there's always a Time the provider reclaims a resource it could be a year it could be two years could be three years doesn't really matter there's always a time a provider will say they're going to shut down

their instance they have to retire it right so there's always a Reclamation time for a resource on a cloud it's just that it's not obvious how long the Reclamation time is it's not very clear I think first thing we need to solve in Akash is yeah but it's also seamless for the customer like a customer doesn't see that resource coming offline because typically Cloud providers will just move that resource through another like actual machine not really no no no I noticed thank you yeah okay give you like six months a year give you a good long reclammation time and they they can't just move data over right because that could cause laws a lot of damage because the

redundancy and replication is very workload specific every workload has different ways of how they do reputation you know be really not get too technical but really comes down to consistency of the comments of the rights right media consistency versus you know events versus uh you know eventual consistency determines how you replicate that so you can't just like take a data blob and put it while it's under processing address this different story right that's why storage networks are easier redundant so the point is there's always a Reclamation time only question is how long do you get so Akash that Reclamation time is not very obvious so the first thing we're going to solve it

is uh providers advertising the Reclamation time and sticking to the Reclamation time so you could get a notice saying that hey your resources will be reclaimed within a week a week or a day or an hour right so and then you can choose a provider based on Reclamation time so something if you're deploying a database and you want to be like I want six months minimum information because we're taking six months to move these database over to a new server right or if you're running something like a machine learning I would be like hey give me one hour I just need to stop this job reclaim to another job right when you know recommation you can actually automate that Reclamation in a in a much easier way uh each uh each transfer each transfers workload

specifically and second thing is if you're running Mission critical you always have to be redundant odd numbers the way I I run my workloads is I have on a card you can specify different providers in the sdl file you can have three providers and running it usually three different regions and have a load balance in front of it that'll do the do the routing and with the health check on the on the container and it's a little complicated but it's typically how you run a production rate infrastructure and from from a state standpoint what I do if I have a postcard server for example uh depending on uh the the storage I mean postgres I

would do a a master sorry new words we would do a leader follower uh style uh style replication model with the hot failover for the follower to to elect as a leader uh using the Quorum in case of a uh in case of a you know a fault so what Akash does is it is extremely strict in enforcing these practices and you have those practices be centralized cloud decentralized cloud doesn't really matter Akash is very very strict if you don't do them you will fail right and that's what happened to you and happen to a lot of people and that causes churn but the point is it's

extremely strict in forcing these practices uh Cloud providers are much more lenient right in enforcing but I believe in order for you to run a redundant system you have to plan redundantly and Akash is is unforgiving in that in the form like think of Akash as this system for Builders that know exactly what they're doing right yeah it is very very powerful but it demands quite a lot of work from you and that is a user experience problem or whatever you want to call it right and that that's getting solved very quickly um and uh we introduced something called a console if you recently console or Acacia Network and in that you'll notice topologies uh as part of the deployment

packages so you can actually deploy a multi-leader A multi-follow or sorry a leader follower topology uh you know that's application specific obviously you could do like you know workloads and DMZ you can do all kinds of distribution and essentially packages the best practices as to how to deploy an Akash and and package us then in a packageable in a Deployable manner using a beautiful longer shorter interface right and so uh so it's Reclamation you know is not as foreign as you think it's just you know it's just not obvious on different uh there you go you can actually really you click on a cache on top deploy now

uh and you'll see like an RPC now validated or Sentry node right you'll start seeing like other topologies and you can actually uh do you know yeah you'll go ahead and do it live demo will be fun take Essentials this is a missing certificate connect your wallet yeah yeah yeah yeah okay no this is cool I'll have to check this out um yeah I mean I've heard a lot of people talk about uh running validators on Akash and I yeah so from from this year it looks like it is possible to do that but one of the one of the issues I think was that you know the key is you're you're validator to keys are effectively on someone else's

machine which you know is the case for for any you know for any validator that you're running in the cloud but here there's you know perhaps like another intermediary in the middle that is not the cloud provider um how does that work like what's the best practice for running a validator on cautious is it a good idea and how do you manage the keys right so you use autogen attributes and some providers like equinix that directly provide some providers like Cherry servers that that people love uh their direct providers their direct providers that can be verified using this attribute so if you're concerned about someone being a middleman you can use a drug provider and second you shouldn't really put keys in any provider doesn't matter if I you

should always have a mechanism to uh you know to offset right so I would recommend using something like a remote Center hot Crux or some of that you know where you actually sign the keys in a completely different system shuts not the system that's you know uh that's processing your transactions right sure but I think I like a lot of validators just have like their signing key on the server and they're not using horcrux like a lot of validators I think do that right right so you know best practices you know remote signers uh you know if you're concerned about middleman use a direct provider uh equinoxes and Charities and we have quite a lot of direct providers um and uh

uh yeah so you know but I would choose the more signers as a best practice for anything as a matter of fact or any any secure setup yeah this is cool I I didn't know that like you had this this uh this tool that just allows you to you know deploy an RPC I mean this is pretty useful I think like for the cosmos ecosystem as a whole to have this yeah especially like bottom providers that are not profitable as in these market conditions it's very very hard for a lot of these folks to pay the bills and I think we should start considering something like attached uh Cherry service is a great provider you know they're the drug deal providing compute for Akash and that I think run quite a lot of uh you know power notes

and Cherry highly recommend them if you're thinking about deploying uh we're also working with a lot of other direct providers to come online uh yeah eight uh work minor and we have you know website Cloud which is really really prominent in um in in in ethereum ecosystems a lot of these providers want to get to get into Cosmos and actually like looking at akashes away because another way to think of Akash is it makes a a average cloud provider uh more professional because they have the tools now that is you know and the ecosystem along with it right so you have fleas you know all

these amazing tools that are integrated into Akash meaning these providers now now have all these tools for free that they don't have to you know develop right so a lot of these providers are looking to plug into a casual Akash gets more you know more more mainstream and more I guess at the world as awareness grows we're going to see this increase in and also Akash is pre-incentives so we have no incentives for providers and that's going to change very soon and we've been extremely careful on how we're looking at incentive design we've looked at every possible incentive design and and we have to be uh you know uh cautious in how we incentivize for virus because so far it's been

not optimal in the whole crypto space starting off with file coin to heliums of the world and you look at a lot of these incentive models that are not really didn't deliver to the expectation yeah so I mean listen if you can help me get delegations I'm happy to set up a to figure this out and set up a validator uh and a cost validator on a cash um and do it yeah although I it's interesting I think Akash doesn't really um it's one challenge there running a cash balance itself even though the validator um the provider does not depend on the

blockchain to execute they may terminate your leads if they don't get paid so depending on when the provider is claiming their payment if the blockchain is is down that could cause a non-payment or issue depending on how to provide yourself I think some of them are like okay if you're not able to talk to a blockchain don't do anything it'll work well just keep running so yeah but that could be could be an issue especially during upgrades Akash didn't have any unplanned outages so far we've been very very lucky and again I don't want to I mean it's been live for over two years it's been great uh every outage we had was plan outage during during mainnet

upgrades and usually did Cosmos right um yeah you all know how that goes with validators the dance who do every every minute of grade um but so yeah I think you can run on a cash validator nakash itself but but but you could run a you could run any other validator on gosh I guess like the issue is like yeah yeah okay cool absolutely right with like with like you know a big challenge again is like playing with akd with with yeah usdc payments that's coming very soon I mean we're very very obvious like hey costs like 10 bucks a month or like very very clear concept yeah yeah so so why did you so why does it cost you stalker

containers and are there other VMS that is it possible to deploy with other VMS like I I've heard from a team that you're familiar with uh that like wasm VMS are very useful because they allow you to uh effectively know the the compute resource required before running and they're also more efficient and yeah are there are there any other are there any plans to to leverage wasm you know either either as like an alternative or a replacement to Docker or yeah what are your thoughts there yeah I thought we looked into it very seriously uh my background I you know my I've been in early contributor to the

cloud infrastructure container native uh world I mean all the way from 2013 you know I was uh 2013 was heavily uh hypervisor world right so you had VMS as the primary mechanisms to ship software and plus software and they were mutable so that was not great especially with downtime right so I was hyper focused on solving the parity problem uh you know essentially when you deploy something how do you ensure the the target machine has all the dependencies and containers happen to provide the best abstraction right so I was working on continuous pre-docker using lxc and Dr Cameron made container super simple and Docker didn't have fault tolerances so I was working on fault

tolerance using a research paper published at Google called Omega so I was actually working on uh right building this optimistic scheduler for Docker right or like containers uh then kubernetes came the scheduler was not as great as a optimistic scheduler but it worked and it gave orchestration and fault tolerance and and and you know I've done some early contributions even my code even I still currently used by kubernetes and and all this amazing software that came about if you look at GitHub you have an idea of my contributions um and um so containers happen to have they're not like the Silver Bullet uh you know

one as one would call them but they're having incredible incredible ecosystem that solves for a lot of problems right and Akash is not built to be container only you can plug in any you know any container sort of like uh you know not Docker only you could put any container runtime you can attach the container and that really depends on the provider and that's also another cool thing about a clutch is you get this optionality we'll get into that later but but it's it's pluggable right and the white paper doesn't white paper actually proposes for a other mechanisms as well as how you package and how you can deploy yes wasn't could be a package but we looked

at wasm um you know some friends of mine create form companies to take was on mainstream and a lot of them didn't really succeed because even though wasm is great on on on on on on on on paper is extremely opiniated in terms of uh how one should write an application and that's that opinion leads to a lot of uh a lot of uh friction for adoption right so I think the ecosystem hasn't gotten to a point where you can actually write full-fledged high performance applications in wasm and because a lot of times he's like well if you're right if you think of a wasn't mode you're really thinking about serverless right or infrastructure and server listening

is heavily distributed and at certain scale serverless gets very very complex and extremely extremely expensive right um so there is now a trend of reverse serverless which is like model applications like Amazon Prime video who actually wrote a paper like last week I think or week four talking about how they save 90 on their Cloud bill this is Amazon Prime video talking about savings by switching from microservice architecture to a monolithic architecture which is anti-wasm right anti-all these uh patterns that we it's very anti-pattern and now people are questioning like is is microservice the way to go so I mean it really comes down to it

really really comes down to what stage and what scale your operation your workload is at right like early days scale when when you're starting off something it's very advantageous to deserve us because you don't want the complexity of managing infrastructure and the cost of the complexity in terms of time exceed the cost of your you're you're going to market the product right so optimize for the least amount of work you need to do in order to get to Market so you know serverless becomes an amazing option but at scale when your problems had shifted from time to ship to actually save costs on infrastructure

like you're doing video streaming you have to go monolithic and you have to go vertically integrated there's really no other way because and the reason why Apple products work really well is they're completely tightly integrated right so uh in in case like Amazon Prime where they could save 90 because you you tend to save quite a lot on serialization overhead uh when it comes to distributed systems because the moment you communicate from one system another there's always serialization deserialization that happens over over a line right and that cost surprisingly is extremely high and you add verifiability and and and resource like estimation costs on top of it that comes with awesome so every time you're like

running you're actually estimating the cost and you're optimizing accordingly so your your overhead for compute is extensive that car that results in cost so so yes so it really depends on like what state right if you don't have scale that overhead is is manageable but at scale it or had become extreme extremely expensive there is no one architecture that's why it's very important to be general purpose it's very important to be unopeniated and Akash even though it's painful at times when you see automatic replication systems that'll all just move your workload wherever but it's like you're so opiniated in replication that you're going to limit your workloads on type what kind of replication now I don't

want to throw fired or things about another project but I have to mention this project called flux right I mean again they're doing the thing they have their opinions they have their approaches and let Market be ultimate sort of like this decider as to who wins but the challenge that I uh you know flux they really wanted a good user experience so when you have a workload on flux actually you know replicates across different servers and switches over automatically now the question is like who decides what is the event for failover like you know like application health checks have to do the failover so you tend to have uh you know failover when your database goes down

you you create a failure event right so in order for failover to happen you need to have an event bus you need to have a custom health check mechanism you need to have how the replication how the database is switched over right what kind of you know for leader election happens what kind of you know uh how how are you streaming the the data right what kind of atomic capabilities two databases are going to incorporate what kind of consistency you're going to have a virtual or immediate right that determines your your so you need a lot of these as when you deploy distributed systems at scale and production you need to be able to do these determinations that's that that's specifically your

workload you can't just put something on the cloud and expect that to replicate nobody does yeah right so so like so that's why I'm not like a big fan of like opinionated models in general as a platform I think platform should be extremely unopiniated then if so you can have abstraction on top of it right like people can build stuff on top and it's it's just more uh yeah I mean I think this touches some of the earlier points that we mentioned as well which is you know how do you just have the the simplest sort of uh infrastructure platform that can allow people to build for these different use cases that build that would have built in this replication or build in you know different VM support and things like that right exactly Fleek is doing only

web workouts they're really good about workloads JavaScript tested yeah and they know exactly how to use akashu Skin JavaScript and they do the opinion they do the abstractions for you they're on similar right Akash AI in fact we're doing this machine learning uh verticalization that knows exactly how to use scalar machine learning workload and removes all the complexity of using a cache right so that is a higher level function in terms of these opinion opinionated uh approach or Workforce as I like to call them yeah so let's let's talk about the GPU stuff because I think that's what a lot of people uh are are hoping to hear about so you know we talked a couple days ago you know about like the state of the GPU market and I

think you know I feel like you're someone who has a pretty good grasp on um the state of the GPU market and you know how demand has been growing and you know is manufacturing keeping up with that demand so like I'm curious to maybe get your high level overview on what that market looks like right now because it seems to be like a lot going on there yeah it's always been a high demand if you remember Akash has been working on gpus for a very very long time if you remember super mini and all these things that we we were we were trying right um uh super menu was supposed to be a super computer for the home so you can have your own own compute so we've been you know exploring gpus for over four

years I would say right um and now we finally have the capability it's very very hard to do an open GPU Marketplace uh so if you look at the space there are essentially uh very few companies and media AMD maybe Intel that actually uh make gpus and gpus are extremely uh powerful additional chips on top of the CPUs that adds acceleration for certain workloads to be it crypto mining or be it machine learning or be it Graphics processing gpus tend to have these these uh accelerated uh you know uh chips to to power these workloads right

and when it comes to machine learning Nvidia happens to have the best gpus because their on-chip memory is is much higher than your typical AMD processors where you have you may have more powerful like clock speed but your on-chip memory is small and on chip memory in the sense the memory on the chip itself that separate from your actual RAM and that's very important to train neural networks because neural networks are essentially there's a lot digging there's large data sets and breaking them into this like this checkpoints by X by going deep and you need and the latency between the chip and the memory needs to be as minimal as possible so Nvidia happens to

make the the chips with more memory or the most memory the most advanced chip which have which we call h100 right that's a purpose-built high memory uh chip um uh which costs like 40 Grand or something like costs like forty thousand dollars and if you can get it that is it's it's like a Rolex or a Ferrari right the MSRP is never the now the price you end up paying for the chip you end up paying a lot more because the supply is extremely constrained uh the constrained by two main factors first is they cannot make these chips fast enough right um because uh you know uh

because the in the the supply chain for these chips really involves a lot more companies than just Nvidia can make them so Nvidia is the one that makes ships but they get the machines from this company called asml uh you know uh and uh uh sorry uh machines the Nvidia gets chips from tsmc or Taiwanese semiconductor corporation that actually makes puts the chips on the transistors on the ship and then you get the software to make uh you know and the Machine is made by ASMR right or something like this whole supply chain uh so a lot of very incredible choke points and with covert it it it you know

the general Chip Shot is because they couldn't make this chips fast enough and second the ex the demand from AI just exponentially grew right to a point where uh AI research has been there was a lot of hype at some point and then the hype died right and then there's a lot of research and innovation especially around this this mechanism called Transformers invented by Google really um and you know a few I think five years ago and from Transformers we started seeing a whole range of large language models or llms we call them and llms have amazing use cases we have

generative use cases your summarization use cases you have classification you have search you have a lot of amazing things you could do with llm and one of the use cases happened to be generative use case and so people could take take a general llm and put a pre-trained model and create something called a generative pre-trained uh Transformer or GPT right so GPT 3 came out in 2000 by open AI and a GPT and open AI you know it's been around for a few years and they built a user interface called chat GPT on top of GPT that they introduced uh last year that obviously you know blew up right uh yeah and they got 100 million users in

three months and the fastest ever growing so it was a UI problem not the technology technology has been there for a long time and we've been following very very closely uh and and that gpd3 needed quite a lot of chips uh which are a100s so for for the longest time opening AI the lower end chips someone said a100 is a high-end chips right but but lower than the h100s it's 400s are brand new they came out just like okay a few months ago yeah but right gp3 LGBT 4 was trained on a100 LGBT 3.5 at least um so and there were I believe I mean don't quote me honest I think they're

like 45 to 50 000 um here 100s as far as I know in the in the world uh Amazon and and Facebook uh has access to about 20 000 of these so we know exactly where these are and a lot of them are actually owned by the U.S government in National Laboratories you know or state governments right um so um and uh they're so so uh so high demand it's now impossible to get a100s on cloud I mean I think you can get a 140 gigabytes Which is less powerful than the 80 gigabytes by 80 gigabytes is just not available anywhere like even if you know the software you can't get them uh people will just buy them up right

h100s are even more hard because they're about six times faster than a100s for certain type of applications especially llm training applications so Elon Musk bought about 20 000 of these right like a 20 or 10 000 in any company that's doing anything serious with llms uh are buying up these chips so are people like selling futures for these like uh people like hoarding them or they are arbitrages there are some of this like speculation Market isolation going on yeah oh yeah absolutely there is there's a whole whole ecosystem whole supply chain in new players and supply chain Now with uh holding on the h100s

um like there's something called distributors in the middle so you have you know your chips and you have these server manufacturers like Dell super super micro and a whole lot of them they have a whole lot of supply and you only get h100s if you buy a server from these people so it's like this is a lot of a lot of stuff here if you can get access to age 100s okay get them right you can actually sell them in in secondary markets I don't know how much they're going on eBay but people are showing on eBay for like 40 Grand yeah like there's one on eBay yeah right I mean I mean it's very obvious right and and guess what when you have when you're sitting on these chips and you're not using them you have a place to go now to to to put them and

and when you have a you know when you have a buyer you take the chip and put it back right so you know sell it to the buyer um and uh so there's a lot of like this incredibly high demand and now with geopolitical scenario not looking so great for for uh globalization there is a reverse globalization there's a lot of protectionism right now happening um and uh and and the main reason why Chinese Chinese are prevent are not uh sorry there's export control since Advanced Micro processors now guidance and executive order about nine months ago preventing any advanced Microsoft a micro uh chip sales to China and Russia and in Iran obviously and any American

from working for these companies for microprocessors so overnight with a signature I think mining killed the entire Chinese GPU industry and um so there is Extreme export control on these chips it's very very hard to get them I think they're even harder than getting a gun yeah well right that's not a high priority to cross here but what's the what's the like biggest problem like so you guys are running a GPU test net right currently uh what's so different about providing gpus you know as opposed to providing just generalized you know CPU uh

infrastructure that necessitates running like a dedicated test net for this and how is that rollout gonna look like once it starts hitting mainnet yeah so it's a new type of resource on the cache Network and gpus are very device specific whereas CPUs don't really care much about what kind of CPU you have you may yeah or Intel but they all work pretty much you know the same right gpus are very different your application is very GPU specific some applications are written only to work on each 100s or a100s or like it's very GPU specific right so their anti-uh uh their opposite to the whole notion of a container where you can

build one and run in anywhere so unfortunately with gpus you have to build a container to run on that GPU so it's it's a very different way of deploying right it doesn't uh uh you know it doesn't give a homogenality you get with gpus and the gpus and kubernetes are extremely new Nvidia is the only company that has some degree of uh you know some degree of uh adoption and some degree of like good software but others don't really have the kind of penetration and if you were in another country and viewer thing you had to put Nvidia is a software company not a hardware company

that's my contrary intake in The nvidia's Innovation is really software the Cuda yeah SDK that they came out 10 years ago that's why they're really good with AI no one else yeah and video is going to Software um I mean I don't think that's such a contrarian view I mean I have like a pretty summary or like understanding of this Market but my understanding is that Nvidia really excels at the software and like they've really created a huge mode around that and sort of um yeah I mean most people think Nvidia is gpus right but yeah anyone can make gpus I think the software that they built yeah not anyone but anyone that's you know but they also have this really tight integration with um with with the chip right so coming

back to your analogy earlier about Apple and how that you know the software and the hardware need to work together I think they've also like done that really well and you know to the same extent that you might consider Apple also to be kind of a software company right because they're building really good software that serves a purpose and experience but building the hardware that is able to deliver that experience I think you know they the software is the interface with with the right person that's using it um correct yeah I mean any I mean they're much better phones than iPhone right like cameras a lot of Androids better than iPhone but the the cohesive vertical integration uh is done so well with Apple product and media products that's why they shine Nvidia is the apple of chips you know yeah so is it cost gonna be providing like raw access

to this GPU power or are you going to be providing access to the models uh yeah what's that going to look like for for users Akash network is going to be unopiniated not custodial always so it's going to be a raw access to be on it but overclock Labs is building Akash AI right which is an abstraction on top of a car says optimize to run models in a very simple way so we're taking the experience first approach use the first approach and one click deploy stable diffusion control net you know llama alpaca whatever model you want and we're going to be extremely expressive we're actually working with uh with the integrating hugging phase so any model on hugging phase you should be able to deploy Akash any more like chain these

are very popular in machine learning application uh you know packages or platforms you should be able to click deploy a cache within like a couple of clicks and pay using credit card now of course credit cards are not are custodial right so they're not not the studio uh that component overclocks the building to specifically focus on machine learning developers that are not crypto native right for them I tried onboard a few of them in like multiple hackathons and that was the worst thing ever to go there and go tell them to get go give your mother's password and track in and get your kyc done and wait for a couple days to get the stockings it's just so bad you know to hold onboarding and we talked to them about Kepler they're like well what is

it a browser bugging in the freak out because nobody wants to install brother plugins I mean you'll be surprised as to how anti browser plugins right so it's just uh the quite a lot of barriers for folks to see the value of Akash from a machine learning that's not crypto needle and overclock is solving that problem one old one is one of the companies solving but specifically Focus towards machine learning applications so you know I I I gotta I was talking to someone earlier about this interview uh and and uh it's like oh I wish I ask Greg and and um that I think this is kind of a good question to to wrap up on you know Akasha is is essentially like a

Marketplace for for distributed computing and you guys are uh offering now like different types of compute Resources with General CPUs and um and gpus but how are you thinking about Akash over the long term and um how do you how do you plan on transitioning to whatever the next big thing will be right the next big Computing um you know resource that will be in high demand if it's not gpus or or CPUs you know maybe it's like you know Quantum compute resources or some other type of of compute resource that might be like at a different abstraction level than what we're used to um yeah what's the long-term Vision here

very cool thank you for the question so the you know the first major milestone for Akash network is parody in the cloud so right now Akash is still from a capability standpoint it's not equal uh with with Amazon or Google right we need to get to parity that's going to be hyper focused uh on the roadmap post parity is how we evolve right right so post parody imagine having 10 000 providers in every city of the world right GPU CPU speed whatever resource imagine being able to provision a workload uh that follows your user no matter where they go so when you have a

future where you have computers essentially are mobile you can think about a robot or a self-prime vehicle or any of these devices the compute is at the edge and in order for age compute to function you need uh you know to solve the communication cost problem in the sense how long the package take to travel between the age device and in the cloud right uh the the device has motion and that the device is static right you always need this like communication and and at that point the only way to achieve that is the massive distribution of compute all across and with the incentive model for these providers to provide computer costs right uh at scale

um and and and and and and and the possibilities are Limitless I call it Innovation Beyond imagination because now you have a network that's extremely agile extremely high performance at the age and second thing is Akash is community uh managed Community built projects we have about 280 contributors out of them 17 folks are employed by overclock Labs if you look at some of our repos like Akash console Akash Network slash uh you know GitHub Cartoon Network console you look at the pull request nine out of 10 pull requests come from outside overclock Labs it is a community built and it's very critical that this can and we're adding about 15

developers building on a cache every month or something we're growing really fast ever since we came up with the announcement of like completely radically open source on our blog in earlier this year we had an amazing participation right so now everything about Akash is said in the public right at a product the roadmap the features the priorities all that is sent in the public all those meetings are published publicly posted transcribed summarize everything done right it's very it's really impressive by the way yeah very important for us to expand that because that is what is going to get us to move faster in the future yes you know you can go faster by yourself in the short term but you can go further when you're together with with the community and

that's why and that's why open source software at a certain scale moves so much faster we look at Linux you look at kubernetes you look at any of these these projects that hit a certain inflection point and started really getting into this community uh contribution that's where Akash is getting there right uh once they take off it becomes an Unstoppable portion becomes the standard now you don't look at any other container orchestration that kubernetes is ridiculous and everybody tried at one point to have their own content conversation right you don't look at any other uh for the most part operating system that's that's not Linux right and you don't go try to rebuild Linux and there are a lot of different operating systems before lands right so and you have this kind of inflection point from a community

contribution standpoint Once Upon achievement it becomes Unstoppable and becomes uh and it becomes the global sign and that's what Akash is going towards right um and at scale we will have massive distribution of compute uh with options that are impossible to get another Cloud today our cash is already offering certain you know certain capabilities where you have nbme storage or if you wanted uh you know a different runtime for your Docker you don't want Docker if you want uh you know they're all kinds of cool like experimentation when it comes to runtimes that are happening some providers want to provide that as an option to the user so you're already seeing this like Branch off uh between what you can get on the cloud to what

you can get on a couch with h100 we end up being the only cloud provider to have h100 so you're already starting to see the branch office no longer just a alternative Cloud that's cheap it's now that cloud that can get resources on that you can get on the on the traditional Cloud right and that's that's going to be the future right and um and it's very very important for us to ensure that there is a healthy Foundation uh that is building uh both for decentralization or announced Evolution the way I look at decentralization is yes you know every protocol when you look at from a technology standpoint uh is decentralized at a higher level but you start peeling layers right you repeal the layer you see governance and you see

centralization the governance and then you peel enough layers you see uh you know ultimately comes down to who's actually building the software that's the biggest attack right once that Builder is attacked the software stops evolving it's very very important for that uh that layer to be secure and that layer to be uh you know fault tolerant and that's why Akash is built by Community Builders overclock Labs disappears today Akash will still be working and we're working very hard to have this on chain you know uh fund uh to to you know sustain the development right we call it a public goods funding right so a lot of the stuff you're working on will come out more publicly very soon are geared towards ensuring that the sustainables event comes

evolving software so that is our biggest more I think Community is is is the key right yeah and that's something money can't buy yeah well before we wrap up here just a few more minutes I want to open it up for questions in the audience there's a couple of questions here and one that I had already also listed uh is you know how do we deal with you know material I mean here we're talking the the person's asking about censorship is it possible to host not not safe for work projects in Akash my question was more around the the topic of illegal content I know that like with um our weave or it's like governance mechanisms there to limit and take down um illegal contents things like csam and

stuff like that uh are are there mechanisms in place on Akash to deal with this sort of issue yes there is uh there is active conversations on content moderation happen on Akash right so kind of moderation is a very sensitive subject um the well you know we need free speech at the same time we should protect people right there's actual Bad actors abusing the system and we saw that happened with SIA and that led to the fall of Sierra right so it's very very important to have uh this mechanism and we're looking at certain ways we go to github.com network in community there's a group called uh Sig moderation especially interest group when you click on that you'll see all

the progress we've made um there is a notion of the ultimately it comes down to the providers right providers are liable right so if you're a provider hosting some something illegal you you are required by law to take it down and you need to have a the MCA mechanism to to protect these rights you have all kinds of like the legal sort of like Frameworks that are already established jurisdiction you're based on so we are going to make it easy for providers to comply and it's very important right and there is uh if we don't make it easy for comply for them to comply they get in trouble so it's very important for us to protect the providers so Akash as a protocol level will never consider is

not built that way you cannot censor but providers ultimately are The Gatekeepers and they have to um they have to make sure that you know they're able to serve you know and protect uh people that they are serving right so yeah uh so like generally what happens is um you know we have okay like all not all content is equal right so you have from a sensitivity you have something that have different political biases right you all saw what happened with with Amazon taken on parlor or things of that nature with Akash if you have something like that as long as it's a provider that still considers this not a not as bad as child

pornography and they want to serve that that's fine with them a culture will still function very well but if every provider it's bad they're going to you know not serve they choose they have complete Liberty uh us to choose who they want to serve right yeah I mean universally bad universally front Pawn they won't serve so as long as there's a single provider running on a cache the workload will get served and that provider is ultimately liable right so yeah you know and if if with authorities then deem that provider to be malicious and they you know there's an IP address they you know the provider can be easily attract so yeah you're not you end up in a scenario where you're going to have a lot of illegal content and providers not you know progress are going to lower the

content and they have to take them down um and uh that's how Network kind of like self sort of like regulates itself without having a police or a someone watching them from from the outside standpoint okay cool um yeah and then one more question that actually came from the telegram group uh was with regards to you know conversations that you know you guys have had with I think Google or Microsoft over the years and um you know do you do do they see you as competition and in your in general or what are large Cloud providers what do they think about Akash are they you know do they consider them as potentially like a path to

like a a a a a suitable path to offload some of their excess capacity or yeah well what's the state of you know your your workings with large Cloud providers I don't think I've ever said anything about Microsoft I don't know where my Microsoft came from but I can publicly say uh you know this is based on tweets from employees of these companies that I'm talking about I mean just if you go to my Twitter you'll see a lot of tweets from Folks at Amazon they had a blockchain Folks at Google you know uh the the web 3 divisions of Google they all are talking about they all think Akash is a critical piece of the future of the club and this is public

information by the way right of course I'm not going to go into details as to what exactly is functioning sure sure people can look at your Twitter yeah right but you know if you look at my Twitter I mean I I keep retweeting tweets from these folks talking about Akash and they're really excited about it uh and uh they all acknowledge I mean cloud has a massive problem when it comes to efficiency of users these are the resources right and they all acknowledge and these companies are realizing web 3 is a future they have established teams for web 3 uh to to make their companies more of everything and it's very very aligned when it comes to Akash right so a big uh a big challenge is like improving liquidity or efficiency with their

existing users when it comes to like prepaid customers I think they look at a cash as a way a solution for that problem and second gpus are very hard to come by and everybody wants to give gpus to the users it's just just can't right because there's no way to get them and all these GPS are just locked away everywhere so Akash is unlocking this like math so under utilization under utilize resource especially for high-end gpus h100s right so you in the future you you won't be surprised to see an Amazon user using a GPU on Google paying using their Amazon account and actually you know not leaving Amazon because today they you know they are if a user cannot get gpus on Amazon they leave Amazon and they go to someone else

so now Amazon has a way to retain the user and you know satisfy the need by satisfying the needs and actually work with a competitor and Akash is the bridge and that's exactly what we want a custom Market Focus where these companies are user photos and that's very very important right so yeah that makes it so much sense and uh it's it's I mean it's not you know it's going to be very very clear in a few years when all these Integrations come out and we are not in the business of announcements we are in the business Integrations we don't announce anything unless it's actually integrated right so and because of the transparent and open Nature all the conversations are public so if you actually participate in our working

groups we have meetings almost every week on like where the product is going and we are very very active if you're not able to there's also recordings we want if you have time and it's also summarized recordings right so so you can go through a lot of that stuff and get an ideas who we're working with but you know you're an open source open Community uh and uh we haven't had any conversation with Microsoft just to be clear uh yeah I don't uh yeah I I know they have good well three teams but just don't uh for some reason haven't you know interacted with many of them it's mostly the the you know number one and number two uh Google and Amazon okay well thanks for that and uh yeah thanks for sticking sticking around for

so long it's I've really enjoyed this conversation learned a lot with the cash and uh we had quite a few people on the live stream uh certainly this was probably amongst our you know top ranking episodes in terms of people that join the live stream so you bring a crowd sir thank you thanks it's awesome thanks again and hope to see you at nebulous Summit yes looking forward to it all right thanks Greg thank you