December 2, 2025 • 43 mins
Summary
In this Block by Block episode, Peter Abilla sits down with Kony, CEO and co-founder of Gaib, to unpack how GPU tokenization can create real yield at the intersection of AI, DeFi and zero-knowledge crypto. Kony explains how his mix of finance, AI engineering and running an actual cloud business led him to see a gap: AI cloud providers have huge demand for GPUs but limited access to private credit, while crypto investors are starved for sustainable, cash flow–backed yield.
Gaib steps into that gap by financing enterprise-grade GPU deployments, securing their cash flows, and wrapping both into tokenized products that anyone can buy on-chain. Kony walks through their two-stage due diligence, risk controls, and how they work with cloud partners, OEMs and third-party custodians to verify hardware and revenue. The conversation covers Gaib’s $2B+ pipeline, DeFi integrations like Pendle’s PT/YT structures, and the long-term plan to bridge Web2 capital markets and Web3, starting with GPUs and eventually expanding into other real-world, productive assets.
Takeaways
— Kony blends finance, AI engineering and cloud-operator experience, which shapes how Gaib designs GPU-backed yield products.
— Gaib’s core mission is to deliver real yield in crypto, funded by AI infrastructure revenues instead of token emissions.
— The team treats GPUs and their cash flows as a new on-chain asset class that can be structured, financed and traded.
— There is a big market gap: many AI cloud providers cannot tap traditional private credit, despite strong demand for their GPUs.
— Gaib pre-finances enterprise-grade GPUs, then gets repaid from the cloud provider’s customer revenues over time.
— A strict, two-stage due diligence process screens partners on company data, clients, revenues, hardware and security.
— Gaib leverages industry networks, OEM/ODM relationships and third-party custodians to verify that GPUs exist and stay where they should.
— Current integrations like Pendle let investors split principal and yield, and more DeFi money markets are planned.
— The GPU pipeline already exceeds 2 billion dollars in assets, still a tiny share of a rapidly growing AI compute market.
— Long term, Gaib wants to bridge TradFi and Web3 and extend the same model to other real-world, revenue-producing assets.
Keywords
Gaib,GPU tokenization,AI infrastructure,real yield,DeFi,Kony,Block by Block podcast,enterprise GPUs,AI compute,Web3,TradFi,private credit,tokenized real world assets,Pendle Finance,PT YT structure,AI cloud providers,structured finance,cash flow backed tokens,zero knowledge cryptography,AI DeFi
Chapters
00:00:00 Introduction to Gaib, Kony’s Background and the AI x DeFi Opportunity
00:05:56 Why Crypto Needs Real Yield and How GPU Tokenization Fills the Gap
00:12:02 Financing Models for GPU Acquisition and Gaib’s Credit Structure
00:17:52 Due Diligence, Risk Management and the Scale of AI GPU Demand
00:23:57 Gaib’s DeFi Integrations, TradFi Bridge and Expansion Beyond GPUs
Follow me @papiofficial on X for upcoming episodes and to get in touch with me.
Watch these interviews and subscribe on Youtube Block by Block Show.
See other Episodes Here. And thank you to all our crypto and blockchain guests.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:01):
Welcome back to Block by Block, the podcast where we go deep with a builder shaping thenext era of crypto, AI and everything in between.
Today we're joined by Koni, CEO and co-founder of Gaib, a project at the intersection ofdecentralized AI infrastructure and zero-knowledge cryptography.
Koni, welcome to the show.
(00:22):
Thanks, Peter.
Yeah.
Let's begin with your background.
um You are in a very interesting intersection of crypto and AI.
What makes your background a very good fit for this intersection that you're in?
Yeah, so as you mentionedWhat Gaib is doing is actually sitting at the intersection of AI, DeFi, Adobe and crypto.
(00:49):
Just to give you a quick information is that the name Gaib comes from, we take inspirationfrom it.
Listen out Gaib.
It's actually an acronym for three things like GPU, AI and blockchain.
That's where we want to focus on.
Then for my background, I was in traditional finance most of the time doing like creditresearch, equity research and then private
(01:12):
Afterwards, I went on to help an exchange on an &A team to help them expand outside ofChina That's why I also been focusing on &A on exchanges, wallet companies, e-sports,
gaming, etc And then I started a crypto VC fund together with my partner to invest intothe AI infrastructure side where we focus on deep tech um Infra, for example, their one
(01:36):
tools cross-chain, bidoware and even some of these like uhAI DeFi infrastructure as well.
So our portfolios include like Eigen and Eigenly, Babylon and the other companies andBecause of that like I have a pretty strong like finance background, but at the same time
I'm also like a half engineer, you know I'm pretty familiar with the whole AI supply chainwhen ChachiBee 3.5 was just launched I was like studying the whole process of how to do
(02:08):
fine-tuning, rack and even at that time the context window forDoing an agent was pretty small.
So I was writing my own agent to make it have like long to memory how to treat how toactually Tokenize different different types of contents and then do a similarity search
and effectively the basis So pretty familiar with pretty familiar with the whole like AIlike supply chain there.
(02:29):
So these background actually gave me uh good experiences and um knowledge inside how AIworks how infrastructure works and more so is when
for our for Gaib's business is actually we're going to tokenizing these GPUs and this AIinfrastructure.
my also my background on finance and crypto also also help a lot on that side as well.
(02:54):
That indeed sounds like a really great background for what you're doing in Gaib.
What was the original problem that drove the interest in creating Gaib as a solution tothat problem?
Yeah.
So there are actually two key reasons why I started Gaiib with my partners there.
(03:16):
So the first reason is I saw that there's actually no actual yield, real yield in crypto.
So we've been playing DeFi for many years.
We have seen a lot of uh token inflated yield inside the market, during the bull market,where you can have very great return.
But when a bear market comes, nowyou will see that all these yield that pays on token emissions actually cannot last, it's
(03:39):
not sustainable.
So what I want to solve is, as Vitalik also mentioned, we can rely on the outborrow insidethe DeFi.
We need to have...
yield that comes from anything external.
So that's where I saw a very big opportunity is actually from the AI sector because we'regoing to AI era.
(04:02):
Everybody is relying on uh these underlying infrastructure, GPUs.
Whenever you make an inference, whenever you do a training on AI model, you needcomputation power and you need computation asset.
And these are the cornerstone assets that can give sustainable yield if you believe in theAI will be adopted globally, sustainably in the future.
(04:22):
So that's why, know, maybe something, this is the yield that we seeing can bring intocrypto.
So that's the first reason.
And the second reason is we want to create a new asset class that allow everybody toinvest into AI.
Because if you think about AI investments, what comes to mind immediately?
I think to most people, it's going to be Nvidia stock, right?
(04:43):
You know, just buy Nvidia stock, hold on to it.
Or sometimes you can, uh there's no other way, right?
So other than buying stock, what can youdo to invest into AI directly.
There's no straightforward methods.
So that's why what I want to do is create an asset class for GPU, the AI infrastructure,these GPU assets.
(05:04):
You can directly invest into them, get whatever yield they generate, get whatever upsideit generates, which is impossible now.
So with crypto, to me, think the key asset of crypto, other than the characteristics ofblockchain being decentralized, being like...
uhuh cannot be tempered, always on-chain.
(05:26):
Other than that, think crypto actually brings a lot of elements on the defi side.
It allows creation of new markets and creation of new assets.
So that's exactly what I want to borrow from crypto and the blockchain to make theseassets tradable and on-chain.
So yeah, these are the two reasons why we started.
uh To recap, it's just one, we want real yield into the market.
(05:50):
Second thing is we want to create a new asset class.
can invest into these AI infrastructure.
Now creating a new asset class sounds very, very ambitious.
um What does that road ahead look like for doing something like that?
I've met with many other projects that are trying to do the same thing, um but there's alot of legal, regulatory, um and also it's a change management thing because it's a new
(06:17):
thing to invest in and people are naturally hesitant.
And so what kind of challenges do you see in creating a new asset class in the GPU space?
Yeah, it's definitely not easy, right?
So if you look at what happened to gold when they were going to mainstream, they struggledfor a long time.
(06:42):
It was only when ETF that got established that people finally got a way to invest intogold easily.
And then we saw exponential growth in terms of the gold economy afterwards.
So that's something also applies to GPUs, these infrastructure sector, becausenumber one to most people, may not understand what is GPU, how do they work, what role do
they play inside the whole AI supply chain.
(07:09):
So the first step is definitely to educate.
When we actually tell them, hey, this yield comes from GPU, comes from AI infrastructure,then they will be like, how?
Where does the yield come from?
There's always a question.
So that's why we have to explain, okay, how does the GPU being rendered out, thecomputation power being used to power all these different AI models and clients.
(07:30):
are paying their money and you as the investor into these assets, you get the money back.
So we have to explain the cash flow, explain the value capture, how GPUs are there.
And the second thing is when we are doing like tokenization, like putting these assetsfrom physical world to on-chain.
So there will be a lot of...
umA lot of doubts like what goes in between like how does tokenization work and how does
payment work?
(07:56):
Where is the transparency?
Well, that's exactly what we want to solve, right?
So we are actually using cryptography and blockchain to solve this problem.
For instance, we have a note network that keeps on tracking every GPU that we tokenize onchain, meaning you as a buyer and investor of these assets, then you can immediately know
(08:17):
like the underlying assetWhere they are?
uh How much how are they performing?
can go all get all these data and this is enabled by blockchain that was They can give youauthenticity and give you like transparency to the other asset which the other traditional
way of investing may not be able to do so and Yeah, then third thing is basically buildingup the markets like I really totalize it you have to let people to understand buy into it
(08:45):
and even be like all the on-chain DeFi components becauseyou want to tap into the existing DeFi infrastructure that we've been built since 2019-20
and even before that we have to build things and develop things in a more composable way.
For example, the tokens can be uh easily integrated with other protocols.
(09:05):
So we have to do lot of these engineering and designing in order to make it happen.
Definitely educating the market is a very crucial step that we have been doing.
That's exactly what you were saying.
When we first came out to do this tokenization of GPU asset since last year around like...
umJuly or this time, thought we are another IONet, another render network or a cash.
(09:33):
They're like, are you doing a compute marketplace again?
Are you tokenizing compute?
We have to explain to them again and again and again.
We're like, we're not, we're different.
We are the financial infrastructure for these assets to live on chain.
That's why we bring these new asset class.
And after six months, after we have been educating the market six months, people are like,wow, this is such a big, big,
(09:56):
like we should look into it, we should invest into it.
So it definitely takes time for the market to understand what we're doing, not only forthe crypto side, but also for the traditional side, because GPU, back financing, GPUs as
an asset, is also pretty new to the traditional markets, which on our end, we're doing atwo-step approach.
Web2 we educate, Web2 also educate, to maximize the reach of these asset class in terms ofthe adoption side.
(10:23):
Yeah, let's go under the hood and talk in detail.
I've met with a number of AI projects that are focused on the GPU space.
And just to give you a sense of the segments that they're targeting.
there's, you you mentioned it, there's the rent a GPU market, which is somewhat of an AWSexcept for compute for AI training.
(10:48):
And then there's also another market whereYou know, what's really interesting about the GPU market is that, at least in crypto, is
that most infrastructure actually borrows money and then they buy the equipment and thenthey pay the, and then there's a debt service that they pay back.
But in crypto, for some reason, we need to have all the money upfront to buy theequipment.
(11:12):
And so I met with another project that actually is providing loans so that GPUs can beacquired and then
the collateral, the GPU becomes the collateral for that loan.
And so that's actually very interesting.
it helps Web3 mature.
Because now we're doing the same things that traditional finance does with borrowing andthen collateralizing that debt and then paying it back with a debt service.
(11:42):
So I think that's very positive.
And then what you guys are doing is very different.
Maybe take us through in detail.
like what it means to tokenize GPU and how the yield and how people can benefit users andpeople that participate can benefit from the yield.
(12:02):
Yeah, correct.
em Yeah.
So I mean the...
I mean the uh model that you mentioned like raising money upfront by the GPUs, use it ascollateral and then pay in the service.
There's something we can do as well, but we are not focusing on that side right nowbecause of the risks level.
So what we do here, um maybe use an example to explain to you in this way.
(12:27):
So you as a cloud company, now you have some GPUs, you have some clients, and then youwant to expand.
um What do you do?
Right now you have get capital.
So you have to get like capital either through three ways.
One is your own pocket, meaning your profit, your revenue.
The other way is to go through like equity financing, raise money from VCs, PEs, equityto, you know, get some money, know, to acquire these GPU assets, host it up and then get
(12:55):
the clients to pay you back.
And then third way is to go through like traditional private lending.
For example, go to family offices, banks, private credit guys and say, Hey,Can we get some money like ah either as a loan or like asset-bet lending side of things?
So to most people they can only tap into the first and second step in the own money andequity financing a lot of them cannot do private lending a public credit ah because GPU is
(13:22):
a session you niche asset class and Not many of them understand it.
um So that's why there's a huge market gap there.
So we exactly here to provide these like um solutions to the gap and what we do is justtake a
you as a cloud company right now, you can come to us like cloud utilizing your GPUstogether with the cash flow and GPUs because you have sold them to the client, right?
(13:47):
So they're making money.
So you cloud utilizing GPU and then with the revenue streams that these clients are payingyou, you pay us back.
essentially, other than upfront, we pay you money to buy the GPUs.
Now you have some GPUs there so you can be cloud utilizing that and then you can use themoney that we give you to buy
some GPUs more as well.
(14:07):
So that's something the model that we do.
And the other model is we can just be like one of the equity owner into the GPUs as well.
For example, if we actually have 50 % worth of whatever these GPU assets, then we canshare the revenue that these GPUs generate, like 50 % of the revenue generated.
(14:28):
So that's more like the revenue share model that we do.
So and the third type is basically a mix of both like minimumterm plus some equity or the revenue share upside that we can work with the companies,
cloud companies there.
So we're pretty flexible in terms of ways that we work with this cloud company.
The whole goal is to make sure that we have the asset backing it.
(14:53):
Second thing is they have a client behind that paying these assets, utilizing it.
We're not just like financing any hardware, it's doing their idle because every minute isa cost.
And most importantly is we onlyfocus on the enterprise-grade GPUs meaning H100, H200, B200.
So the reason is they are number one, they are hard to access by most people if anybodywants to invest in them.
(15:16):
uh They're pretty used, they're used pretty often these days for the big models, datamodel there.
So they have good return profile but most people cannot tap into.
That's why we want to focus on that.
The second thing is these GPUs are the enterprise-grade GPUs are usually being sit uh inthe big data center.
So we can have third party guarantee on these assets that they won't be being Unpluggedand they move away to somewhere else that we cannot track because I said as I mentioned
(15:45):
earlier We have no network that came on tracking these underlying assets to make sure thatthey stay where they are and also to make sure that like there's no Admonities with the
asset value that you know, we're backing the tokenization here So yeah, so we've tokenizedas I mentioned we took line the asset and the cash flow make it into a U-brand asset that
people can invest in tool.
(16:08):
So you're painting a market where there's the supply side and the demand side.
The demand side are people that buying the, correct me if I'm wrong, the demand side arepeople that want yield on these tokenized GPU assets, is that correct?
And then the supply side is you working with these GPU providers and service, um thesedata centers.
(16:34):
umLet me ask a question about that because that that that feels like a uh Traditional web to
kind of process where you create an SPV and then you you you tokenize your assets, etc theI'm curious about the due diligence required because that feels like a lot of work when
And how does that process start?
(16:58):
Does the data center come to guy or do you reach out to them and?
umyou know, how does that start?
And then what does that due diligence looks like look like?
Because I think the level of quality and depth of the due diligence will give confidenceto the market uh to participate in the GAIB network.
(17:19):
Correct, correct, correct.
Yeah, so...
ahStrict due diligence is actually very important on our end because we want to make sure
that like whenever every dollar we give out to these cloud companies actually go in theright hand right place and make sure that we can get them back, right?
So our due diligence process works in two stages.
(17:41):
One is a premium DD meaning we'll be sending out a questionnaire to the cloud company ifthey uh wants to work with us.
Say, hey, we're gonna get some basic information.
Number one is company information, when did you incorporate, who is your client, how muchrevenue you're making, what is the GPU that you're owning, how much they're worth, how
(18:02):
long they've been running for, and then how much financing they're looking for.
after these criteria, the information that we've got, we do an internal review andscreening.
If you find it fit, then we move on to the next step.
And how do you find it fit is, number one,We work with the companies that have GPU first.
(18:24):
So we don't work with anybody like we don't have anything.
And the second thing is we also work with companies that have a client paying them money,making cash flow.
So such that oh our assets, we can make sure that our money can be paid back.
So after these two cards are being matched, then we move on to the next more in-depth duediligence process where we'll be leveraging not only in-house expertise, but also external
(18:47):
expertise here to conduct this.
So uh in-house, because wehave like colleagues and team members that used to do like &A including myself &A equity
financing and also credit financing so we understand the whole we have a very strictframework about the credit framework credit analysis in-house as well like in terms of how
much leverage they're using what is how much money they're making how long these GPUs arebeing used for and then also we'll be leveraging third-party professionals including
(19:12):
auditors to help us do like commercial due diligence financial due diligence to understandthe companies
financial situations and also we will have third-party appraisers to appraise on a GPUasset value to make sure that you know these assets are number one not not that I don't
(19:34):
want to say it's deprecated, they're actual GPUs there and they're actually worth X amountof money and afterwards after all these being done the process is done then we'll move on
to structuring the deal so are we doing like 67 % LTVhow much is the overall customization required, how long is the duration, how much do we
ask for in terms of the payback uh period as a monthly payment.
(20:00):
So we do all the use structuring afterwards, like if they fit.
And then ah after the structuring is done, then we move on to tokenization, offering thisproduct on the platform and through our Web2.1.3 network to anybody who wants to invest
into them.
And the reason why our internal in-house can actually have the expertise and have thedeviance capability is other than us.
(20:21):
come from a financial background.
also have my co-founder actually runs a cloud company himself.
So we have the first-hand information about how much GPU prices are, who is usually theclient inside the industry.
If you claim to have 1,000 GPU, we can immediately cross-check and know whether you have1,000.
Either from our connections, the OEM, ODM side, or...
(20:45):
Yeah, basically the industry is pretty small.
we have like, we know immediately like, is actual asset out there.
So we share these like first-hand information.
I think, I believe we were one of the few teams in the industry that has an actual AIcloud company business running.
So we can leverage both, both analytics piece, like the other professionals and ourin-house to actually conduct the whole process.
(21:10):
And to us, one method the most is risk control.
So we only do.
things that we feel the team and the checkboxes are off-field, then we do it.
Otherwise, we won't risk it.
ohA couple of questions from that.
Thanks for sharing that.
what's the criteria for who you work with?
(21:33):
Is there a certain revenue size that you'll say yes to and then everyone underneath that,below that you say no to?
Because as you were talking, I was thinking about smaller data center providers.
It could be kind of mom and pop, maybe with a couple of H100s or even individuals with anH100, but
(21:54):
and I've rented out some of my GPU space for AI training, but I'd like to tokenize it andthen make some money out of it.
Or arbitrage, and maybe get the money upfront so that I could buy in more equipment.
Do you work with individuals or, when you think of the market, small, medium, or large,and then there's mom and pop, like at the very bottom,
(22:24):
Tell us kind of the, what is the sweet spot for?
Yeah.
yeah.
So um we don't work with individual um right now because it's pretty hard to do the riskcontrol.
For example, if you rock, then it's hard for us to trace you down.
ah Yeah.
And the second thing is we tend to work with like small and medium about size of theselike cloud companies, meaning the asset they have is at least like 30, 40 million dollars
(22:54):
worth of signs above uh for us to do like meaningful entry there.
The reason for working with these big cluster is that these assets are usually beinghosted in a big data center like third-party data center where we can have a tripartite
agreement with them saying now oh after we tokenize these assets we actually own a part ofthat so the initial cloud company cannot just take it out as they wish so we have
(23:23):
additional protection on the asset level side of things but if it's a moment in pop orindividual it's hard for us to
that even in a very legally sound country like the States.
So we tend to work with more institutional size these days.
We may extend it a little bit if the platform grows further and uh we can find better waysto reduce the risk to extend to smaller scale but before that we'll be focusing on the
(23:54):
sweet spot side that I was mentioning just now.
Do you have a BD team that works on the supply side?
And then on the demand side, you mentioned the tokenizing part.
And I'd like to go into that part because after you've done the due diligence and thenyou've decided to uh tokenize a data center, or the GPU is in the data center, and you
(24:16):
have a tripartite agreement with the provider.
And then the tokenization part is selling it or providingmaking it available for individual traders to purchase.
Is that correct?
Yeah, yeah, yeah, correct, correct.
Yeah, so...
more about that side.
(24:38):
Yeah, so for the cloud company and for the supply side, yeah, we do have a BDT.
So we have a growth lead that used to work at Big Deer at the corporate strategy team.
and then he's basically leading most of the discussion with data centers and cloudcompanies.
And sometimes we do talk to, we do do active reach out, but most of the times these arelike close referrals because the GPU and cloud companies is very constant,
(25:08):
like umsmall like because people know people and uh When one company worked with us they
understand how we work they like us then there was just spread the words Spread the wordsaround the circle So that's how we got like over 10 data centers and cloud companies
working with with us recently So when we started there was only one which is my club my mymy co-founder's company and then regarding the the other side of tokenization side uh so
(25:31):
yeah, so um for that endAfter we tokenize it we just put on to a platform for people to subscribe and people want
to buy it and we have like DeFi integration for example right now we have like Pendointegration for our pre-deposit campaign now you can just after you you buy the tokenized
asset you can go to Pendo to do the PT and YT structure that you like and Yeah, so a lotof DeFi integration are coming up uh more than that lending markets other than Pendo will
(26:04):
be live pretty soon as well.
That's really interesting.
em So you mentioned on the supply side, started with one, do you call them customers orpartners or em cloud partners?
So you started with one cloud partner and now you have 10.
(26:27):
Give us a sense of how big that is, how many GPUs and like the size.
We have over 2 billion dollars of asset in the pipeline that we can tokenize.
And then these 2 billion dollars is less than 1 % of the whole total market.
The GPU market is expanding exponentially in the past six months, past six months, past ayear.
(26:52):
Because we're now going into a real AI adoption phase, more AI applications being built inthe past six months.
These AI companies are making money.
Now they can berenting more GPUs to expand the growth and scale up the operation.
m It is very different from two years ago.
Two years ago, most of the companies are burning money.
(27:14):
No revenue at all.
They were on a foundational model, training, raising game that everybody was trying tojust outbid uh ChachiBit, Lama3, that was number two, number three, and going forward.
All these foundational models.
But then,Right now we're seeing the main main utility comes and the revenue comes.
(27:37):
So the growth and demand for the underlying AI infrastructure actually expanded a lot.
And that's why our pipeline also expanded a lot from when we started.
Yeah.
And do you see, so you mentioned the 20 billion and that's less than 1 % or 2 billion,which is, mean, the opportunity is, the TAM is very, very large.
(28:04):
It's less than 0.1 % because actually this is an easy trading market because if you lookat Manifestion 7, they invest almost $300 billion every year to AI infrastructure and that
investment will continue to expand and we're just tapping into a very, very small amountof that these days.
(28:25):
And as these foundation models become more efficient um and as the algorithms that arecreated are become more efficient, do you see at some point we hit an asymptote where we
may not need compute power as much as we did?
Or do you see it just continuing to grow, that we just need more and more compute?
(28:50):
um I don't think we just need more more compute because um as Jensen was saying, for thenext generation model, we need at least 100 times of computation power more.
And by more, it doesn't mean...
um The model is getting bigger.
It basically means that the model now can do more things.
(29:12):
For example, I'll just give you one concrete example.
When DeepSeq R1 came out, V version V and R1 came out, people were like, oh damn, themodels are getting more efficient, Vita should be dying, GPUs are already buying the GPUs
anymore.
But after six months, what happens now?
(29:33):
People realize, oh no, we actually need more computation power.
because the DeepSec R1 and the other similar reasoning model actually needs morecomputation power to think and to research and aggregate information, utilize that and
finally give you an answer.
I would like to use this uh analogy, like the invention of steam engine doesn't make codea legacy, but they actually use more codes in order to power the steam engine.
(30:05):
Meaning, when we actually make these AI models more...
umEfficient we will just use them to do more things for example why we can use reasoning
model these days is because now when you do way more iterations to give you more conciseprecise and concise Answers you don't require that much computation power anymore So
that's why you can do more things and the same thing happens to feed your generation imagegeneration text music generation as well, so we'll just see more demands for using AI and
(30:30):
the computation sideinstead of the reduced usage there.
Yeah.
one thing I'm seeing is companies are creating very domain specific LLMs and that's justgoing to continue to grow.
(30:55):
And they're doing something very interesting that I guess the way I'm thinking about it isChad GPT and these large foundation models, they're good for probably 90 % of the prompts
and questions that we have.
But for like the last 10%, um I'm seeing a lot of companies, they see that as a sweetspot, that they can create something very, very specific for the last 10 % that Chad GPT
(31:23):
cannot.
um that's really exciting.
I'll give you an example.
One, um I'm talking to a company that is creating a um large language model for comedy.
And so they're training this AI.
what funny looks like, what funny sounds like, what's funny to some cultures but notothers.
(31:45):
um And teaching an AI humor is actually a really, really hard work because AI is reallygood at remembering, recall, and then inference, but it's not super good at um qualitative
things like what is sarcasm, what is, um you know, what's a joke.
(32:08):
what's funny.
It's not very good at that.
And so they're actually doing something very pioneering anyway.
But what they're doing is they're also hiring comedians and they're hiring uhscreenwriters to help train this model of like what's funny.
And um that's never been done before.
And that's super exciting.
(32:28):
um And I can see more and more of that type of work just eating up computing resources,which is why we need more.
Definitely, Like even now, um...
If you're asking about more foundational models like robots, we haven't seen concreterobotics models out there that do more generalized functions.
(32:53):
And even for video generation, most of the video generation are very broad spectrum,general, not specific.
If you want to have a video generation that's mainly for, let's say, some artificialeffects for a movie, so you have to train that and put that perspective.
And even for like medically used AI models that not only answer generic questions, butmaybe doing X-ray analysis, doing m cancer cell spotting, you know, these all need to have
(33:24):
more dedicated computation power to train that model to become useful.
So yeah, I totally agree with you.
We will only see more specific use cases, specialized models being set out.
And these, we are all burning computation power.
along the way.
Let's go back to the demand side.
(33:49):
If I'm a participant on Pendle or some of these other DeFi applications, what do I see?
And do I know that I'm participating in the Gaib network?
I guess, what does that look like?
yeah, yeah, umThe basic thing is if you want to engage with us, just go to a website, directly deposit,
but then you want to use DeFi, then you can go to Pendle.
(34:17):
I believe now, I haven't checked today, I believe now we still rank top three in terms ofthe highest yield on Pendle.
If you look at AID, AUSDT, then we should be ranked top three on the Pendle main page.
You can buy that.
the PD token, ah can also go to trade our YT or you can provide equity into the pool.
(34:42):
anything you want depends on what is your risk and how you use what you want outsidethere.
And for the other DeFi applications, we have also announced the guideline on how to engagewith them on our website and uh on our medium post, our blog post there.
So these will actually guide you through uh how to engage with some of these partners.
(35:06):
that have on board at Gayib and integrated with Gayib's products on it.
Got it.
Okay, so AIDAUSDT, is the, uh I guess the, got it, got it.
um And that's the primary product that em individuals will see as they participate inDeFi, is that correct?
(35:30):
AIDAUSDT.
This is only temporary product.
It's more like a pre-deposit product.
We'll be having our mainnet product on soon.
That's why you see AIDA is the alpha and USDC is the uh asset underneath.
Basically, you engage with that by depositing USDC or USDT uh inside it.
(35:53):
And right now it's available only on Pendle or other DeFi applications too.
Right now it's on pendle.
I think a more full integration should be realized as well and Curve integration iscoming.
and the other on the other chain There are other like DeFi dApps that we will beintegrating soon Yeah.
(36:15):
Now if I'm on Pendle and I'm interested in, what is this AIDA USDT thing?
It sounds interesting.
The yield looks good.
I wonder where the yield comes from.
Yep.
And then how do you answer that?
I mean, we've talked about it where you provide financing for data centers and then theyreceive payment and then they use the payment to pay for the financing.
(36:42):
But where is that?
As I'm looking, it doesn't necessarily tell me where the yield comes from.
Yeah, on the token side, does not.
You have to uh go back to a website and also a document and they will show.
(37:03):
And then when the main net product is on, we have transparency board where you see whereare these different deals, where are these GPUs, where's the yield come from basically.
So uh it's going live pretty soon.
got it.
You talked about Mainnet is coming up at some point in the future.
(37:24):
What stage is Gaib right now?
Is it testnet?
We are at Mainnet Alpha, so everything is functional.
We're doing our pre-deposit campaign and afterwards we'll just launch our Mainnet uhproducts version.
Tell us pre-deposit, what does that mean and what is the details of the campaign?
(37:47):
Yes, the AID Alpha is a pre-deposit campaign where depositors can deposit the stablecoinsthat we support.
Right now it's USDC, USDT, and USDR on our website.
Enter the respective vote.
then you'll be getting additional token incentives uh from our end when we actually launchthe token uh for being the initial equity supporters into the vault.
(38:16):
And the capital in the vault will be used to uh deploy into TBOs and some of the financingdeals that are in our pipeline as well.
Got it, got it.
That's really, really interesting.
I can see this taking off in a very big way.
mean, is getting access to these types of deals is, most people don't know where to getaccess to these kinds of deals, but the GP or market is very, very attractive.
(38:46):
And I imagine the interest is,you know, not just on the, on the token side, it's not just from crypto people, but I
imagine if more web two and traditional finance people knew about it, they would beinterested.
What efforts are you making in that, in that segment?
Correct, because of financial background, uh also have been talking with lot moretraditional family office, institutional, even private credit funds who are not into this
(39:17):
sector yet.
So we offer them the same opportunity to invest into these assets ah in a way that theyfeel comfortable.
If they don't want to hold tokens, we can put it under Clustity and directly invest into afund structure we legally set up in that way.
Just do it in the family structure that side as well.
So that's definitely one area that we have been working very closely on.
(39:42):
And we do have some investors uh that are doing this with us right now.
And ultimately, we want to become, what GAI wants to become, is actually a platform.
So right now it's us bringing these different GPUs use online, getting capital from Web2,Web3.
But ultimately, we want to have more operators or underwriters, similar to LIDL.
(40:02):
You have a few operators.
on the LIDL to operate the staking contracts on Ethereum.
So we also want to have multiple operators to be the ones bringing the different GPUdeals, bringing different capital onto our platform.
And even so, we can expand to the other asset classes.
Now we work with GPU, but then maybe 10 years later, we'll be seeing more ASICs, just likewhat happened to mining machines, So we can expand to that.
(40:31):
And even if there's people, if this ishave demands for the other like mining machines, we can also get these asset class on.
So we as a platform is pretty as a no-stake.
We focus on AI asset because now we believe that if we want to get invested into theseasset classes, there's no way you do so.
(40:52):
And we want everybody to have a chance to get in early.
And yeah, that's why we exist.
Yeah.
Well, Kony, thank you for joining us.
Is there anything else that you'd like to talk about that we haven't yet?
No, I think pretty much.
Yeah.
So we covered why, you know, Gayib is here.
What's the problems are you solving?
(41:13):
um Yeah, but please do follow us uh on our Twitter, on our website.
There will be way more exciting things that we're going to announce.
Some of our new initiatives will be coming as well.
So yeah, whether you are interested into AI, interested in DeFi or yield, um just followus.
(41:36):
Join us on the AI revolution.
I'll share all of the URLs in the show notes.
Well, Kony from Gaib, thank you so much.
Thanks Peter.
Yeah, thanks for having me.
It's my honor.
Welcome back to Block by Block, the podcast where we go deep with a builder shaping thenext era of crypto, AI and everything in between.
Today we're joined by Koni, CEO and co-founder of Gaib, a project at the intersection ofdecentralized AI infrastructure and zero-knowledge cryptography.
Koni, welcome to the show.
(00:22):
Thanks, Peter.
Yeah.
Let's begin with your background.
um You are in a very interesting intersection of crypto and AI.
What makes your background a very good fit for this intersection that you're in?
Yeah, so as you mentionedWhat Gaib is doing is actually sitting at the intersection of AI, DeFi, Adobe and crypto.
(00:49):
Just to give you a quick information is that the name Gaib comes from, we take inspirationfrom it.
Listen out Gaib.
It's actually an acronym for three things like GPU, AI and blockchain.
That's where we want to focus on.
Then for my background, I was in traditional finance most of the time doing like creditresearch, equity research and then private
(01:12):
Afterwards, I went on to help an exchange on an &A team to help them expand outside ofChina That's why I also been focusing on &A on exchanges, wallet companies, e-sports,
gaming, etc And then I started a crypto VC fund together with my partner to invest intothe AI infrastructure side where we focus on deep tech um Infra, for example, their one
(01:36):
tools cross-chain, bidoware and even some of these like uhAI DeFi infrastructure as well.
So our portfolios include like Eigen and Eigenly, Babylon and the other companies andBecause of that like I have a pretty strong like finance background, but at the same time
I'm also like a half engineer, you know I'm pretty familiar with the whole AI supply chainwhen ChachiBee 3.5 was just launched I was like studying the whole process of how to do
(02:08):
fine-tuning, rack and even at that time the context window forDoing an agent was pretty small.
So I was writing my own agent to make it have like long to memory how to treat how toactually Tokenize different different types of contents and then do a similarity search
and effectively the basis So pretty familiar with pretty familiar with the whole like AIlike supply chain there.
(02:29):
So these background actually gave me uh good experiences and um knowledge inside how AIworks how infrastructure works and more so is when
for our for Gaib's business is actually we're going to tokenizing these GPUs and this AIinfrastructure.
my also my background on finance and crypto also also help a lot on that side as well.
(02:54):
That indeed sounds like a really great background for what you're doing in Gaib.
What was the original problem that drove the interest in creating Gaib as a solution tothat problem?
Yeah.
So there are actually two key reasons why I started Gaiib with my partners there.
(03:16):
So the first reason is I saw that there's actually no actual yield, real yield in crypto.
So we've been playing DeFi for many years.
We have seen a lot of uh token inflated yield inside the market, during the bull market,where you can have very great return.
But when a bear market comes, nowyou will see that all these yield that pays on token emissions actually cannot last, it's
(03:39):
not sustainable.
So what I want to solve is, as Vitalik also mentioned, we can rely on the outborrow insidethe DeFi.
We need to have...
yield that comes from anything external.
So that's where I saw a very big opportunity is actually from the AI sector because we'regoing to AI era.
(04:02):
Everybody is relying on uh these underlying infrastructure, GPUs.
Whenever you make an inference, whenever you do a training on AI model, you needcomputation power and you need computation asset.
And these are the cornerstone assets that can give sustainable yield if you believe in theAI will be adopted globally, sustainably in the future.
(04:22):
So that's why, know, maybe something, this is the yield that we seeing can bring intocrypto.
So that's the first reason.
And the second reason is we want to create a new asset class that allow everybody toinvest into AI.
Because if you think about AI investments, what comes to mind immediately?
I think to most people, it's going to be Nvidia stock, right?
(04:43):
You know, just buy Nvidia stock, hold on to it.
Or sometimes you can, uh there's no other way, right?
So other than buying stock, what can youdo to invest into AI directly.
There's no straightforward methods.
So that's why what I want to do is create an asset class for GPU, the AI infrastructure,these GPU assets.
(05:04):
You can directly invest into them, get whatever yield they generate, get whatever upsideit generates, which is impossible now.
So with crypto, to me, think the key asset of crypto, other than the characteristics ofblockchain being decentralized, being like...
uhuh cannot be tempered, always on-chain.
(05:26):
Other than that, think crypto actually brings a lot of elements on the defi side.
It allows creation of new markets and creation of new assets.
So that's exactly what I want to borrow from crypto and the blockchain to make theseassets tradable and on-chain.
So yeah, these are the two reasons why we started.
uh To recap, it's just one, we want real yield into the market.
(05:50):
Second thing is we want to create a new asset class.
can invest into these AI infrastructure.
Now creating a new asset class sounds very, very ambitious.
um What does that road ahead look like for doing something like that?
I've met with many other projects that are trying to do the same thing, um but there's alot of legal, regulatory, um and also it's a change management thing because it's a new
(06:17):
thing to invest in and people are naturally hesitant.
And so what kind of challenges do you see in creating a new asset class in the GPU space?
Yeah, it's definitely not easy, right?
So if you look at what happened to gold when they were going to mainstream, they struggledfor a long time.
(06:42):
It was only when ETF that got established that people finally got a way to invest intogold easily.
And then we saw exponential growth in terms of the gold economy afterwards.
So that's something also applies to GPUs, these infrastructure sector, becausenumber one to most people, may not understand what is GPU, how do they work, what role do
they play inside the whole AI supply chain.
(07:09):
So the first step is definitely to educate.
When we actually tell them, hey, this yield comes from GPU, comes from AI infrastructure,then they will be like, how?
Where does the yield come from?
There's always a question.
So that's why we have to explain, okay, how does the GPU being rendered out, thecomputation power being used to power all these different AI models and clients.
(07:30):
are paying their money and you as the investor into these assets, you get the money back.
So we have to explain the cash flow, explain the value capture, how GPUs are there.
And the second thing is when we are doing like tokenization, like putting these assetsfrom physical world to on-chain.
So there will be a lot of...
umA lot of doubts like what goes in between like how does tokenization work and how does
payment work?
(07:56):
Where is the transparency?
Well, that's exactly what we want to solve, right?
So we are actually using cryptography and blockchain to solve this problem.
For instance, we have a note network that keeps on tracking every GPU that we tokenize onchain, meaning you as a buyer and investor of these assets, then you can immediately know
(08:17):
like the underlying assetWhere they are?
uh How much how are they performing?
can go all get all these data and this is enabled by blockchain that was They can give youauthenticity and give you like transparency to the other asset which the other traditional
way of investing may not be able to do so and Yeah, then third thing is basically buildingup the markets like I really totalize it you have to let people to understand buy into it
(08:45):
and even be like all the on-chain DeFi components becauseyou want to tap into the existing DeFi infrastructure that we've been built since 2019-20
and even before that we have to build things and develop things in a more composable way.
For example, the tokens can be uh easily integrated with other protocols.
(09:05):
So we have to do lot of these engineering and designing in order to make it happen.
Definitely educating the market is a very crucial step that we have been doing.
That's exactly what you were saying.
When we first came out to do this tokenization of GPU asset since last year around like...
umJuly or this time, thought we are another IONet, another render network or a cash.
(09:33):
They're like, are you doing a compute marketplace again?
Are you tokenizing compute?
We have to explain to them again and again and again.
We're like, we're not, we're different.
We are the financial infrastructure for these assets to live on chain.
That's why we bring these new asset class.
And after six months, after we have been educating the market six months, people are like,wow, this is such a big, big,
(09:56):
like we should look into it, we should invest into it.
So it definitely takes time for the market to understand what we're doing, not only forthe crypto side, but also for the traditional side, because GPU, back financing, GPUs as
an asset, is also pretty new to the traditional markets, which on our end, we're doing atwo-step approach.
Web2 we educate, Web2 also educate, to maximize the reach of these asset class in terms ofthe adoption side.
(10:23):
Yeah, let's go under the hood and talk in detail.
I've met with a number of AI projects that are focused on the GPU space.
And just to give you a sense of the segments that they're targeting.
there's, you you mentioned it, there's the rent a GPU market, which is somewhat of an AWSexcept for compute for AI training.
(10:48):
And then there's also another market whereYou know, what's really interesting about the GPU market is that, at least in crypto, is
that most infrastructure actually borrows money and then they buy the equipment and thenthey pay the, and then there's a debt service that they pay back.
But in crypto, for some reason, we need to have all the money upfront to buy theequipment.
(11:12):
And so I met with another project that actually is providing loans so that GPUs can beacquired and then
the collateral, the GPU becomes the collateral for that loan.
And so that's actually very interesting.
it helps Web3 mature.
Because now we're doing the same things that traditional finance does with borrowing andthen collateralizing that debt and then paying it back with a debt service.
(11:42):
So I think that's very positive.
And then what you guys are doing is very different.
Maybe take us through in detail.
like what it means to tokenize GPU and how the yield and how people can benefit users andpeople that participate can benefit from the yield.
(12:02):
Yeah, correct.
em Yeah.
So I mean the...
I mean the uh model that you mentioned like raising money upfront by the GPUs, use it ascollateral and then pay in the service.
There's something we can do as well, but we are not focusing on that side right nowbecause of the risks level.
So what we do here, um maybe use an example to explain to you in this way.
(12:27):
So you as a cloud company, now you have some GPUs, you have some clients, and then youwant to expand.
um What do you do?
Right now you have get capital.
So you have to get like capital either through three ways.
One is your own pocket, meaning your profit, your revenue.
The other way is to go through like equity financing, raise money from VCs, PEs, equityto, you know, get some money, know, to acquire these GPU assets, host it up and then get
(12:55):
the clients to pay you back.
And then third way is to go through like traditional private lending.
For example, go to family offices, banks, private credit guys and say, Hey,Can we get some money like ah either as a loan or like asset-bet lending side of things?
So to most people they can only tap into the first and second step in the own money andequity financing a lot of them cannot do private lending a public credit ah because GPU is
(13:22):
a session you niche asset class and Not many of them understand it.
um So that's why there's a huge market gap there.
So we exactly here to provide these like um solutions to the gap and what we do is justtake a
you as a cloud company right now, you can come to us like cloud utilizing your GPUstogether with the cash flow and GPUs because you have sold them to the client, right?
(13:47):
So they're making money.
So you cloud utilizing GPU and then with the revenue streams that these clients are payingyou, you pay us back.
essentially, other than upfront, we pay you money to buy the GPUs.
Now you have some GPUs there so you can be cloud utilizing that and then you can use themoney that we give you to buy
some GPUs more as well.
(14:07):
So that's something the model that we do.
And the other model is we can just be like one of the equity owner into the GPUs as well.
For example, if we actually have 50 % worth of whatever these GPU assets, then we canshare the revenue that these GPUs generate, like 50 % of the revenue generated.
(14:28):
So that's more like the revenue share model that we do.
So and the third type is basically a mix of both like minimumterm plus some equity or the revenue share upside that we can work with the companies,
cloud companies there.
So we're pretty flexible in terms of ways that we work with this cloud company.
The whole goal is to make sure that we have the asset backing it.
(14:53):
Second thing is they have a client behind that paying these assets, utilizing it.
We're not just like financing any hardware, it's doing their idle because every minute isa cost.
And most importantly is we onlyfocus on the enterprise-grade GPUs meaning H100, H200, B200.
So the reason is they are number one, they are hard to access by most people if anybodywants to invest in them.
(15:16):
uh They're pretty used, they're used pretty often these days for the big models, datamodel there.
So they have good return profile but most people cannot tap into.
That's why we want to focus on that.
The second thing is these GPUs are the enterprise-grade GPUs are usually being sit uh inthe big data center.
So we can have third party guarantee on these assets that they won't be being Unpluggedand they move away to somewhere else that we cannot track because I said as I mentioned
(15:45):
earlier We have no network that came on tracking these underlying assets to make sure thatthey stay where they are and also to make sure that like there's no Admonities with the
asset value that you know, we're backing the tokenization here So yeah, so we've tokenizedas I mentioned we took line the asset and the cash flow make it into a U-brand asset that
people can invest in tool.
(16:08):
So you're painting a market where there's the supply side and the demand side.
The demand side are people that buying the, correct me if I'm wrong, the demand side arepeople that want yield on these tokenized GPU assets, is that correct?
And then the supply side is you working with these GPU providers and service, um thesedata centers.
(16:34):
umLet me ask a question about that because that that that feels like a uh Traditional web to
kind of process where you create an SPV and then you you you tokenize your assets, etc theI'm curious about the due diligence required because that feels like a lot of work when
And how does that process start?
(16:58):
Does the data center come to guy or do you reach out to them and?
umyou know, how does that start?
And then what does that due diligence looks like look like?
Because I think the level of quality and depth of the due diligence will give confidenceto the market uh to participate in the GAIB network.
(17:19):
Correct, correct, correct.
Yeah, so...
ahStrict due diligence is actually very important on our end because we want to make sure
that like whenever every dollar we give out to these cloud companies actually go in theright hand right place and make sure that we can get them back, right?
So our due diligence process works in two stages.
(17:41):
One is a premium DD meaning we'll be sending out a questionnaire to the cloud company ifthey uh wants to work with us.
Say, hey, we're gonna get some basic information.
Number one is company information, when did you incorporate, who is your client, how muchrevenue you're making, what is the GPU that you're owning, how much they're worth, how
(18:02):
long they've been running for, and then how much financing they're looking for.
after these criteria, the information that we've got, we do an internal review andscreening.
If you find it fit, then we move on to the next step.
And how do you find it fit is, number one,We work with the companies that have GPU first.
(18:24):
So we don't work with anybody like we don't have anything.
And the second thing is we also work with companies that have a client paying them money,making cash flow.
So such that oh our assets, we can make sure that our money can be paid back.
So after these two cards are being matched, then we move on to the next more in-depth duediligence process where we'll be leveraging not only in-house expertise, but also external
(18:47):
expertise here to conduct this.
So uh in-house, because wehave like colleagues and team members that used to do like &A including myself &A equity
financing and also credit financing so we understand the whole we have a very strictframework about the credit framework credit analysis in-house as well like in terms of how
much leverage they're using what is how much money they're making how long these GPUs arebeing used for and then also we'll be leveraging third-party professionals including
(19:12):
auditors to help us do like commercial due diligence financial due diligence to understandthe companies
financial situations and also we will have third-party appraisers to appraise on a GPUasset value to make sure that you know these assets are number one not not that I don't
(19:34):
want to say it's deprecated, they're actual GPUs there and they're actually worth X amountof money and afterwards after all these being done the process is done then we'll move on
to structuring the deal so are we doing like 67 % LTVhow much is the overall customization required, how long is the duration, how much do we
ask for in terms of the payback uh period as a monthly payment.
(20:00):
So we do all the use structuring afterwards, like if they fit.
And then ah after the structuring is done, then we move on to tokenization, offering thisproduct on the platform and through our Web2.1.3 network to anybody who wants to invest
into them.
And the reason why our internal in-house can actually have the expertise and have thedeviance capability is other than us.
(20:21):
come from a financial background.
also have my co-founder actually runs a cloud company himself.
So we have the first-hand information about how much GPU prices are, who is usually theclient inside the industry.
If you claim to have 1,000 GPU, we can immediately cross-check and know whether you have1,000.
Either from our connections, the OEM, ODM side, or...
(20:45):
Yeah, basically the industry is pretty small.
we have like, we know immediately like, is actual asset out there.
So we share these like first-hand information.
I think, I believe we were one of the few teams in the industry that has an actual AIcloud company business running.
So we can leverage both, both analytics piece, like the other professionals and ourin-house to actually conduct the whole process.
(21:10):
And to us, one method the most is risk control.
So we only do.
things that we feel the team and the checkboxes are off-field, then we do it.
Otherwise, we won't risk it.
ohA couple of questions from that.
Thanks for sharing that.
what's the criteria for who you work with?
(21:33):
Is there a certain revenue size that you'll say yes to and then everyone underneath that,below that you say no to?
Because as you were talking, I was thinking about smaller data center providers.
It could be kind of mom and pop, maybe with a couple of H100s or even individuals with anH100, but
(21:54):
and I've rented out some of my GPU space for AI training, but I'd like to tokenize it andthen make some money out of it.
Or arbitrage, and maybe get the money upfront so that I could buy in more equipment.
Do you work with individuals or, when you think of the market, small, medium, or large,and then there's mom and pop, like at the very bottom,
(22:24):
Tell us kind of the, what is the sweet spot for?
Yeah.
yeah.
So um we don't work with individual um right now because it's pretty hard to do the riskcontrol.
For example, if you rock, then it's hard for us to trace you down.
ah Yeah.
And the second thing is we tend to work with like small and medium about size of theselike cloud companies, meaning the asset they have is at least like 30, 40 million dollars
(22:54):
worth of signs above uh for us to do like meaningful entry there.
The reason for working with these big cluster is that these assets are usually beinghosted in a big data center like third-party data center where we can have a tripartite
agreement with them saying now oh after we tokenize these assets we actually own a part ofthat so the initial cloud company cannot just take it out as they wish so we have
(23:23):
additional protection on the asset level side of things but if it's a moment in pop orindividual it's hard for us to
that even in a very legally sound country like the States.
So we tend to work with more institutional size these days.
We may extend it a little bit if the platform grows further and uh we can find better waysto reduce the risk to extend to smaller scale but before that we'll be focusing on the
(23:54):
sweet spot side that I was mentioning just now.
Do you have a BD team that works on the supply side?
And then on the demand side, you mentioned the tokenizing part.
And I'd like to go into that part because after you've done the due diligence and thenyou've decided to uh tokenize a data center, or the GPU is in the data center, and you
(24:16):
have a tripartite agreement with the provider.
And then the tokenization part is selling it or providingmaking it available for individual traders to purchase.
Is that correct?
Yeah, yeah, yeah, correct, correct.
Yeah, so...
more about that side.
(24:38):
Yeah, so for the cloud company and for the supply side, yeah, we do have a BDT.
So we have a growth lead that used to work at Big Deer at the corporate strategy team.
and then he's basically leading most of the discussion with data centers and cloudcompanies.
And sometimes we do talk to, we do do active reach out, but most of the times these arelike close referrals because the GPU and cloud companies is very constant,
(25:08):
like umsmall like because people know people and uh When one company worked with us they
understand how we work they like us then there was just spread the words Spread the wordsaround the circle So that's how we got like over 10 data centers and cloud companies
working with with us recently So when we started there was only one which is my club my mymy co-founder's company and then regarding the the other side of tokenization side uh so
(25:31):
yeah, so um for that endAfter we tokenize it we just put on to a platform for people to subscribe and people want
to buy it and we have like DeFi integration for example right now we have like Pendointegration for our pre-deposit campaign now you can just after you you buy the tokenized
asset you can go to Pendo to do the PT and YT structure that you like and Yeah, so a lotof DeFi integration are coming up uh more than that lending markets other than Pendo will
(26:04):
be live pretty soon as well.
That's really interesting.
em So you mentioned on the supply side, started with one, do you call them customers orpartners or em cloud partners?
So you started with one cloud partner and now you have 10.
(26:27):
Give us a sense of how big that is, how many GPUs and like the size.
We have over 2 billion dollars of asset in the pipeline that we can tokenize.
And then these 2 billion dollars is less than 1 % of the whole total market.
The GPU market is expanding exponentially in the past six months, past six months, past ayear.
(26:52):
Because we're now going into a real AI adoption phase, more AI applications being built inthe past six months.
These AI companies are making money.
Now they can berenting more GPUs to expand the growth and scale up the operation.
m It is very different from two years ago.
Two years ago, most of the companies are burning money.
(27:14):
No revenue at all.
They were on a foundational model, training, raising game that everybody was trying tojust outbid uh ChachiBit, Lama3, that was number two, number three, and going forward.
All these foundational models.
But then,Right now we're seeing the main main utility comes and the revenue comes.
(27:37):
So the growth and demand for the underlying AI infrastructure actually expanded a lot.
And that's why our pipeline also expanded a lot from when we started.
Yeah.
And do you see, so you mentioned the 20 billion and that's less than 1 % or 2 billion,which is, mean, the opportunity is, the TAM is very, very large.
(28:04):
It's less than 0.1 % because actually this is an easy trading market because if you lookat Manifestion 7, they invest almost $300 billion every year to AI infrastructure and that
investment will continue to expand and we're just tapping into a very, very small amountof that these days.
(28:25):
And as these foundation models become more efficient um and as the algorithms that arecreated are become more efficient, do you see at some point we hit an asymptote where we
may not need compute power as much as we did?
Or do you see it just continuing to grow, that we just need more and more compute?
(28:50):
um I don't think we just need more more compute because um as Jensen was saying, for thenext generation model, we need at least 100 times of computation power more.
And by more, it doesn't mean...
um The model is getting bigger.
It basically means that the model now can do more things.
(29:12):
For example, I'll just give you one concrete example.
When DeepSeq R1 came out, V version V and R1 came out, people were like, oh damn, themodels are getting more efficient, Vita should be dying, GPUs are already buying the GPUs
anymore.
But after six months, what happens now?
(29:33):
People realize, oh no, we actually need more computation power.
because the DeepSec R1 and the other similar reasoning model actually needs morecomputation power to think and to research and aggregate information, utilize that and
finally give you an answer.
I would like to use this uh analogy, like the invention of steam engine doesn't make codea legacy, but they actually use more codes in order to power the steam engine.
(30:05):
Meaning, when we actually make these AI models more...
umEfficient we will just use them to do more things for example why we can use reasoning
model these days is because now when you do way more iterations to give you more conciseprecise and concise Answers you don't require that much computation power anymore So
that's why you can do more things and the same thing happens to feed your generation imagegeneration text music generation as well, so we'll just see more demands for using AI and
(30:30):
the computation sideinstead of the reduced usage there.
Yeah.
one thing I'm seeing is companies are creating very domain specific LLMs and that's justgoing to continue to grow.
(30:55):
And they're doing something very interesting that I guess the way I'm thinking about it isChad GPT and these large foundation models, they're good for probably 90 % of the prompts
and questions that we have.
But for like the last 10%, um I'm seeing a lot of companies, they see that as a sweetspot, that they can create something very, very specific for the last 10 % that Chad GPT
(31:23):
cannot.
um that's really exciting.
I'll give you an example.
One, um I'm talking to a company that is creating a um large language model for comedy.
And so they're training this AI.
what funny looks like, what funny sounds like, what's funny to some cultures but notothers.
(31:45):
um And teaching an AI humor is actually a really, really hard work because AI is reallygood at remembering, recall, and then inference, but it's not super good at um qualitative
things like what is sarcasm, what is, um you know, what's a joke.
(32:08):
what's funny.
It's not very good at that.
And so they're actually doing something very pioneering anyway.
But what they're doing is they're also hiring comedians and they're hiring uhscreenwriters to help train this model of like what's funny.
And um that's never been done before.
And that's super exciting.
(32:28):
um And I can see more and more of that type of work just eating up computing resources,which is why we need more.
Definitely, Like even now, um...
If you're asking about more foundational models like robots, we haven't seen concreterobotics models out there that do more generalized functions.
(32:53):
And even for video generation, most of the video generation are very broad spectrum,general, not specific.
If you want to have a video generation that's mainly for, let's say, some artificialeffects for a movie, so you have to train that and put that perspective.
And even for like medically used AI models that not only answer generic questions, butmaybe doing X-ray analysis, doing m cancer cell spotting, you know, these all need to have
(33:24):
more dedicated computation power to train that model to become useful.
So yeah, I totally agree with you.
We will only see more specific use cases, specialized models being set out.
And these, we are all burning computation power.
along the way.
Let's go back to the demand side.
(33:49):
If I'm a participant on Pendle or some of these other DeFi applications, what do I see?
And do I know that I'm participating in the Gaib network?
I guess, what does that look like?
yeah, yeah, umThe basic thing is if you want to engage with us, just go to a website, directly deposit,
but then you want to use DeFi, then you can go to Pendle.
(34:17):
I believe now, I haven't checked today, I believe now we still rank top three in terms ofthe highest yield on Pendle.
If you look at AID, AUSDT, then we should be ranked top three on the Pendle main page.
You can buy that.
the PD token, ah can also go to trade our YT or you can provide equity into the pool.
(34:42):
anything you want depends on what is your risk and how you use what you want outsidethere.
And for the other DeFi applications, we have also announced the guideline on how to engagewith them on our website and uh on our medium post, our blog post there.
So these will actually guide you through uh how to engage with some of these partners.
(35:06):
that have on board at Gayib and integrated with Gayib's products on it.
Got it.
Okay, so AIDAUSDT, is the, uh I guess the, got it, got it.
um And that's the primary product that em individuals will see as they participate inDeFi, is that correct?
(35:30):
AIDAUSDT.
This is only temporary product.
It's more like a pre-deposit product.
We'll be having our mainnet product on soon.
That's why you see AIDA is the alpha and USDC is the uh asset underneath.
Basically, you engage with that by depositing USDC or USDT uh inside it.
(35:53):
And right now it's available only on Pendle or other DeFi applications too.
Right now it's on pendle.
I think a more full integration should be realized as well and Curve integration iscoming.
and the other on the other chain There are other like DeFi dApps that we will beintegrating soon Yeah.
(36:15):
Now if I'm on Pendle and I'm interested in, what is this AIDA USDT thing?
It sounds interesting.
The yield looks good.
I wonder where the yield comes from.
Yep.
And then how do you answer that?
I mean, we've talked about it where you provide financing for data centers and then theyreceive payment and then they use the payment to pay for the financing.
(36:42):
But where is that?
As I'm looking, it doesn't necessarily tell me where the yield comes from.
Yeah, on the token side, does not.
You have to uh go back to a website and also a document and they will show.
(37:03):
And then when the main net product is on, we have transparency board where you see whereare these different deals, where are these GPUs, where's the yield come from basically.
So uh it's going live pretty soon.
got it.
You talked about Mainnet is coming up at some point in the future.
(37:24):
What stage is Gaib right now?
Is it testnet?
We are at Mainnet Alpha, so everything is functional.
We're doing our pre-deposit campaign and afterwards we'll just launch our Mainnet uhproducts version.
Tell us pre-deposit, what does that mean and what is the details of the campaign?
(37:47):
Yes, the AID Alpha is a pre-deposit campaign where depositors can deposit the stablecoinsthat we support.
Right now it's USDC, USDT, and USDR on our website.
Enter the respective vote.
then you'll be getting additional token incentives uh from our end when we actually launchthe token uh for being the initial equity supporters into the vault.
(38:16):
And the capital in the vault will be used to uh deploy into TBOs and some of the financingdeals that are in our pipeline as well.
Got it, got it.
That's really, really interesting.
I can see this taking off in a very big way.
mean, is getting access to these types of deals is, most people don't know where to getaccess to these kinds of deals, but the GP or market is very, very attractive.
(38:46):
And I imagine the interest is,you know, not just on the, on the token side, it's not just from crypto people, but I
imagine if more web two and traditional finance people knew about it, they would beinterested.
What efforts are you making in that, in that segment?
Correct, because of financial background, uh also have been talking with lot moretraditional family office, institutional, even private credit funds who are not into this
(39:17):
sector yet.
So we offer them the same opportunity to invest into these assets ah in a way that theyfeel comfortable.
If they don't want to hold tokens, we can put it under Clustity and directly invest into afund structure we legally set up in that way.
Just do it in the family structure that side as well.
So that's definitely one area that we have been working very closely on.
(39:42):
And we do have some investors uh that are doing this with us right now.
And ultimately, we want to become, what GAI wants to become, is actually a platform.
So right now it's us bringing these different GPUs use online, getting capital from Web2,Web3.
But ultimately, we want to have more operators or underwriters, similar to LIDL.
(40:02):
You have a few operators.
on the LIDL to operate the staking contracts on Ethereum.
So we also want to have multiple operators to be the ones bringing the different GPUdeals, bringing different capital onto our platform.
And even so, we can expand to the other asset classes.
Now we work with GPU, but then maybe 10 years later, we'll be seeing more ASICs, just likewhat happened to mining machines, So we can expand to that.
(40:31):
And even if there's people, if this ishave demands for the other like mining machines, we can also get these asset class on.
So we as a platform is pretty as a no-stake.
We focus on AI asset because now we believe that if we want to get invested into theseasset classes, there's no way you do so.
(40:52):
And we want everybody to have a chance to get in early.
And yeah, that's why we exist.
Yeah.
Well, Kony, thank you for joining us.
Is there anything else that you'd like to talk about that we haven't yet?
No, I think pretty much.
Yeah.
So we covered why, you know, Gayib is here.
What's the problems are you solving?
(41:13):
um Yeah, but please do follow us uh on our Twitter, on our website.
There will be way more exciting things that we're going to announce.
Some of our new initiatives will be coming as well.
So yeah, whether you are interested into AI, interested in DeFi or yield, um just followus.
(41:36):
Join us on the AI revolution.
I'll share all of the URLs in the show notes.
Well, Kony from Gaib, thank you so much.
Thanks Peter.
Yeah, thanks for having me.
It's my honor.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Betrayal Season 5
Saskia Inwood woke up one morning, knowing her life would never be the same. The night before, she learned the unimaginable – that the husband she knew in the light of day was a different person after dark. This season unpacks Saskia’s discovery of her husband’s secret life and her fight to bring him to justice. Along the way, we expose a crime that is just coming to light. This is also a story about the myth of the “perfect victim:” who gets believed, who gets doubted, and why. We follow Saskia as she works to reclaim her body, her voice, and her life. If you would like to reach out to the Betrayal Team, email us at betrayalpod@gmail.com. Follow us on Instagram @betrayalpod and @glasspodcasts. Please join our Substack for additional exclusive content, curated book recommendations, and community discussions. Sign up FREE by clicking this link Beyond Betrayal Substack. Join our community dedicated to truth, resilience, and healing. Your voice matters! Be a part of our Betrayal journey on Substack.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com