Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
The Data Layer AI Actually Needs. Now in Early Access.
Recorded: March 12, 2026 Duration: 1:02:00Space Recording
Full Transcription
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music Music Music Music Music Music Music Music Music Music Music I'm The Oh Hello friends, good morning.
Can I get a thumbs up if you can hear me?
Nice, thank you. Welcome Avery.
GM folks.
GM GM.
I see we have Pranav in the audience.
What are you doing today or is it...?
Hello, hello. GM.
Alright. Uh oh. Spaces might be ragging us. I still see Pranav in the audience.
Can you hear me? in the audience.
Can you hear me?
Yeah, I can hear you, but you still look like you're a listener.
No, I did show speaker.
Yeah, okay, I'll be right back because my app is lagging. Okay.
Okay.
All right.
I can see Avery and Pran off on stage with me this is good
all right looks like we have some here
hello you hear me okay? I can. Welcome.
All right. Thank you for your patience everyone.
While we get everyone on stage, we'll get started here in just a couple minutes. Okay. okay
is um
is Samir's show
is on stage
for anyone else
it shows on my for anyone else? It shows on my...
For me, at least.
Yeah.
Yeah, it must just be me this morning.
I've already quit the app and restarted it,
but it still shows you as a listener.
Yeah, same here.
But as long as he can speak, we are good.
I see him as speaker.
Samir's a speaker in my...
All right.
We're going to try this.
I think we're going to get one more person now. Yeah. All right. We're going to try this. I think we're going to get one more person, though.
Yeah.
All right.
I see Austin has requested.
I'm going to add you as a speaker.
Okay. We've got the double zero count. Very nice. Okay.
Is it just me or are we all here?
Ready to go.
Awesome. Awesome. Well, I mean, i think we're right on time that's uh it's pretty epic
for this early in the morning so i guess um we'll get it kicked off if um everyone actually i don't
know um austin can we hear you?
I cannot hear Austin.
Can anyone else hear Austin?
Oh. Hey guys, can you hear me?
I can.
All right.
Okay, so I think that's everyone then, right?
Yep.
I think my space is a super... Oh, no, there we go.
Wouldn't let me mute or unmute for a second.
Yeah, it's rugging me this morning too, so we're in for a wild ride, but
I think we're just going to have to roll with it.
All right, well, let's try to get this
kicked off then. Thank you everyone for joining us at such an
early hour. It's 8 o'clock over here on the West Coast.
... early hour. It's eight o'clock over here on the West Coast. Oh, wouldn't let me unmute for a second. All right. This should be fun. Let's go ahead and
kick it off. Just wanted to bring everyone together and just kind of tell a little bit of the story about how we came to be here.
So let's start off. Why don't we start with Pranav, if you want to kind of introduce yourself.
We'll go around the room.
Hello, hello. It's early here in West Coast and probably late somewhere in the other corner of the world, where a bunch of our communities are dialing from. I'm Pranav, I'm GM of Shelby,
leading Shelby, and previously I was leading infrastructure at Aptos Labs and super excited,
super stoked that early access is out there and great to have some amazing co-speakers, co-conspirators
in helping us bring this to life.
And so looking forward to this conversation.
Awesome.
Well, thank you so much for being here.
Austin, let's kick it over to you.
Welcome, Austin.
Oh, hold on.
Austin got kicked out.
All right.
Welcome back, Austin.
All right.
Let's see.
Can you guys hear me now?
Is this good?
All right.
Yes.
Perfect.
Yeah, I was just introducing you, so welcome back, Austin,
if you wouldn't mind introducing yourself.
Oh, man, space is best today. Great.
Yeah, I'm Austin Federa.
I am one of the three co-founders of the Double Zero Network.
Welcome, welcome. good to have you. And let's see, Samir,
it looks like hopefully you're still here. Samir, welcome, how are you? I'm good, thank you. I hope you can hear me. So I'm Samir, I'm crypto research lead at Jump Crypto. I've been in the crypto space as part of the
trading organization for about seven years now. And I'm one of the co-builders of and collaborators
of Shelby, like where we are, where Jump Crypto is building one of their Shelby clients in much the same way as we're building
some of the other things in the crypto space.
Awesome.
Well, yeah, great to have you here as well.
And last but not least,
we have Dr. Avery Ching.
Welcome.
Hey, folks.
Avery, co-founder and CEO at Aptos Labs.
I know we only have 45 minutes,
so let's just get into it.
That's right.
All right, everyone.
So glad to have you here.
Let's dive right into it.
We're going to do a warm-up lightning round.
This is, I don't know, maybe 30, 40 seconds each of you.
We're going to start with, I don't know, you know, kind of like a hot take, right?
So Pranav, let's kick it off with you.
Let's just kind of walk through, from your perspective,
what is the biggest assumption about AI infrastructure that's going to be proven wrong in the next two years?
Oh boy, this is such a rapidly evolving area.
But I feel we are reaching a point where just throwing more compute on LLMs equal AGI.
I think that equation is kind of breaking.
And I'm looking forward to having a kind of completely new breakthrough in the
foundational model like my i'm quite excited about the jeppa model that jan lukom is uh is pushing
for like build a world model not language language itself is quite insufficient even to communicate
and language does not mean complete intelligence so that's that's part one and then more practically
i would say a lot
of inference should move towards the edge rather than towards a massive kind of centralized
infra. So I think that those are my two hot takes.
Yeah. No, I mean, I tend to agree with you. All right. It looks like Austin is just coming
back up. So I'm going to repeat the question just in case you weren't able to hear it. We're starting off with just, you know, kind of hot takes. I'm going to kick it over to you. And the question is, what's the biggest assumption about AI infrastructure that's going to be proven wrong in the next two years?
Proven wrong? Okay. First off, disclaimer, my space is kind of borked at the moment. So I think a lot of things are on pretty heavy delay. So hopefully I'm not speaking over anyone at the moment. What's going to be proven wrong? Okay, so I think one of the main things we're going to see proven wrong is that the incumbent companies that are leading today have this unassailable gap in anything other than
power generation. I do think there's a real, real moat in access to gigawatt hours, but that the
fundamental idea that, I mean, we see this on a daily basis, right? Like Claude is better for one
month, and then ChatGPT is better for another another month and then DeepBook comes up. And so I do think the idea that like the incumbents have already been established is, I mean, we thought that about AOL as well back in the 90s.
You've got mail.
Yeah, 100%. No, I think that's an absolutely great point.
Samir, I think hopefully you heard the question. What are your thoughts? What's going to be proven wrong?
So coming at this from a pure engineering perspective, Amdahl's Law, which is, for those not familiar with it, it can be stated as the overall performance gained by optimizing a single part of a system is constrained by its most significant bottleneck.
And that has always been and continues to be a movement of information and not actually
compute and GPUs and power.
And what you see is that the vast majority of investment in all AI infrastructure is
mostly in bigger data centers, more power, faster GPUs and inference devices,
ASICs, FPGAs, and so on. But no one has really invested in or put any real thought into actually
how you keep them fed with information to make them useful. That's where we are building and
where we think that there's going to have to be additional investment.
Yeah. Yeah, I agree. I mean, I think what I'm hearing is we're going to have to
solve this whole speed of light thing.
All right. So I think that leaves us with Dr.
Ching. What are your thoughts? All right. Two things. So one thing just
being what Samir said,
storage is going to be hot again.
And that's a pun intended across the board.
I worked as a PhD in high-performance computing on storage.
Very unsexy area.
Nobody cared what we're doing, to Samir's point.
But definitely the biggest bottleneck in the system
is moving data around.
It's just love us by orders of magnitude.
And so I totally agree with that.
The other hot take I have is that today we're all using frontier models
for the most part for our inference.
I think that's going to change very, very fast in the next two years
where a lot of our inference is going to be done on open source models
because they're going to get cheaper, faster,
and they're going to answer more questions
that the frontier models are just not allowed to answer
for regulatory or otherwise reasons or business reasons.
And that's going to be a major shift in the next couple of years
as computer hardware gets better as well.
Fascinating.
Okay, so there's a lot that was just said there
that I wish I could dive right in.
But in the interest of keeping us on schedule, So there's a lot that was just said there that I wish I could dive right in.
But in the interest of keeping us on schedule, I think I'm going to kick it back over to Pranav now.
And I think we'd like to talk maybe just a little bit more about the origin story of Shelby.
Yeah, I think the origin story is kind of, you know, meeting of minds, so to speak.
The listeners might remember that, you know, Aptos' origins are back in the DM Libra project,
where a bunch of us previously have built Facebook infrastructure for billions of users,
exabyte scale data.
And Jump Team, Jump Crypto has also built a lot of infrastructure for their kind of high-frequency trading platforms. And sometime in 2025, we are just brainstorming and looking at the state of the decentralized systems.
And kind of we saw trade-offs everywhere, significant trade-offs everywhere.
And kind of we saw trade-offs everywhere, significant trade-offs everywhere.
If you want to do anything significant or there was just an accepted kind of status quo that if anything is decentralized,
it's kind of going to be slower, it's going to be more expensive, it's going to be more friction,
there's going to be more friction to use.
And we kind of rejected that premise and said, no, it doesn't have to be that
way. And then we also saw that anything meaningful that we were doing in our enterprises were all
driven by high performance data. And looking at the enterprise, I mean, looking at the
decentralized space, that layer was completely missing. And we're like, why don't we build that out?
That's kind of the origin story I remember.
But Samir, what do you remember?
Yeah, I think part of this has been to jump crypto has been building in the crypto space for a little while.
One of the projects that kind of this has been inspired by is Firedancer.
And what that is, is taking technology that has been built as part of the trading platform
and figuring out how to use it in Web3.
And we were realizing that there's an awful lot of stuff that we have built at Jump that is really interesting and very high performance.
And there are ways and it has strong relevance in a lot of what is happening today, especially in the AI space.
But then working with Aptos, it's really about the experience of the engineering teams.
We know what the Jump teams are good at.
We're good at large data.
We're good at fast systems.
And we know that Aptos is really good at infrastructure,
like global consensus infrastructure,
building systems out in public
and having these things face the real world.
And it really made sense for us to work together here.
And it fundamentally came down to us having this idea and then realizing pretty much simultaneously,
like this was something different and something that was actually a really interesting product
and that there's no real viable competitor to. Like this doesn't exist right now. We're solving
a real problem that does not have a solution. And there have been a lot of attempts in this space, and they've all been less than ideal for one reason or the other.
And we think this one really meets the standard that we, as in Jump itself, would use as a product once it exists.
Yeah, this is really interesting.
Avery, I think I want to hear from you a little bit.
You know, we've heard from Samir, you know, kind of obvious why, you know, Jump Crypto would want to get involved in something like Shelby.
But walk us through, you know, from the Aptos Labs side, like, you know, what did you see that made you feel like, you know, this was the right bet to take?
that made you feel like, you know, this was the right bet to take?
And why now?
And why now?
So as we started to work with blockchain, it became pretty apparent
that you can only store so much data on blockchain,
even though AppSys is extremely cost effective and efficient.
You know, there are pretty significant
orgs of magnitude that don't make sense for certain sizes of data.
Namely, you know, when you go to like megabytes, gigabytes,
ten terabytes, exabytes,
so on and so forth.
And it's pretty obvious that there are still strong use cases for why you want verifiable
data, why you want data that's going to be traceable and also have lineage behind it
for different kinds of Oracle purposes.
Imagine large Oracle's instead of blockchain based oracles on pricing only or supporting these AI use cases where a lot of the revenue share is going to be done on chain and with future models.
So it became pretty clear that this is going to be a use case.
And then working with our partners at Jump, we saw the same vision, I think, and obviously two teams that have great respect for each other and also delivered incredibly fast and getting to test net today.
teams that have great respect for each other and also delivered incredibly fast and getting
to test net today. I think it's a pretty natural fit to think about how do we kind of grow
the use cases for the centralization, how you build it on a core infrastructure like
Aptos and how now you can get like sub-second reads on a system that's coordinated across
a blockchain, which is just amazing.
Yeah, absolutely.
Yeah, I think there's this kind of, you know,
this match made in heaven, right?
These two early collaborators in, you know,
such an innovative problem space.
So, yeah, really glad that, you know,
these two entities got together
and are contributing on something, you know,
much, much bigger than all of us, right?
So, you know, I think I want to pivot just a little bit and Austin, maybe we can hear from you.
You know, I want you to talk us through, you know, from like the, from the double zero perspective,
like what were your observations, you know, more at that kind of fundamental, you know,
the network layer that others were not seeing. And maybe you could just kind of talk us through,
you know, what you're seeing,
like what's the bet that you're making
in terms of this purpose-built fiber?
Yeah, you know, we sort of have seen this thing
where a huge number of distributed
and decentralized systems have spun up
with a fundamental trade-off,
which is that, hey, we give you decentralization,
we give you control of your data, which is that, hey, we give you decentralization, we give you control
of your data, control the execution environment, but in exchange, everything's going to be kind of
slow. And, you know, it's this trade-off that we've sort of felt was inevitable for most of
the history of blockchain, is that, you know, Bitcoin, obviously incredible, but quite slow
relative to other types of payment systems.
And this even scales out when you're looking at execution layer blockchains as well,
like Solana, Aptos, lots of fast networks out there.
They're quite slow in comparison to even the matching engine on Coinbase or Binance or something like that.
And so the vision of 00 really was was to say, a lot of the
reason for this is not because of some fundamental system architecture problem. It's because what
we're dealing with is systems operating over the public internet. And whether they're Web2 giants
or AI companies of today, you know, or their financial exchanges, nothing of note really
operates over the public internet.
You may think of Facebook as a large public internet company and sure, to some extent
that's true, but they're also the largest builders of private fiber and private networks
of anyone in the world because they need to distribute data at global scale.
And if your Instagram reels require dedicated high-performance infrastructure, I think it's
pretty crazy that we're trying to build the world's financial systems on something that
isn't even suitable for serving Instagram reels.
And so that's kind of really the major unlock here.
And so when the folks at Shelby sort of were in the early stages in getting this idea together
and saying, look, we're trying to build decentralized hot
storage that will rival the performance and cost of centralized hot storage. You know,
the fundamental question from the folks that I was talking with was like, that's awesome.
What does the networking layer look like? And it was pretty, pretty clean, you know,
pretty clear early on that there was going to be a lot of work we could do together where 00 can provide this fast base layer of connectivity and performance for delivering
Shelby data to customers, as well as for helping with the verification and all the checksum data
that the network has to do to make sure that the storage actually is there and is still maintained.
has to do to make sure that the storage actually is there and is still maintained.
And, you know, from our perspective, our expertise is certainly not in building data storage
systems by any means.
It is simply in moving data around the world and that transmission layer.
And so, you know, there's a very, very natural fit here between the two projects.
Yes, I'm starting to see it all now.
You know, it's all kind of forming in my mind's eye.
Okay, so we've talked a little bit about, you know, kind of the origin story,
you know, what brings these three parties together for this grand adventure.
So we understand now, right, like we've talked about how the problem is real.
You know, the timing is right. We've got a product live now.
And so let's talk through a little bit more about what actually changes if this gets built properly.
I think I'd like to go right back to you, Asit, if you don't mind.
You know, as you've discussed, right, we're not just optimizing existing routes here.
We're building entirely new ones.
And so let's talk a little bit about, you know, like, what is insufficient? Like,
what breaks today for AI? And why can't it just be fixed in software?
Yeah, certainly. So I think there's a there's sort of a great example here. If you look at
BitTorrent. So BitTorrent is probably the world's most successful peer-to-peer protocol that runs on the internet today.
And BitTorrent, you know,
is primarily used to distribute copies of Linux,
as we all know and love.
But, you know, one of the things about BitTorrent
is it's really great at distributing data
if timeliness is not important and if the order the data arrives
isn't important. And that's kind of why it's used by a bunch of open source projects, because they
don't have to pay for bandwidth in the same way. But then you compare that to, you know, if you
were trying to use something like BitTorrent to watch Netflix or YouTube, it would be a pretty
bad experience, because the entire way
the protocol is designed is not about timeliness and responsiveness. And this is kind of the story
of most data distribution networks today is that they run on private dedicated connectivity.
Cloudflare runs on private connectivity. Amazon and AWS runs on private connectivity.
And so I think the really, the vision, the mission here is kind
of the holy grail, which is like, what if we could start breaking apart and disintermediating
the stack that the cloud providers have provided, which is incredibly useful, but offers, you know,
comes with a huge amount of lock-in as well. And so, you know, if Shelby is able to build this storage system that truly
competes with what can be provided by centralized storage providers, and the 00 network is able to
provide global fast connectivity, that means the delivery of that data is just as fast,
if not faster than you get from any of these cloud providers. That's a market that that's a recipe
for breaking open in the market the same way that you know, what happened at the end of the early
90s is of course, in this case, like the DOJ stepped in and broke up Microsoft. But without
that breakup, it's very unlikely you would have seen the huge amount of innovation that came in
the early 2000s that led to Google and Facebook and all
these other companies. And, you know, we're obviously not trying to legally break the cloud
providers, but if the economic monopoly that many of these companies have ends up not even going
away, just being softened just a little bit, it can break out innovation for entirely new types
of concepts of products, which is exactly what blockchain is
all about. This is a way of fundamentally building projects and products from a different set of
first principles. And if that base layer infrastructure to live up to the sort of
philosophical visions of that still really isn't here today at the level we need it to be. And so
I think storage is one major component. Connectivity
is one major component. Compute is obviously the third leg of the triad. And so this is just,
you know, a part of that much larger journey. Yeah, wow. I'm learning so much personally
today. This is really fascinating. So, okay. So we've talked a little bit about the underlying need for
complete fundamental change. I guess I think
maybe Pranav, you can talk to us a little bit more about
going back to Shelby and talking through what does Shelby
unlock for AI team building a product today
that they, let's talk through, like,
what's possible with Shelby that cannot be done in current infrastructure?
Yeah, and I'll maybe first step back, build up on what Austin said. And if you just look
at first, let's just look at the technical view first. Like, there is a, you know, we
have a global kind of dedicated bandwidth
and dedicated connectivity,
which would move a ton of data around.
Sameer talked about the high performance
kind of storage stack that Jump has built,
and that storage stack can pretty much bring your data
from disk to network and kind of saturate
all available network bandwidth.
We have a decentralized coordination layer
which can globally coordinate data movement
from point A to point B in kind of almost real time.
So when we put it all together, that's a solution space.
Like, but what's the problem space?
What's the demand side of things?
And this is a very kind of,
this is Shelby's is like at the right place
at the right time.
Product and not just for decentralized world, but just for overall AI workloads anywhere.
The emergent phenomenon is that GPUs and AI focused compute capacity, be it ASICs or FPGAs or GPUs or others,
they are coming together
wherever there is power is available abundant,
if power is available cheap,
there is political will,
there is data center build out capacity available,
and there are massive gigawatt capacity data centers
are being popped up all over the world.
If you look at enterprises,
all of their operational data still exists
in different cloud silos or their own data centers
where they are operating and building.
And that all the cloud silos or regional data
used to be enough because you were able to find
all of your compute capacity also
in that particular region
and everything worked beautifully so far.
Now, the new demand is that
I have my data in this region, I need to move it to a different part of the world to access
GPU capacity or other AI compute capacity. And this is creating a multi-dimensional problem.
One dimension is to move data faster to the compute capacity. If you're a capacity provider,
you want to fetch data fast, not wait out.
And currently there is this bunch of data copies
moving around that is high egress costs being paid.
There's a lot of friction in managing data.
And now you kind of go back to this,
what Shelby offers technically.
You technically can just put one copy of your data on Shelby
and access it anywhere in the world,
bring it to any compute capacity,
bring it to anywhere where you need to serve out data.
So this kind of global network of data,
global data fabric, so to speak, this does not exist.
And this is the biggest unlock that Shelby is going to offer the global data fabric, so to speak. This does not exist.
And this is the biggest unlock that Shelby is going to offer.
On top of that, you add data verifiability,
where the data came from, who's the owner,
its immutability guarantees, cryptographic proof.
It's just the perfect solution for IP-protected data, user-consented data,
and data that needs to move around to
find compute capacity.
So that's kind of multidimensional unlock that's waiting to happen with Shelby.
Yeah.
And I think, you know, I think you're starting to touch on something that I think maybe we
can have Avery talk us through a little bit more as well.
So, you know, crypto has been talking about verifiable data for years.
And enterprise, you know, in AI, right, has been completely ignoring it.
So what changed and why are both sides now kind of, you know,
pointing at the same problem?
So, I think there are two things. One thing is, of course, getting
the software, the compute and the hardware all together in the same place. That's something that
Shelby's going to excel at. You need to have all three and data movement is the most expensive of
those operations. The other thing is that, you know, crypto protocols have had verification,
but for a long time, the performance hasn't been there.
So taking 10 seconds to load your data and read it,
and, you know, or a hundred seconds in some cases
and paying exorbitant fees,
it's like open cloth today, right?
Like it's a very fun product to try and to play with,
but it's just not very productive for most people.
You need to provide enterprise kind of great product
out there that's gonna support, you know,
incredibly fast reads and writes,
and supports enterprise-grade reliability.
When clouds go down, Shelby does not, because it's got multi-cloud defenses from day one.
And it's got the pricing model that's going to beat hyperscalers on egress.
So these things just didn't exist in the past.
That's why this product is completely new and differentiated. And we've seen so many different use cases in builders that are
thinking about new things to do with an infrastructure that is just fundamentally
different from anything else that exists in the market today. That's right.
So it's been a minute since we've heard from Samir.
So let's hop over to you.
And just to preface this,
you know, jump build systems that operate in microseconds.
And everyone says AI needs better infrastructure.
So I guess my question to you is,
you know, why hasn't the market already solved this?
I think it's one of the things that Austin mentioned is that the hyperscalers have already
solved this.
Like they have built giant networks, they have built giant data centers, and they have
solved this for themselves.
As an individual firm, they have built this whole thing as a vertical,
and you can use it as long as you are completely willing to be locked entirely into their ecosystem.
As anyone who's tried to go and purchase GPUs or get blocks of GPUs experience, it's great,
but it's not cheap. They definitely charge you for everything you use.
But in the more general case, I think it hasn't been solved because of problem of coordination
and investment. And there's multiple layers here. You? There is the, you start with something like, you start with double zero, which is the ability to have a better infrastructure
to move information around.
And then you build Shelby,
which is the ability to actually have data
that moves on that infrastructure.
And only then can you build coordinated compute
across that entire infrastructure.
But until relatively recently,
we've all been building with internet constraints, right?
This is a system for all of its amazing features
that was designed and built in the early 80s
and just isn't capable of carrying volume of information
that we're trying to move around.
So until we spend time like investing and building
and deploying all of this basic core infrastructure
that we can all use,
you can't even begin to make this work.
And I think something I've already said
right at the beginning is like,
this isn't sexy kind of work, right?
This is sort of behind the scenes, back office,
incredibly important,
but it's not the kind of thing that gets,
that drives investment dollars currently until,
but I think as we're seeing now,
a lot of the AI focused firms
are starting to realize like this is a problem
and we are well placed to solve it
like as they realize that they have it you know everyone is bringing up such
interesting i mean really fascinating points and i i wish we could dig more into this like i wish we
had you know two hours i think this is the folks on stage here have so much fascinating insight
that we could really dive deep.
But unfortunately, we've only got under 15 minutes left for today.
So I do have to transition into our last segment
and a couple things that we wanted to discuss.
So I'm going to hand it off to Pranav here. And for context, I want to talk a
little bit about early access. We did have some strong questions that came in ahead of the show
as well. We're going to do a little bit of Q&A. But first, I want to talk a little bit more about
I want to talk a little bit more about early access.
So Pranav, who should be looking to get in on Shelby Early Access?
What are the folks building?
What are the things we want to see people build?
And yeah, talk a little more about early access.
about early access?
First of all, I'm personally blown away
by the interest in early access.
And Akasha, you can share some stats later.
So super, super excited that there is such a warm reception
to Shelby early access.
And I feel, and there are like three categories I've seen
there's so much, so many developers want to build
so many cool applications.
And I'm seeing kind of three buckets
where either a developer is looking for data mobility,
like I have a lot of data,
I want to kind of bring it to GPU spot instance
and just, I'm just starting off,
I don't want to pay too much money.
And so there is data mobility demand,
there is data verifiability use cases,
and then just general economics
are so much better using Shelby.
And so those are the three categories.
The one that I am most excited about is that
as we move towards utilizing the AI models,
not just for knowledge, but for action, which is having the more agentic systems that can take
actions, I want, I see, I clearly see from the developers who are building and what we are seeing is that Shelby is poised to become the data rails for the agentic
systems which are coming on board because there is inherent proof, verifiability, payments integrated.
So agents can easily put their data on Shelby, do the interactions, do the processing and move on
to the next step. So that is one very incredibly exciting area.
And I'm looking forward to more and more use case solving for that interaction model.
So, yeah, this is interesting.
Should I share some statistics now or should we save that for later?
Oh, please go ahead.
Yeah, so I just ran the numbers.
And currently we have just shy, in 12 more registrants,
we'll have 13,000 registered developers on the developer portal.
We've got over 10,000 who have submitted an application.
We've got over 10,000 who have submitted an application.
And I will say we are being very, very selective in the early access, in the early phases of early access.
We want to make sure that the, you know, the experience is, you know, a premium one.
We want to make sure folks have plenty of room, you know, to kind of test things out and actually bring real data and workloads.
And one thing I wanted to mention before we move on, I know there are a lot of folks that
are listening right now who are in our Discord.
There's been a lot of interest around application status and whether you're going to be approved.
I just want to throw this out here.
If you have not been rejected yet, there is still a possibility that you'll be approved.
And if you're in the Discord
and you've been tagging me, looking for reviews of your applications, I just want to say if I've
interacted with you and you have not yet been rejected, that's a good sign. I'm looking for
folks to build something interesting. So I just wanted to throw that out there. And I know I'm
going off script a little bit, but there's been a ton of interest and some folks are worried they're not going to get it in. There's still plenty of time, but we are just looking for folks that are building interesting things that are at least show us an existing history of building stuff on open source and we'll get you in.
Okay, so sorry for that little plug there.
But I want to kick this over to Samir.
Let's talk a little bit more about, from your perspective,
you've been involved in a very early collaborator
in Shelby's development.
What is some of the upcoming features or capabilities
that you're most excited for builders to get their hands on?
So I think this is about these kind of solutions
that have existed before.
I talked about BitTorrent and there's other distributed storage things,
but most of them just haven't really been interestingly viable.
So if you have an idea where you need a couple of extra zeros in terms of performance,
like this is what Shelby is.
Like it's the same idea, but now it's useful.
And this is everything from,
we already had an awful lot of people doing things like distributing video,
but there's distributing IOM models and everything in between.
I actually need to unlock really significant information distribution across whether it's
to people or to compute edges.
And I think a large part of what we've been thinking about is we don't even know what people are going to build. When blockchain first came out, I don't think
anyone had a really good idea of how that would end up. And the idea is we're going
to give you the tools and you can come up with some really exciting ideas with them.
Yeah, now so this touches on something I think maybe Avery you could speak to a little bit more.
You know, just with the thought of like not really knowing what folks are going to build.
You know, thinking around what we've seen so far on Shelby.
So Dr. Ching, what has surprised you the most about, you know, how folks are actually using it versus maybe what you expected?
And what are some of the use cases
that you're most excited about?
So when we kicked off this project
together with JUM last year,
AI had not taken off to the degree it has today, right?
So Claude's revolution started a couple months ago.
And the whole industry has changed since.
I think just going back to Samir's point, we didn't know exactly what
this thing would be useful for, but it became apparent as the compute needs are going to be
mostly guy-driven in the future, especially on inference, it seems like shellfish is a perfect
fit, right? That problem is trying to solve getting data from anywhere in the world to the
closest and cheapest compute infrastructure without paying an exorbitant fee,
and then supporting very reliable infrastructure
that can tolerate cloud downtimes and outages
just turns out to be like a really important
infrastructure piece out here
that I didn't expect honestly from the very beginning.
And now that we're building towards those use cases
and seeing more use cases around not just the training of the inference side, but also even the ability of training data to be used in meaningful ways and having shared access for public data sets that can be have revenue models and royalties associated with them, supporting kind of use cases for creators to kind of re-earn and monetize their existing content.
of use cases for creators to kind of re-earn
and monetize their existing content.
Those things have been really strong interest
for me personally and for, I know, for the team.
And seeing those use cases start to pop up
and big partners coming out and telling us
that they need this work from us
has just been very, very encouraging.
So I couldn't be more excited about, I think,
the use cases that are being explored now
and also about the future use cases as AI grows by a couple trillion dollars in the next couple of years
in terms of hardware costs and spend on the compute side.
And there's going to need to be commensurate infrastructures spend and investment into the storage side as well.
Yes. Yes.
Okay, so this is all...
God, I know I keep saying this,
but I wish we could stay here for hours.
But to round us off before we jump into a little bit of Q&A,
Austin, I want to hand it over to you.
Maybe you can just talk us through from the network side.
What is a team on Shelby? What are they going to feel? to hand it over to you. Maybe you can just talk us through from the network side. What
is a team on Shelby? What are they going to feel? What's part of their experience that
they wouldn't find anywhere else?
Yeah, I mean, there's so much when it comes to networking. Networking is one of those
things that I think most people who've grown up with modern computer science just treat
as this magical black box of packets leave the system and just magic takes care of the rest of it. And there's a lot
under the hood. The public internet is this incredible resource that we all know and love,
but it's not built for performance. And so if you are dealing with data movement over the public
internet, if you're dealing with it over a cloud provider,
oftentimes what you're getting is just a subpar experience.
And so for normal builders who are sort of building
sort of a, you know, like a lightweight project on Shelby
that requires some data and storage,
and they're using it for a whole bunch of reasons,
you'll just find that you can access data in more locations
and at lower latencies and higher
performance than you may have had before. But I think the real unlock will be for folks who are
like, I need to store a petabyte of data and read and write it extremely quickly and have that all
be incredibly responsive and know exactly like every time I hit this data set from this server, it's 48.32 milliseconds to get a response back with my data back. And that number is consistent 24 seven 365. It's that class of application that's just never been possible to build on blockchain before. And quite frankly, even if you're building it in a cloud provider today, there's so much jitter on those internal networks, you're not getting that type of certainty that comes out of that anyway.
so much jitter on those internal networks, you're not getting that type of certainty
that comes out of that anyway. And so it really is the idea that you can finally build mission
critical systems using decentralized technology that, you know, will work and respond as you
expect them to. So, you know, the unsexy pitch here is it'll be just like you always imagined,
but faster, which, you, which is sort of the
point of networking. You should not have to think about it. It should just do what you need it to do.
That's right. Okay, so let's just dive into Q&A. We're almost pretty much at time. So hopefully,
we can get through a couple of these questions. And for context for everyone in the audience, we asked for questions ahead of the show.
And we've picked a handful of them here.
And so what I'll do is I'll read a question and I'll direct it to who I think might be the right person.
But if someone else wants to jump in, that's fine as well.
And so to start off our Q&A, I've got a question here from Will Fricks 6. And this question reads, if AI agents become the dominant users of the Internet, do we need a completely different data layer? Dr. Ching, why don't you kick us off?
Oh, I love this question. And it's not an if question, it's a when question. And AI agents will become the dominant users of not only the internet, but also the blockchain. And it's going to require a completely new infrastructure across the board, right? And so one thing that we've been doing in Aptos is to support kind of AI agents as the native interface. Agent RHEL is what I call it. I think we're doing the same thing for Shelby, obviously.
is what I call it. I think we're doing the same thing for Shelby, obviously. And so that's
something that's pretty obvious from an interface perspective, but it also requires us to rethink
the entire infrastructure itself. As you have the AI agents interacting, it means that the safest,
strongest, most reliable platform should be the ones that they choose. The ones that can make the
most money on should be the ones they choose. And so it's not just a data layer, but different compute layer, different data layer,
different kind of infrastructure that, you know, we all built for humans in the beginning,
but it's very clear that the users that interface with it are going to be agents, 100%.
I could not agree more. Thank you. Next question reads, well, this is
a question from 0x
Ben GSS.
How does Shelby deliver
sub-second global
reads without crazy
egress costs? Samir,
why don't you take this one?
That's a two-part
solution. The first
thing, I think,
as Austin has already talked about,
is double zero.
You can move data globally
without internet bottlenecks.
I think people are so used
to the performance of the internet,
they don't realize that's just not
how it's supposed to be.
This is why all the hyperscalers
build their own networks.
And we have access to our own version of that with double zero.
So that's how we do it quickly.
How we do it cheaply is actually,
it's another hyperscaler thing.
Icarus is a lot cheaper than hyperscalers
would lead you to believe.
And you actually look at the cost of doing this.
You can go and rent a major hosting provider and just go and see how much that actually costs.
And it's actually pretty reasonable.
But the problem has been coordinating them.
You've got to bring all of this together.
And you coordinate with Aptos,
with the software that runs on it that Jump Crypto is building. And now you get both benefits. You
get high performance data movement at a much lower price. And we actually have the ability to create
a real competitive system that rivals the performance of the hyperscalers themselves.
Exactly. I could not have said it better myself. Thank you. All right, we've got another question
from 0xbengss, bringing the good questions today. This question, I think, Austin, maybe you could answer. In 6 to 12 months, what centralizes always faster assumption will
Shelby break?
You know, let me kind of like start off with the shattering the illusion that what Amazon
or any cloud provider runs is a centralized system.
Now, of course, they're centralized in control, in ownership, in programmability,
but Google is one giant distributed system.
It's also a centralized distributed system, but it is a distributed system.
And so there's been this sort of like ethos for a while of like,
decentralized systems can't be as fast as centralized systems.
But I think really the problem there is we're not using these words correctly.
So distributed systems are the majority of systems today.
The vast majority of the architecture of any Web2 company is a distributed system.
It's just admin under one admin key.
And so if people look around, they say,
oh, well, Google delivers 50 millisecond response times anywhere in the world. No decentralized
project could do that. It's like, well, no, Google is a giant distributed system as well.
And so I think one of the main assumptions that's going to be shattered here is that you have to pay a latency or a performance tax
when you're going with a system
that lacks that centralized control.
100%. 100 emoji.
All right, so I wish we could get through
the list of all of our questions.
I think we maybe have time for one more.
So Pranav, maybe you can take us home with the last question.
This question is from Holly Goner, and it says,
what differentiates Shelby from other modular data infrastructures?
Yeah, and I think probably what people might have
gleaned already from this conversation, like Shelby is an entirely different
category, entirely new category of data infrastructure
that does not exist. So I think in some sense, comparing
to the existing modular storage systems,
decentralized storage systems is almost not fair.
It's like comparing kind of, you know,
freight trains to the next-gen hyperloop system.
And so I think it's just completely different set of use cases,
different performance parameters,
different deployment model,
and just an entirely different category.
That's right.
Okay, I know I said that was the last question. I lied.
I do have one more question. This is my question, but I
feel like I deserve to ask at least one. This is a question for
everyone. We'll go around the room and then we'll wrap up.
But so my question, how do I put this?
I mean, I guess I'm just going to, I'll just come right out and ask.
You know, how does it feel, right?
Like, I don't want to put this.
How does it feel to be?
Okay, I'm just going to say it. How does it feel to be so awesome, Pranav?
Oh, so awesome because I'm talking to you, Akasha.
Being on the same stage as you, it always feels amazing.
Far too kind.
It's super exciting. I think just a stage we are in, which shall be development and what can happen and the potential opportunities that unlocks is very exciting to watch.
Yeah. And Austin, how about you? How does it feel to be so awesome?
I mean, look, the thing that makes 00 awesome is all of the contributors that power the actual network.
We at the foundation
just build a software system that meshes it all together. The real power of all of this system
are the 15 plus independent contributors to the network. So yeah, I think that's, you know,
how does it feel to be so awesome? I don't know. You should ask someone that's actually contributing to the network. I just
talk on Twitter spaces.
Yes, so humble.
Sorry, I'm going to get in trouble if I
go with that joke. Let's see. Samir,
you probably already know what I'm going to say here,
but how does it feel to be so awesome?
Oh, I'm going to say I think it's the jump crypto,
jump trading technology that's doing the awesome bit here
and all the people that are contributing to Shelby.
But we're just trying to let people experience the kind of technology that we get to experience
every day.
And we use it all the time for our own systems and just realizing that this is just the way
it should be.
And the awesomeness here is seeing any given thing
that you might experience currently in the crypto space,
the way we do it at any major enterprise
is you put about three or four more zeros
onto the end of the performance and latency budgets,
and that's the normal business use case.
So we want to bring that awesome to you.
Oh, I love it.
Okay.
So last but not least, Dr. Avery Ching,
how does it feel to be so awesome?
I think what really feels awesome, honestly,
is seeing super strong and talented teams coming together
to build something that, you know,
to solve problems in the space that haven't been solved.
And that, you know, having Jump and Double Zero
and Shelby come together in AppTest, come together in a way that, you know,
like these strong collaborators and tech teams can build things that I think very few,
if anyone in the industry can build.
And that feeling of just collaboration and solving hard problems, you know, gets me super excited to go to work every single day.
Yeah, could not agree more.
Perhaps the real Shelby is the friends we made along the way.
All right, so I think that's pretty much all the time we have for today. We are just a little bit over, but I think we did pretty good, honestly. There was a lot. There's a lot to get into chance to get to everyone's question. I saw several requests for folks
to come up. I'm sorry we couldn't get to you today, but please stay tuned
for more spaces in the future.
Just to close this all up,
as you've probably heard, in case you haven't,
Shelby Early Access is live today.
Link is in the description.
If anything that you heard today resonates with you,
please go sign up right now.
You can go to developers.shelby.xyz.
You can join us in the Discord.
That's discord.gg slash shelbyserves.
I want to thank everyone, our guests here,
Pranav, Austin, Samir, Dr. Ching.
Thank you everyone for joining us. We are going to sign out for today. And until next time.
I was trying to end it with music, but I can't. I'm sorry, guys. We're just going to have to end on awkward silence.
Sounds great.
Thanks.
Cheers, y'all. Thank you.
