Why You Can't Build Sovereign AI Alone/David Keane, Akash Agarwal, Abhi Ingle - SCX.ai, SambaNova, Australia ~ Spark of Ages Ep 50 - Spark of Ages

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
David Keane (00:00):
What we've seen is that for many countries, the
idea that we simply use AI givento us by some Californian
company or of course by aChinese business may not be what
the country wants.
What it means is that they'vebuilt the technology in such a
way, one, it's pre-assembled ina rack.

(00:22):
I mean, we've tried to assemblean NVIDIA rack and it's
requires a Harvard MBA, let meput it this way, and an MIT
degree with it to get that done.

Rajiv Parikh (00:30):
Thinking more of the MIT degree.
Because usually we just readabout it and talk about it.

And to build a financial model,

do we really build financial models?

Abhi Ingle (00:37):
My job is to make these gentlemen successful.
Okay, if I do a good job withthat, they're gonna come back
and buy more systems from me.
It's not about me putting mybrand front and center.

Welcome to the Spark of Ages podcast.
In this episode, we're on thecritical mission of how to build
sovereign AI.
That's right, sovereign AI.
And we're speaking with thefounders of Southern Cross AI,
or SCX.ai.
We're gonna call it SCX fromnow on, who are building their
company in Australia to securethat nation's data future with

(01:15):
help from a friend of our show.
This conversation is about thefoundational architecture that
defines the AI-powered economy.
So here's my guest.
I got David Keane.
David served as the founder andCEO of Southern Cross AI at
Inference as a service platformdedicated to establishing
sovereign, scalable, andcost-efficient AI infrastructure

(01:37):
tailored for Australianrequirements.
Previously, David led BigTinCanas CEO managing director for
over 14 years.
So a fellow founder who stuckwith it, David is an alumnus of
Macquarrie Graduate SchoolMacquarie?
We'll get there.

We'll work on it.
I'm gonna get you an Australianaccent, Rajiv, before we finish
this.

Macquarrie?
I thought I got it.
Macquarrie.

Macquarrie.

Macquarrie Graduate School of Management.
I listened to it on YouTube andI guess I didn't do it, right?
Okay.
Akash Agrawal.
Akash serves as the chiefstrategy officer and co-founder
of Southern Cross AI.
He's been on the show before.
He founded and led AI andvarious AI and security
companies.
He previously was SVP at SAP,where he led the IoT application

(02:20):
security business.
Akash holds an MBA from HarvardBusiness School.
Then there's my great friendAbhi Ingle, who I've known for
over 20 years.
Abhi serves as the chiefproduct and strategy officer at
SambaNova Systems.
Before joining SambaNova, hewas a member of the executive
team at Qualtrics.
Prior to Qualtrics, Abhi spentover 17 years at AT&T serving as

(02:42):
senior vice president andbusiness leader of multiple
multi-billion dollar businesses.
Abhi also holds an MBA fromHarvard Business School.
So, gentlemen, welcome to theSpark of Ages.

Abhi Ingle (02:53):
Thank you.
Thank you for having us.
Thank you.

Well, we're really excited to have you.
We're going to talk a lot aboutthis really interesting
concept.
Everybody knows AI is what youget with ChatGPT or Claude or
they talk about various AIinfrastructure vendors.
But David, Akash, could youexplain what Southern Cross AI
or SCX is about and how it fitsinto the AI ecosystem?

(03:17):
Why'd you start this firm?

What a great way to start.
So, yes, first of all, I'm theone with the strangest accent on
this call.
So you're going to have to bearwith that.
That's going to be a challengefor the viewers out there to see
how they can understand thiscrazy Australian accent.
But I think, Rajiv, it's suchan interesting time in
technology overall.
You know, it doesn't take arocket scientist to understand
the impact that AI is going tohave on our society and of

(03:41):
course our economy, but on thelives of every human being.
And in fact, many people wouldsay that we're already halfway
there.
If you think about how peopleuse AI today as everything from
their tool they use to help themdraft an email or a note to
their boss looking for a payrise, right through to many
people, particularly youngpeople, using an AI as almost
their personal advisor to guidethem down some of life's biggest

(04:04):
decisions.
You know, AI is already havinga big impact on the lives of
many of us.
But when you think about thatdeeply, what you start to
realize, of course, is thatwhilst the world is amazing and
many of us have so many thingsin common, there are also
differences in the world.
Countries and cultures andenvironments can be different.
And what we've seen is that formany countries, the idea that

(04:28):
we simply use AI given to us bysome Californian company or, of
course, by a Chinese businessmay not be what the country
wants.
It might be if AI is going tohave such an impact on the lives
of everybody.
And I'll just speak here aboutAustralia for a minute, every
Australian.
You know, we want to have somecontrol in that.
We want to know what'shappening.
We want to know what happens toour information.

(04:51):
Is it processed in a particularway?
Is it used in a particular way?
And we want to have someinfluence over that.
So SCX was launched with thegoal in mind to build
Australia's first sovereign AIcloud.
And that means that we'redelivering the outputs from what
people now term an AI factory.
It was a phrase I've I've hearda lot now.

(05:11):
But basically a business thatbuilds the capacity to then
create the AI that is used byevery citizen for business, for
government use.
And of course, amazing nextgeneration AI builders that are
creating the future throughtheir tools.
But do it in a way where itfits the Australian context, the

(05:32):
Australian culture, theAustralian version of English,
and does that in a way thatworks for all.
So, you know, we have seen thistrend impact a lot of
organizations and countries, andwe're just so pleased to be
able to bring that to Australiawith the launch of scx.ai.
And Akash, what's, you know,you've we've talked a lot about
this.
Akash is someone who bringstremendous value to the team and

(05:54):
I think brings that SiliconValley native experience as
well.
And, you know, whilst it's easysitting in Silicon Valley to
think about and understand whatAI does, but you know, when
Akash was with me in Australiajust recently, you realize that
not everyone in the world hasthat depth of knowledge yet,
right, Akash?

Akash Agarwal (06:09):
Yeah, that's correct.
So I think there's, you know,as David pointed out, with SCX,
there's obviously one of thebiggest differentiators is that
we're partnering with companylike Samba Nova.
So we've gone all in and we'reusing next generation technology
to bring what we call inferenceas a service.
So, you know, layering on topof the sovereign capabilities
that David defined, you know,part of being sovereign is being

(06:33):
in the country, making sure thedata doesn't leave the
country's perimeter.
You know, we're also makingsure that we're bringing kind of
a disruptive technology.
We believe that ASICstechnology that Samba Nova is
based on is fundamental toinferencing.
And, you know, we are aninferencing cloud.
We're inferencing as a service.
So we don't stop in Australia.

(06:53):
We are providing inferencingacross the world.
So, you know, if you have anapplication and want to use
inferencing, you can leveragethe SEX service to do that.

So, like David, you talked about the notion of a
sovereign data cloud.
Is it kind of like whathappened with GDPR, right?
The European standard, whereyou want to have your data in
the country?
Or is it, you know, because theway AI works so closely with
data, you want it super close tothe customer and therefore you
want to put it in Australia?

(07:24):
Or is it even as Akash talksabout it, is this is one
location that's doing lower costinference as a service,
offering it internationally?
Just help me understand it.

Yeah, look, I think it's really insightful that
question you ask.
So of course, we saw thishappen in the history of
computing many times when thingsoften started in one country,
particularly here in NorthAmerica, and then things get
distributed around the world.
Often it was for latencyissues, it was for cost issues
in terms of backhaul.
There were a lot of reasons whytechnology got distributed.

(07:54):
And I'm sure Arby can give ussome really great examples of
how ATT solved that.
But, you know, like there werethose traditional challenges
that you weren't distributed forthose reasons.
I think what's happening hereis something that's slightly
different.
Those things still apply.
But we're talking here aboutintelligence, guys.
Intelligence.
Every time I kind of say that,I think, well, what am I talking
about?
Intelligence.
It's not intelligence.

(08:14):
But for many businesses andorganizations, it will be.
And the idea that you can justoutsource your intelligence
overseas is a challenge for somecountries, right?
First thing.
Having said that, our goal atSCX is to apply what I like to
call unit economics.
I love the phrase tokenomics,and I'm sure Abby can talk to us
about that.
But the the concept of look, ifyou can make this stuff in a

(08:39):
way that delivers outputs withthe performance and price that
really works, you can sell itanywhere.
So I think the answer is nuancefor a Jeep.
It's like, yes, sovereign kindof means sure where it is and
what it is, what is thisintelligence?
How is it being done?
And we'll talk about it today.
But we also launched, as partof our program, was a unique

(09:01):
version of one of the models,which we call Project Magpie.
And we can challenge those onthe podcast to send you a
comment on Twitter, Rajiv, ifthey know what a magpie is.
That's going to be a test.
That's a test.
It's a test.
I'm not going to say what itis.

We can even make that as part of our game.
We can add it to it.

Oh, we could.
We could.
I'm not going to give it away.

And no, you're not allowed to use AI to answer
that question.

Uh-huh.
It's right.

You'll get the answer very quickly.
Yeah.

So you talk about Project Magpie.
And go ahead and give us aquick brief because I have a
whole bunch of questions foryou.

Oh, yeah.
So this is Australia's first,and I think one of the first in
the world, of fine-tuned largelanguage models where we've
actually fine-tuned thereasoning side.
So those of you, again, who goand use one of these AI models,
sometimes you'll see it likethinking.
People think of it as the innermonologue of an AI talking to
itself.
The term reasoning is the termthat is often used in the

(09:52):
industry to describe what it'sdoing, but it's processing some
stuff before it goes and givesan answer.
And I quite like the term innermonologue.
It's like getting itself readyto answer in a particular way.
So we've gone and fine-tunedthe reasoning layer on a large
language model to think like anAustralian.

How would you think like an Australian?
Does that mean like when Ithink of a beach, I should go to
Bondi Beach?
When I think of mountains, Ishould go to the Blue Mountains.

Is that I was going to say it involves beer but and
cricket.
Beer and cricket, I was goingto say.
But no, look, I think there's alot to what it means to think
like an Australian.
We have to confront theseissues because Australia,
particularly, is a very diverse,very multicultural country with
people from all over the worldthere.
But there are a few things thatmake Australia Australia.
The way people engage with eachother, their approaches to life

(10:36):
are slightly different.
And yes, part of it isknowledge, so they know what
Bondi Beach is, but part of ittoo is the way the culture and
knowledge fit together.
So our view was we've got atremendous data set.
Fine-tune the reasoning.
Don't focus on the answer.
The answer will take care ofitself if you fine-tune the
reasoning.
And we're really pleased, andthose out there on the podcast
can visit chat.scx.ai.

(10:57):
They can give Project Magpie ago, and they can see what it is
like to think like anAustralian.
And we think that's thebeginning of it because people
want to feel that their AIunderstands them.
They want to feel that way.

I can see it from the consumer context.
But you said a lot of this isbusiness and government as well.
So what would those examplesbe?

Fine-tuned on the legal system, fine-tuned on the
way the government works,fine-tuned on a tremendous
amount of information that isappropriate for how Australians
should do, not should, but do,do act.
And I think that givescompanies and governments the
ability to implement AI fasterwith more confidence.
And I think confidence is a bigthing for some of these

(11:38):
organizations that will helpthem to do it faster and better.
So Magpie is proof that we cantake data, we can fine-tune an
outcome, and then we can run iton the high-performance
inferencing machines that uh weget from San Vinova.

And let's talk about that.
So this is a great answer.
This is really helpful tounderstand.
So as you talked about, there'sa national need, there's an
in-country need in Australia inspecific, and then again around
the world.
So you're filling a criticalgap in sovereign data
infrastructure, right, inAustralia.
So what led your choice topartner with Sanbanova systems

(12:13):
for the deployment of their lowpower, high throughput, ASIC
architecture hardware, as wellas their managed cloud platform?

It's a big decision because when you're building a
business, you have to havetypically something as fresh as
this idea.
You have to build somethingthat's built to last.
You have to have the partnersin place that can scale with
you.
But you also have to have thefundamental business plan that
works.
And anybody out listening tothis podcast who's built a
business, you know how importantthat is.
If you don't have a businessplan that works, eventually it's

(12:39):
going to catch up with you.
The hype alone is not going tosave you.
You've got to build businessesthat make money.
And so I can share with withthe with everyone, we we spent a
significant amount of timetalking to all of the folks that
build the chips that createthese AI outputs from big to
small.
We built what I believe are themost comprehensive financial

(13:00):
models that talk about the earlyphases and the later phases of
a business like this.
And we looked at that with aview to making money, Rajib,
which is really interestingbecause I think a lot of the
time people haven't been asfocused on making money.
They've been more focused onburning cash and growth, right?
Yeah.
That's good.
I'm not saying that's bad.
We intend to grow.
We intend to grow veryaggressively.

(13:20):
But you have to do it in a waythat at least shows you can make
money.
And we put a lot of effort intobuilding what I see as the most
comprehensive business plan inthe in the world for doing this.
And we looked at all theproviders.
We just saw clearly in ouranalysis that Samba Nova was the
partner that could deliver forus now, but also could continue

(13:43):
to provide that benefit as wecontinue to scale the
inferencing side of ourbusiness.
And you know, that that wasclear to us.

So let me ask Abi, I mean, Abhi, I know you've
been working closely with Akashand David.
You you guys have known eachother for quite some time.
So it's even helpful that priorto this, you guys have known
each other, right?
And then you came into this,and you're leading key parts of
a top-notch firm in the area ofAI hardware as well as software
capabilities.
So maybe talk about two things.

(14:09):
One is you've done work insovereign data AI
infrastructures, and why youthink the Soviet Nova platform
is really viable for this kindof capability.

Abhi Ingle (14:19):
I'll talk a little bit about sovereign AI and why
it's viable and then talk aboutour partnership with SCX as
well.
So, first, you know, buildingon what David said and what
Akash was saying earlier,sovereign AI is all about
reclaiming control andestablishing context.
And that's really what Davidwas really telling you.
It's not just about thecontrol, it's about context.
And you know, every time David,you said think like an

(14:40):
Australian, that 80s song, thinklike an Egyptian, kind of went
through my head, you know.
But I try to I try to tune thatout.
But it's really control aboutthe data, the models, and the
systems that power them, andmore importantly, the people who
operate them.
Okay, that is all veryimportant.
And we're seeing this shifttowards sovereign
infrastructure, where securityis an afterthought, it's the

(15:00):
foundation, but it's not aboutjust security, which is sort of
where GDPR sits.
It's about context, it's aboutcontrol, it's about who operates
it.
Because as David said, nocountry would ever outsource its
entire power needs to somebodyelse.
You would not actually rely onthe mobile network of another
country.
That's where intelligence isat.
You cannot have it somewhereelse.

(15:22):
And that's why we're superexcited to work with David.
And I'll tell you, working withSCX has made us actually
stronger because I'll tell youthe models that David has built
and the scrutiny he put us to isprobably the best workout that
we've had for a while.
And one of the things thatworks for them is unlike certain
portions and certain countriesin which power is at abundance,

(15:44):
space at abundance, you know,perhaps the issues around global
warming are not perhaps thatintense, right?
And money is in abundance, saythe United States, for example,
other markets do care about theenergy efficiency deeply.
They deeply care about the factthat even if it's they don't
create the model, the opensource models they're bringing,

(16:05):
they have actually tuned the wayDavid is doing with Project
Magpy.
And we're partnering, veryproud to partner with him on
that front.
And then lastly, the people whooperate it, why did he do it?
Because he understands thecontext.
He is talking to the banks onthe ground.
We can supply the energyinfrastructure, we can supply
the energy efficient shifts, wecan apply the blindingly fast
inference, we can provide theracks that allow him to shuffle

(16:27):
models in a much smallerfootprint, and we can fit into
his existing data centers or inplaces they have no data
centers, which is uniquely us.
But at the end of the day,without the local understanding
expertise that David brings,there is no win here.
That's one of the reasons why Ithink sovereign AI is really a
very strong partnership.
Without David powering it,without David's brand on it, in

(16:48):
a Southern Cross brand on it.
The word Samba Nova might meansomething in Paul Water in the
United States.
In Australia, it means nothing.
And that's why we rely on themto take us to market.

That's amazing.
So you're talking about theneed for sovereign data or
sovereign AI.
You have the need for lowercost tokens per kilowatt.
Maybe you want to just explainwhat that means, tokens per
kilowatt?

Abhi Ingle (17:08):
So basically, that is a fundamental element of what
Dave was talking about earlier.
How does one make money?
So if you think about it, allintelligence in this case is how
efficiently can you convertelectrons in to a such an
intelligent out.
And intelligence in thissituation is measured in a form
called tokens.
Think of tokens roughlyequivalent to words, like
two-thirds of a word.

(17:29):
Okay?
They're gonna think roughlythat way.
The more efficiently you cangenerate them and the faster you
can generate them, the moreresponsive the story is, the
more responsive the responses,and how much you can actually
get out of it.
And that's what David wasworking on.

Yeah, and can you go in a little bit about the
difference between training,which a lot of people do on
NVIDIA-based architectures,versus the focus that you have
at Salmanova, and David andAkash were both talking about
the notion of inferencing as aservice.

Abhi Ingle (17:56):
So training is also a spectrum.
I talked about pre-training,which is when you build a
certain foundational model.
So ChatGPT, for example, whereGPT is a foundational model,
anthropic is a foundationalmodel, Deep Seek, as you know,
the big Chinese model, is afoundational model.
When you first create themodel, the initial training that
happens requires hundreds ofthousands of AI machines lashed

(18:19):
together.
That's when you actually createthe model that allows you to be
able to use in the future.
The training is training.
When it's actually used, that'sinferencing.
So when you type in a sentence,it completes, or you type in a
query and it completes it.
Or David puts it into GPT OSS,which is the open source model
released by OpenAI, which he hasnow trained to think like an

(18:40):
Australian.
That is the use of inferencing,and that can work much more
efficiently, anywhere betweentwo to five times more energy
efficient, and anywhere betweenfour to nine times faster on
Sabinova systems.
But to be clear, we complementthe Nvidia infrastructure which
is required for training as wellas for proprietary models.

I was going to say, Arby, it's a great example.
I was going to add to that.
So this project Magpie wasfine-tuned on NVIDIA GPUs with
support of NVIDIA, but run forinferencing on SAMO chips.
So I'm sure the folks out therewill understand that split.
It's a really interestingsplit.
If you get it right, that canbe one of the ways they can
power your business.

You can really nail the cost equation better,
right?
And so, Akash, maybe you canhelp me with this too.
Like enterprise AI often fails,right?
They're only 5% of custompilots reach production.
And it's typically due to toolslacking memory or failing to
integrate deeply into existingworkflows.
So, how is SEX leveraging itsvertical integration, deep
domain expertise to overcomethis critical

(19:40):
pilot-to-production chasm forAustralian enterprises?

A lot of it is based on cost as well.
So people run these things, andwhen they realize that, you
know, they've got to extend thesame use for the entire
enterprise and the cost willbecome cost prohibitive.
You know, some of these thingsare put on brakes.
Another reason why they're puton brakes is precisely this
whole area of sovereign.
You know, people don't want theintelligence.

(20:05):
You know, you ask the company,where's the data going?
You know, how are we trainingthis model?
Do we own this model?
So a lot of people right now,you know, the early adopters,
and this is what I wanted totalk about, have been sort of
primarily Silicon ValleyUS-based companies.
I've got a list of all themajor people that are using
tokens in a major way.
And I've looked at them basedon all the neo clouds out there.

(20:27):
And if you look at one thing incommon, you find most of the
companies are venture-backedbased in Silicon Valley.
So they've understood and doingthat.
The mass adoption of AI has notreally taken place in the
enterprise.
That is why you hear aboutthese big numbers and big data
centers being built, becausepeople are anticipating that
enterprises will start usingthat.

(20:48):
A lot of enterprises are usingit in pilot phases, and they are
afraid of a few things.
They're afraid of cost, they'reafraid of data sovereignty,
they're afraid of, you know,making sure that when they are
training these models, that theknowledge of their company and
their business is not going andbeing utilized by these models
for the benefit of othercompanies.
So that's where I think the thedebate is out there about

(21:12):
whether you should use aproprietary model or you should
use an open source model thatyou fork and you train and to
your own specific need, justlike we did for you know, using
OSS to create a project MagPi.
You know, companies can dothat.
They can create their ownversion of MagPies.
And so JP Morgan can have amodel that's for general use

(21:33):
within JP Morgan's employees forgeneral intelligence, that's
you know, very much investmentbanking centric, they're
customer-centric with all ofthose kinds of vernaculars.

What you're saying here is like I can go with a
closed model, like an open AImodel, right?
And I could run it on one ofthe different architectures.
Like I could run it on Azure,or I think you can now run it on
AWS Core Weave.
If I'm building it for mybusiness, I could also have an
open source one, like a DeepSeek, and I could run it on a
Salmonova, right?
And then there's value to mebecause I can see the open

(22:04):
source.
Is there is there value to mein the training?
Is value to me is the speed andthe execution?
Where's the value?

The value is in threefold.
One, you can train that modelspecifically for your needs.
So you can train the model,fine-tune the model, very
similar to what David did.
He took the open source versionof OpenAI's model.
OpenAI has released a couple oftheir models and made them open
source.
They're not their big models.
And by the way, another thingthat we're learning from
enterprises, using OpenAI, theirbig model.

(22:31):
Most of the things in thatmodel are useless to a company.
So why have that burden?
So, one, you can use these opensource models and think of them
as forking them, sort ofcutting them and saying, I only
want a segment of this model andthen I'll train it with my
data.
And then it will learn thecontext, it will learn my
nuances and provide value to mycustomers or to my enterprise.

(22:53):
So that's the benefit of doingthat.
So that's, I mean, that's abigger debate, whether open
source is going to win orwhether these closed sources
are, and what percentage of themarket each will get.
So look, it's pretty clear thatthey will coexist.
You know, it's potentiallymaybe they'll get 50-50, maybe
they'll be 70% closed and 30%open.
What I've found, at least withenterprises, some of the

(23:14):
enterprises don't know what canbe done with open source models.
So, for example, we're talkingto one customer, they came back
to us and said, We're going touse Mistral.
And I said, Can you give me theuse case and let me show you
that same use case running onone of the open source models?
What's the benefit of that toyour question?
One is you control it.
Second, the cost.

You're not paying the intellectual property
necessarily.

Yeah, well, you're not going back to where
open AI is taking you orMistral's taking you.
They're not taking you to SanBonova.
Let me tell you that right now.
They're taking you to theircloud and to their own
inferencing provider.
And that is fundamentally goingto be a challenge, right?
Potentially, unless they'reable to compete on the cost.
And, you know, one of thethings that Abe pointed out

(23:54):
that's very, very important isone of the things that we are
also benefiting from is that wecan use technologies like Samba
Nova in an existing data center.
And that is very, veryimportant to understand because,
you know, there are a lot ofdata centers today that will get
retrofitted to becomingrelevant for AI.

(24:15):
Some of them will not be ableto be retrofitted because they
require extreme power andcooling if you went with sort of
the NVIDIA type approach.
This is one of the bigdeterminants for a company like
us that, you know, we can getgoing and use, as David likes to
say, a 2018 data center andstart working in there.

(24:37):
And that doesn't mean we'relegacy and we're old school.
What it means is that they'vebuilt the technology in such a
way, one, it's pre-assembled ina rack.
I mean, we've tried to assemblean NVIDIA rack and it's
requires a Harvard MBA, let meput it this way, and an MIT
degree with it to get that done.

Thinking more of the MIT degree.
Because usually we just readabout it and talk about it.

In to build a financial model.
Do we really build financialmodels?
Yeah, and actually, I'd love tohear what everyone's view is on
this.
But what I'm seeing too is twothings.
One is more and more, if you'rebuilding an AI company, you
don't really care.
And Abby can close his ears,but you don't really care what
chip someone's built it on.
You think all the viewers ofthis podcast, I'm sure, are out

(25:18):
there with great ideas to buildthe next amazing AI solution,
right?
And I'm sure they're thinking,I'm going to use one of these AI
models to power it, then I'mgoing to build the workflows and
the structure all around it.
They don't care which chip it'smade on.
They just don't care.
What they care about is is itfast, is it reliable, and is it
cost effective?
They don't care what chip.
Because most of them don't evenknow what chip it's built on,
by the way.
You don't know.

(25:38):
You just need to haveconfidence, the right output
from the right model with theright cost, the right
reliability with the rightperformance.
And I think that's what'scoming in our market.

So basically, everyone's competing at every
level.
There's the chips, there's themodels, there's how it's
delivered, there's the data.
Everyone's competing.
It's like a many-to-manycompetitive model.
It seems like there's onewinner right now when you when
you read about this, one winnerarea, but not necessarily so,
right?

I'm old enough.
I'll I'll put my hand up toremember the early days of the
internet.
And there was a little companyout of the Bay Area called Cisco
Systems that came along andpretty much for a number of
years owned the build-out of theparticularly routers, but
switches and other kinds ofdevices as well.
And they pretty much owned it.
But eventually, the people whowere buying the internet service

(26:28):
don't know and don't carewhether it's Cisco or Juniper or
someone else building thatinfrastructure.
What they want is the output ofit.
And we may see something likethat happening here just faster.
Rather than taking years forthat to happen, it might happen
a lot faster than it happenedback then.
And, you know, I think that'swhat I'm seeing from the

(26:49):
customers we're talking to.
Control, Kash use that we wantcontrol, they want management,
they want security, they wantconfidence, they want unique
outputs that are tuned to theirway of working.
But which chip it is is up tous as the provider to be able to
build.

Let me poke on that a little bit.
So I'm gonna, this is foreveryone.
The pace of AI innovation isaccelerating globally, right?
Like one of the things that Abitaught me about is how Deep
Seek models are achieving highlycompetitive performance with
far fewer resources.
Like it was mind-blowing,especially this chain of
reasoning capability.
That means that time to marketfor new capabilities is a

(27:25):
critical competitivedifferentiator.
So now, if time is the ultimatescarcity in the AI race, where
is your product roadmap focusedto allow Australian AI native
companies to consistently beatglobal competitors?

What a great one.
I want to add even morecomplexity to your question
before I answer it, which isthink about the region around
Australia.
You've got Indonesia with 270million people, you've got the
Philippines, you've got Vietnam,you've got Thailand.
These are developing nationswith super smart people and an
aggressive focus on building thequality of life for those

(28:01):
people.
And for a long time, they'vestruggled competing against
developed nations likeAustralia, fully developed
nations like Australia, becausethey didn't have access to all
of the knowledge and all of theinfrastructure to do that.
I was at a meeting with theAsian Development Bank recently,
and they're already gettingapplications for funding from
developing nations for AIinfrastructure.

(28:22):
Because those nations believethat with AI, they take away, it
narrows that, I won't saynarrows, eliminates the gap that
was holding their citizens backfrom achieving better lives.
And so I think, Raji, for us,we have to recognize that across
our whole region, there iscompetition for these capacities
and the ability to move.

(28:42):
So Australia needs to movefaster.
I think it's got somechallenges.
It hasn't been moving fastenough.
I think there's a saying we usein Australia, we call it the
lucky country, is what they callit inside Australia, because it
has been lucky for the last,you know, hundred years, hundred
plus years.
Natural resources, smallpopulation, beautiful geographic
area, all those lucky things.
But, you know, now the world iscompetitive.

(29:02):
We're taking away some of thosebarriers that have held back
other nations.
So I'm really intrigued.
And I think it's important forAustralia and I think other
countries in the world torecognize that competitive
nature.

Abhi Ingle (29:12):
I'll actually agree with David that the consumer of
the product actually doesn'tcare what ship it is.
I actually agree with him onjust so you're aware.
But as an inference provider,he surely cares.
As Akash said before, if youactually had an infrastructure
provider who could allow you touse an existing data center,
which is one of the points thatAkash focusing on, that's a big

(29:33):
deal.
You're avoiding a huge capexfor upgrading those data
centers.
If you're in an area wherethere's not a data center, you
could put things in a container,that's a big deal.
60% of the operating cost of aninstance system after you spend
the capex is energy.
If somebody told you that achip could provide it two to
five times faster, a consumerlike you might not care.
But David and Akash, who aresharp minded business people,

(29:57):
surely care.
That's where we focus, and weare very aware.
Of our role in the ecosystem.
We're very comfortable withthat role.
Unlike some of our competitorswho put their brand front and
center, my job is to make thesegentlemen successful.
Okay, if I do a good job withthat, they're going to come back
and buy more systems for me.
It's not about me putting mybrand front and center.
The second point about gettingthem to market quickly that

(30:19):
David was talking about,everything Akash said is about
actually getting them to marketquicker.
If you don't have to upgradethe infrastructure, guess what?
I can stand them up.
I do a fully managed servicebecause it gets them to market
in three to four months versus18 to 24 months in the
alternative.

Right.
Abby, you guys are more than aprovider of systems, right?
One of the innovations that youbrought in was to say, look,
let's show people the best usecase of all of our systems is to
actually build a cloud, whichyou can buy from us or you can
buy from a hyperscaler that buysfrom us or white labels, I
don't know, from us.

You know, what we're going to do is what Abhi
is saying is that look, this isnot just a chip.
Think of it as an entiresystem.
So what they're doing as wellis making sure that all of these
open source models areinstantiated on their system.
Then they pass that on to usthrough this cloud.
So that just think of like youand I spinning up a server on
AWS and boom, we're up andrunning.

(31:21):
Anybody in my company can dothat.
It's AI and open source modelsand all this stuff is not quite
there.
It's getting there.
So one of the other things thatworking in the model that Abe
described is that we are goingto be bringing to Samba Nova and
whoever our partner is lots andlots of use cases.
We already have them.
I think David and I were on thephone with some company that

(31:42):
has a video model that basicallyhe agreed to sending us the
tokens to process the inferencefor that.
And, you know, I didn't knowthis, but David told me right
away that this model is notsupported by Samba Nova.
So I asked him why.
And he said it's just a matterof time and bandwidth.
And I said, Well, we should getthis right in front of Samba
Nova.
And, you know, David said he'dalready sent that over.

(32:03):
So I think the other thing iswhat will happen is by having
this model that Samba Nova has,by partnering with kind of
providers like us, they're goingto get stronger as well because
we're going to the customersand they can't possibly go to as
many customers.
And we will benefit from othercustomers of theirs, their other
OEM partners that will bringrequirements.
So we'll be able to support.

Let me push you a little bit further.
So, like the question is abouttime is the ultimate scarcity.
You talked about magpie, right?
Which is the Australianadaptation of the model.
Is that the idea?
Is like you're going to dosomething for Indonesia too?
You're going to do somethingfor Thailand.
You're going to like you'regoing to really nail this
region, right?
Because Australia has what, 30million people?

27, 27.

27 million people.
So now, but now you're hittingregions with hundreds of
millions of people.

There's two answers.
There's two.
It's a really good question.
And I love these challengingquestions because there's two
answers to that.
One is I believe that thosedeveloping countries will need
access to highly efficient,well-priced AI inferencing
tokens.
To talk about Arby mentionedtokens, they need that today.
So I think there's going to betremendous demand coming from
those countries.

(33:08):
And I think we'll get a goodportion of that because we're in
the region.
Secondly, we have anopportunity to work with some of
those countries to help them asthey start to build their own
sovereign services, which theywill do.
It's inevitable.
It's inevitable.
They will do it.
And we have an opportunity tohelp them with intellectual
support and, you know, just helpthem to build that out.
So I think we've gotopportunities in in both those

(33:28):
areas.
But, you know, look, I thinkback to your point about time,
yeah, we've got to help everyoneto make more use of this time.
And the people that areflexible in mindset will get
that.
The interesting question formany of us in the industry is
how long will it take the massmarket to get that flexible
mindset?
When will everybody embracethat idea that they can do more?
I'm intrigued, guys, by yourview on this one, which is, you

(33:51):
know, we get a lot of peopletalking to us.
I've I've had a bunch in thislast week who are saying, geez,
how can you help us to replaceour staff?
And it sounds a bit rough, butI'll just talk about it
honestly.
You know, we want to cut ourhuman resource cost.
Can this AI stuff help me toreplace jobs?
Whether you call that digitallabor or whatever you want to
call it.
My response back is, well,look, that's easy.

(34:12):
Cutting costs, sure.
But what about doing more?
Don't think about it as cuttingcosts.
Think about it as doing more.
What could you now achieve ifeverybody in your organization
was working at that speed youspoke about, Rajiv?
Then what would we do?
So I think this is a questionfor everyone on this panel and
our world is how do we get, howdo we educate people to think

(34:33):
that they can produce thisubiquitous abundance you've
heard that.

I think that's one of the things, right?
Everyone's always concernedabout cutting costs.
That's the easiest thing to do.
I say we're cutting costs.
You heard Benioff talk about,well, I didn't have to hire
3,000 engineers this yearbecause of AI.
Really?
Are you sure?
Because everyone's spendingmore than ever on AI engineers
and in applications.
One example I can give you ismy own company.
It's under a couple of hundredpeople, and my engineering

(34:58):
team's relatively modest.
They're focusing on the agentarchitecture.
And what's amazing is insteadof them building the
applications, the operationsteam is building the
applications.
Because now with AI and agents,I don't have to necessarily go
to a separate team to build it,test it, iterate.

(35:19):
I can actually have my teambuild it, the team that actually
knows every nuance, build theirown applications.

Thoughts?
Quick thoughts before I moveyou to the next thing.
My big thought on that, man,we've been speaking to a bunch
of the venture investingcommunity as well as the private
equity community.
And I can share with you twothings.
They are really concerned.
You talked about somethingthere, Rajiv, the way you talked
about it.
Now imagine that at scale,suddenly you can build what you
used to buy.
And so you've talked aboutBenny Off because in the

(35:47):
conversation, you know, that'san expensive piece of software,
Salesforce.
It's pretty expensive.
So what about a world where youbuild your own CRM?
I mean, we're not far away.
I wouldn't be advising everyoneon this call to build their own
CRM today, but we did it.

We built our own CRM for we can certainly connect
multiple applications togetherif I want, or like you say, I
can build my own.

I went to a visit of a well-known private equity
person, I won't talk about whoon this call, and they said this
particular firm had $4 billion,this particular partner, $4
billion of SAS assets.
And this person said, not surewhat to do with those SES
assets.
They weren't the biggest of thebig, you know, but still, $4
billion of SES assets.
There could be a risk.
Yeah.
Abiy Akash, quick ones.

Yeah.
I mean, look, I think that, youknow, we've heard Google's
engineers using, you know, AIcoding tools.
But, you know, I think it justmoves the humans up the value
chain to do much more complextasks.
So I think that some elementsof things can be eliminated, but
I don't think you know there'sgoing to be mass exodus of
people.
So that's kind of my view.

New things to do.
Abi, I you I mean, when I'vegone to some of Sabinova's
developer events, I'm blown awayby some of the stuff that
people are showing off.
Like one line of code is doingwhat a whole application needs
to do.
So your thoughts?
I know you're a technooptimist, so bring us home.

Abhi Ingle (36:58):
When I look at the long arc of history, every new
technology development has saidto these concerns, Rajif.
What happened to all the horsesand the horse carriage drivers?
You know, I think this is ascenario where I agree with what
Akash is saying.
This is an area subject toJavon's paradox.
The cheaper we make it, theeasier we get it, the more
people you could use.
And let me ask you something.
Would you want to put a curb oncreating intelligence?

(37:19):
That's what I ask.
To me, this is where what Davidstarted off with.
To me, intelligence is thehardest asset of all for us to
replicate.
We found a way to replicate it.
By God, let's harness it andfree people up from some of the
drudgery that Akash was talkingabout earlier.
I think will there bechallenges?
Undoubtedly.
Will there be somedisplacement?
Undoubtedly.

(37:40):
Show me a technology productthat has not gone through this.
I think it comes down again tous having the foresight to plan
for it and allowing people tofind alternate ways to get
initiated and find moreproductive users for their time.

But Avi, he's what happens when you've got 7
billion people that all haveaccess to ultimate intelligence.
So, what does that do for oursociety?

Abhi Ingle (38:00):
The way I look at it is look, there's so much
prosperity created by peopleknowing how to code.
At the most optimisticprojection, we had about 40, 50
million coders.
You know, Akash or David mightknow better.
Maybe.
At the stretch, we've nowcreated an ability for 7 billion
people to automate.
That's fundamentally what we'redoing, and that unleashes so

(38:22):
many people who are just assmart, just as capable.
They just didn't have thatspecific technical capability,
which requires years and yearsof practice.
I think we've unlocked 8billion people to dream and
create.
And that is the promise withgenerative AI in my mind.

Yeah, I love it.
I'll just give you my simpleexample.
Let's say I have 200 people andI have 20 that are on the
development side.
The development folks can workon the architecture, making sure
things are integrated well, aresecure, right governance, right
access levels, et cetera.
All of a sudden, I have 180people who can now take their
knowledge and turn it into codethat learns and scales.

(39:01):
They can add their nuance andthey can break into new places.
It's a whole new game.
And we're going to inventthings that people haven't
thought of.
And I think that's the issue.
In their minds, we can'tforesee what it could be.
We just look at linear paths,and maybe that's what it is.

Maybe it's what it is.
But that's RB's lamp lighters.
I mean, I think that is thething that humans have always
found a way to replace the lamplighters.
And we should find a way heretoo.
Although I can't wait to seewhat they think of, Reggie.
What are they what are theygoing to think of?
It'll be fun.

This brings me to our next section.
So all of us have MBAs, or wethink we have some form of MBA.
So we're going to talk aboutthat today.
So we're leaning into one ofthe most polarizing debates of
the business world.
I'm going to move you a littlebit off the AI subject and move
to this.
The value of the MBA.
Is the master of businessadministration still the gold

(39:49):
standard credential?
Or is it an expensive, outdatedrelic?
With the rise of theentrepreneur, the digital
investor, and AI, the $200,000degree is facing more scrutiny
than ever.
So we've compiled somecontroversial opinions that both
celebrate the MBA's enduringpower and dismiss it as a
two-year vacation for the elite.
Get ready to decide is the MBAa necessary launch pad or an

(40:13):
irrelevant liability in themodern economy.
So here we go.
Quick, quick answers from youguys.
Okay.
Yes, no, quick answer.
MBA is the fastest, mostcapital efficient way to acquire
a co-founder or strategicinvestor.
The two years are an unmatched,high-stakes vetting process for
talent and capital access.

(40:34):
I'll go David Akash Abi.

I can only give you my long answer for you, yes,
yes, and no.
From an Australian perspective,if the answer is no, that's
from the Australian perspective.
I can't give you the equivalentanswer from a US perspective.
I think the answer is no fromAustralia.

Okay, thank you.

Gosh.

The answer from Silicon Valley, I think, is no,
but the rest of America, perhapsyes.
So I think that, you know,again, everything is nuanced.
It depends on the individual,it depends on the person.
If you're a builder, perhapsyou can get going by just
building.
If you have a builder and youare only narrowly focused on
building and you don't seeanything outside building, then

(41:12):
you need a co-founder with abusiness.
And that that person may or maynot need an MBA.
So it's a little bit nuanced,but I think there are many
people in Silicon Valley thatcan argue, you know, with a
clear no.

So I got two nosed, Abby.

Abhi Ingle (41:24):
You know, I'm not going to purport to speak for
the world or for Silicon Valleyor the everybody else.
I'll speak for myself.
It was a yes because I was acomputer science and math double
major.
I had zero business background.
And I personally went to an MBAbecause I got tired of people
telling me what to do and be inthe back room.
And as an engineer by training,I wanted to actually have the

(41:45):
ability to make those decisionsfor myself.
So for me, it was a yes, butI'm not going to purport to
speak for the world.
I know for me it complementedmy engineering background with a
set of valuable problem-solvingleadership and elements.
And besides, I got no greatpeople like you and Akash.

So I love it.
What's the meme, Rajid?
It's a Silicon Valley meme,right?
Which is the picture of theengineer who's saying, you know,
I'm really pleased AI is herebecause I used to have to deal
with those rotten MBAs thatwould cause me trouble and I
didn't understand the productand didn't know the market, and
they were just like annoying andslowed me down.
Then the next frame of the memeis the business guy going, Oh,
I'm so glad we have AI.

(42:21):
I don't have to deal with thoseengineers anymore that were
causing me trouble and never didwhat I wanted and always caused
me problems.
So uh interesting times.

We could do vibe coding, we could do vibe
marketing, vibe businessdevelopment.
Everybody's just vibing witheach other.
Let me go to the next question.
In the age of AI, soft skillsare all that matter, and the MBA
curriculum is too focused onquantitative modeling.
The future leader needsemotional intelligence and
narrative crafting, which aformal MBA fails to deliver.

(42:50):
Who wrote this?
Anyways, go ahead.
We'll go backwards.
Abi, you start.

Abhi Ingle (42:54):
I don't know which MBA you're talking about, but
the MBA that I received was allabout narrative crafting.
In fact, people with Britishaccents always did better in
their grades, in my opinion, atHPSC.
It was all about the narrative.
Way better.
David, you might have actuallydone well out there because they
might have thought you werehaving a British accent, even
though it's Australian.
And it's very clear to me.
So from my perspective, it doesdepend.

(43:14):
I do believe that narrativecrafting has always been
important.
Emotional intelligence is justas important.
I do, though, believe thathaving a fundamental
understanding of the conceptsand then applying AI to do the
drudgery for that does allow youto guide it.
Even those people who franklycan't code and are being to
code, having a little bit of abackground in knowing how to
code actually makes you able touse the AI coding tools better.

(43:37):
So my answer is going to be atleast the MBA I went through
taught me a lot about thinkingon my feet.

No way.

Okay.
Akash.

Yeah, I think it's very important.
I think I would have toslightly disagree.
You know, I felt that it wastoo much focused on analysis and
two by two matrices andbuzzwords.
You know, at the end of theday, success is about working
with people and finding ways tocompromise, finding ways to fill
the gaps.
And business is about people asmuch as it is about anything

(44:05):
else.
And I think some of theprograms, you know, need to kind
of over-index in that now thattechnology is coming and
intelligence is in abundance.
This is not in abundance.
So I think that programs haveto really up their level and
really focus on that.
You know, you can't just sayfire the CEO.
That's a classic HBS response.
Change this, change that.

(44:25):
You can't do any of this stuffbecause in reality, you're
working with a set ofconstraints, and those
constraints are people,personalities, and their
opinions.
You can't just do that.

Not everyone is Eric Peterson.
Okay, go ahead, David.
Your answer.

Well, I mean, those two folks are pretty good
representations of the HarvardBusiness School.
You've done well.
You'll be getting a check inthe mail from HBS, I'm sure.
Look, you know, obviously it'sa combination of skills that
makes people successful.
You've got to be able to doeverything, particularly in
today's world.
And back to what Ivy said, youcan.
If you use AI as yourassistant, your true assistant,
you work with it, then noexcuses.

(44:58):
You should be able to do itall.
And maybe today's MBA is do itall.
It's all the skills.
It's a combination, it's abucket of all that stuff, and it
produces the best people.
And those people use the besttools, as humans have always
done, to produce the bestoutcomes.

Love it.
Okay.
This time I'm gonna go, I'llchange the order.
Akash, I'll be in David.
The last question in the MBAsegment.
A top-tier MBA is the only wayto gain instant credibility when
pivoting into complex fieldslike venture capital or private
equity.
You can't learn sophisticateddeal structuring or fund
management on YouTube.

Well, you can learn it on the job.
So, you know, again,technically you don't need an
MBA if you've worked in afinance company with an
undergrad and you're aroundthat, or you can learn it as an
entrepreneur.

Remember, this is pivoting.
The question's pivoting.
Pivoting.

Yeah, I mean, I guess it does help with the
pivoting.
It's like a catalyst.
Is it the only way?
No, but it certainly definitelygreases your chances of success
doing that.
And you know, you could arguedo the classrooms really teach
you about realistic financialsituations that companies are
in?
I mean, those are again casestudies, and they're you're
you're trying to solve themmechanically through a vacuum.

(46:10):
And again, you could argue youcould feed that same information
into an AI agent and it couldgive you a prescriptive answer.
But a prescriptive answer isnot what really happens.
Every deal has nuances that aremore complex than kind of
what's shown on paper.
So that's my thing.
It's a start, but it's not acomplete answer.

Abhi Ingle (46:29):
This one I'll go the other way.
I absolutely don't think it'snecessary.
I mean, look, there are morethan enough successful examples
of founders who didn't know Jackabout financial structural
who've done just as well.
I think today an amazing VC issomebody who's been a founder a
couple of times and has gonethrough this process.
I think that's just as valuablea skill set.
And I think there's severalfounders who I think would put

(46:50):
many an MBA to shame for whatthey've achieved.
I think it just comes down towhat David said earlier.
It's about the set of peopleworking together.
Whether you're a founder with atechnical background or an MBA,
if you know how to surroundyourself and know how to work
with people and supplement yourego and saying he knows
something better than me, or sheis so far better at me than
this particular area, I thinkthe result is a better outcome

(47:13):
for everybody.
And so I don't think this isnecessary in this day and age.

Great point.
David, I'd say that's notnecessary.
But look, I think workingtogether, as Abby was saying, is
everything.
This is the new skill.
You said it before, Rajiv.
It's in a world where you haveubiquitous intelligence, again,
crazy to say, crazy to say, butlet's just use that phrase.
In a world where you haveubiquitous intelligence at the
MBA level, then it's going to beabout how do you work with
people.

I'd love these opinions.
I I would say from what I'mseeing today, I think Jeffrey
Busgang from Harvard BusinessSchool just ran a webinar with
alumni talking about how they'reusing AI in the classroom.
And it is what you all weretalking about, about the
layering of skills.
Like if you think of the MBA aswhat it was before, I mean,
they're going in now.

(47:55):
Basically, you're askingquestions of the case bot,
essentially, before you walk in.
And now instead of setting upthe case, you're walking into
talking about it at a greaterlevel of depth because you have
AI to have those initialconversations and really jump
in.
So it's about the MBA beingused in an appropriate way,
building upon your skills andlearning who to work together.

(48:17):
I mean, look, because of thoseclasses, I got to know Abi, and
years later, I'm still chattingand working with him.
So it's people.
It provides a tremendous valuefrom that point of view.
These are great answers.
I appreciate it.
And so will all the topbusiness schools.
So now we're going to move tothe Spark Tank.
All right.
So this is where you get toreally put on your fun hat.
Today we're joined by threeexceptional leaders who are

(48:41):
building sovereign, scalable,cost-efficient AI infrastructure
at Southern Cross and over atSalmanova.
So we have founder, CEO DavidKeene, Chief Strategy Officer
and Co-founder Akash Agraval,and Chief Strategy and Product
Officer Abi Angle of Salmanova.
So for now, we're setting asidethe high-stakes world of AI to
look at a different kind ofmania.

(49:03):
Pickleball.
This sport has blown up,creating an entire culture of
devotion and sometimes extremebehavior.
Here's the deal.
One is a complete fabricationdesigned to sound just plausible

(49:25):
enough to make you second guessyour judgment.
I'll count down three, two,one, and all three of you will
reveal your answerssimultaneously.
Ready to separate pickleballfact from absolute fiction?

Let's go.
Let's go.
Let's go.
Let's do it.

I'm waiting for Akasha to clap his hands.
The pickleball massive.

Let's go.
I got the pickleball.

He's got it.
He's got it.
He is the pickleball coach ofthe year.
All right.
Two shoots and a lie.
Pickleball maniacs.
Here's question one.
At the 2024 National PickleFest, two fans legally changed
their names to Dink Vader atSmasherella after hitting 300
volleys without dropping a shot.

(50:05):
Number two, a group called thePickle Breakers set a Guinness
World Record in 2025 for thelongest pickleball marathon
ever.
36 straight hours of mixeddoubles in Carrollton, Texas,
raising $18,000 for charity andallegedly surviving on energy
gels, beef jerky, and picklejuice.
Number three, pickleball'spopularity explosion led to some

(50:30):
U.S.
towns establishing officialquiet hours after neighbors
complained that the sound of thepock drove their dog into
chaos.
Okay.
Three, two, one.
One for David, two for Akash.

Nobody hits 300 shelves in a row like that.

Come on.
So David says one.
Wait, Akash, what do you say?

Two, two.

Two, and Abby says one.
Okay.
David says nobody hit 300volleys while dropping a shot.
And he is right.
So that is the lie.
While Dink Vader wouldabsolutely dominate, there's no
record of legal name changes ata pickle fest.
Though pickle themed nicknamesare wildly common among super
fans.
So they're not maybe you can'thit 300 volleys.

Abhi Ingle (51:17):
What is Akash's pickle name?

Akash, I don't have a pickle name.

You don't have one like DJ Cash.
You don't have the cut.
No, I don't have a picklemaster.
Okay.
The pickle breakers 36-hourmatch raised funds for Taylor's
Gift Foundation, combiningathletic endurance with a
heartfelt cause, a rare mix ofexhaustion and empathy.
And the pickleball noiserestrictions, number three, are
very real.
Some cities even conductedacoustic research to reduce

(51:41):
paddle clatter.

Well, they won't let you build a pickleball court
in certain Bay Areaneighborhoods anymore.
That's right.
In Atherton, the the countryclub there is under severe
pressure to reduce the noise.
So what some innovators havedone is built these new paddles
that absorb the noise.

Oh, that's a good idea.
Number two, the oldestpickleball player, Joyce Jones,
holds the Guinness World Recordat age 95, saying her only
secret is pickleball, Pilates,and prunes.
That's the that's answer one.
Number two, a Wisconsin manonce had pickleball patties
carved into his wedding cakedesign, with the newspaper

(52:20):
quoting his spouse, whoapprovingly called it love at
first dick.
And number three, in 2025, fourplayers in Montana attempted a
marathon pickleball session thatlasted 25 hours, just pausing
every hour for five-minutehydration breaks, and one
spontaneous interpretive danceto Eye of the Tiger.
Okay, so first one oldestplayer, second one, pickleball

(52:44):
patties, third one marathonpickleball session 25 hours, and
Eye of the Tigers.
Ready?
Which one is a lie?
Three, two, one.
I'll be as three, David.
It's two.
Akash, you are three, three.
Okay, David.
Why do you think number two?

Nobody's wife would agree to that on their wedding
day.

All right.
No, Akash, would your wifeapprove of this?

No, she wouldn't, she wouldn't.
But you know, I know there arepeople out there because my
wife's not playing pickleball,but there's some wives that
playing pickleball, so theywould agree.

Which one did you pick again?

I picked three.
I picked three.

Why'd you pick three?

It's just too long.
I think it's just too long.
25 hours, whatever.
Yeah.
Well, guess what?
David is right.
Wow, David's got two out oftwo.
David is the lead.
It's thinking like anAustralian.
Exactly.
He's thinking like anAustralian.
He's fine-tuned the model.
Yeah, he's fine-tuned thepickleball model.
Yeah, there you go.

Okay, so David thought correctly.
Although wedding cakes haveindeed featured pickleball
paddles, there's no record ofLove at First Inc.
as a headline just yet, but itcould happen.
Number one, Joyce Jones' recordat 95 cemented her as a living
icon.
Guinness recognized her forlifelong vigor and funky neon
sneakers.
And number three, theMissoula-Montana marathon was

(54:01):
real, cementing the idea thatendurance pickleball is a thing,
not a punchline.
All right, let's see who cancome back on this way.
Here's number three, number onepoint, or number one truth or
lie.
In 2023, a group in Wisconsinbroke the Guinness World Record
for the longest pickleballvolley rally, lasting more than
14,000 consecutive hits, withplayers taking turns using slow

(54:25):
motion dinking to keep the rallyalive.
Number two, pickleball fansonce organized a competition
called the paddle limbo, whereplayers see how low they can go
under the net without droppingtheir paddle or losing a point.
The championship record islimboing under a 10-inch net.

(54:46):
And number three, in 2025, USpickleball reported that over
4,200 new pickleball locationsopened nationwide, catering to
the sport's explosive growth.
Which one is the lie?
Three, two, one.
Two Abby, two David, two.

We better be right, guys.
We're with you on this one.

Why do you think Paddle Limbo won't work?

It's just too low.
Ten inches is too low.
Oh, we're just too old, guys.
We just too old.

Obviously, that was a layup.
Thank you, staff.
So all three of you win on thatone.
So I get to give each one ofyou a point.
While Paddle Limbo sounds likea hilarious fan idea, it's not a
recognized or sanctionedcompetition.
Number one, the Wisconsin rallyset a record for the longest
volley rally, verified byGuinness and covered by local
news, illustrating fans' staminaand creativity with slow motion

(55:39):
techniques.
And number three, the 2025growth report confirmed a
massive surge of new courts withover 4,200 added the prior
year.
Pickleball is just exploding.
So here's the final one.
And in this one, let's make itinteresting.
I will make it so that thewinner gets two points.
So this is a chance to beatDavid.
Which one's a lie?

(56:00):
A 2024 tournament in Floridacreated a pickleball costume day
with players dressed as giantpickles, paddles, or even jars
of brine, drawing crowds andeven viral TikTok videos.
Number two, in 2023, apickleball fan from Oregon held
the world's first everpickleball hot dog eating
contest between matches at alocal tournament, drawing over

(56:22):
500 spectators.
Number three, the quote dillball has been unofficially
adopted as the sports mascot,complete with custom jerseys and
a theme song played at majortournaments since 2023.
You ready?
Three, two, one, go! This isthe chance to win.
David has three.

(56:43):
I see Akash at three.
I see Abby at two.
Okay, Abi, support your case.

Abhi Ingle (56:49):
Oregon hot dogs.
I mean, it just didn't seemlogical to me at all.
Like the crazy things thatpickleball fans do, like
dressing up in pickles, isplausible.
The last one is so plausible.
The second one, what if hotdogs got to do with pickleball?
I have no idea.
Just didn't seem logical to me.

Akash, take them down.
Take down Abi.

Yeah, people eat hot dogs with everything.
So, you know, why notpickleball?

David, do you have anything else on that?

I had to pick three because if if there was a
different national theme songfor pickleball, Akash would have
already played it.
It'd be playing in his houseall the time, so it can't be
that.

Yeah, yeah.
Yeah, David spent some timewith me.

If there was one, Akash would be playing it.
Yeah, Akash would know it, Iguess.

That's just a good point.
That's a good point.
I didn't know about it either.
I don't know any song that'spickleball song.
David's right.
I would have known about itwhether I'd play in front of
David or not, and I haven't.
Maybe I missed it.
I've been busy lately here.

Well, the correct lie and the winner is number
three, is false.

So David has won it again.

Perfect score.
So David is the winner.
He gets the extra two points.

Just like the cricket, guys.
Just like the cricket.

While Dill Ball would be an incredible mascot,
no official or unofficial mascotwith that name exists in the
sport, which is correct becauseapparently Akash would know
that.
And the pickleball costume day,the number one one events have
cropped up at varioustournaments, especially in
Florida, sparking festiveatmosphere and a viral social
media buzz.
And number three, the OregonHot Dog Eating Contest was a

(58:14):
well-documented, quirky sideevent at a pickleball tournament
showing fans' appetite for funand food.
There you go.
You guys did great.
That was a fun pickleball game.
It'll certainly do well on ourYouTube shorts.
So I appreciate you guyshumoring me for that.
Let's get on to uh somepersonal closures.
Okay, David, if you had to picka superpower based purely on

(58:35):
making daily life moreconvenient, not saving the
world, what would it be?

The flesh.
Speed of the flesh.
Get a lot more done, see morepeople, experience the world.

Okay.
What's something you used to bereally into that you now find
completely baffling about yourpast self?

Worrying about things that don't matter in
life, really.
I mean, I think maybe I was nottaking into account people's
feelings as much.
So I think that those arecritical.
I think that, you know, perhapsI didn't pay as much attention
to that.
So, you know, it's it's sort ofa reverse answer in the sense
that I didn't worry about them,but I worry about them now.
So I think that's veryimportant as you get older and a

(59:14):
little bit more what peoplethink of what you might say.
So I think that's very, veryimportant.

That's an incredible answer.
Abi, what's something you'regrateful your younger self did
or didn't do that's paying offnow?

Abhi Ingle (59:27):
My mom enrolled me when I was in the fifth grade
summer in a computer codingclass, and I was pissed off.
And I'm just so grateful andthat she did that.
And actually, today's her 12thdeath anniversary.
So I thank her for everythingthat I have and everything I am
every single day.
Amazing.

Definite love to your mom.
She's an amazing person.
Okay.
If you could give everyone inthe world one piece of
information or one realization,what would it be?

Well, I was gonna say they should visit scx.ai to
experience the future ofsovereign AI.
But apart from that, it's thatthe world is all about people,
it's about making sure that.
That you have people in yourlife like the folks on this
podcast that you love workingwith and you can build a life
around other people.

Rajiv Parikh (01:00:06):
Akash, what's the most interesting thing you've
learned recently from a randominternet rabbit hole?

Akash Agarwal (01:00:13):
There was someone, was it you, Dave, who
told me this about the Bitcoinexample?

David Keane (01:00:17):
The Bitcoin duplicator.

Yeah, the Bitcoin duplicator.
So what was that, David?
Can you share more about thatone?

If anybody's on this podcast, don't visit the
Bitcoin duplicator where theyask you to put in your
blockchain ID and they promiseto give you back twice the
number of bitcoins that you putin because it always works.

Yeah, that was the one.
That was the one.
So hopefully I won't fall.

You chase that one down?

Yeah, I didn't chase it down, but I was very
curious.
I was very curious.
Right down.
Yeah, exactly.
Right down to zero, yes.

Okay, Abby, what's something you do to feel like
yourself when life gets chaoticor overwhelming?

Abhi Ingle (01:00:51):
I go for a long bike ride with friends and just
exhaust myself, or I go for awalk with my dog and my wife.
There's nothing more relaxingfor me than that.

I thought you were gonna say I go to Rajiv's
house.

That too.
That was the third thing if Iwas allowed a third thing.
That was his backup.
When the other two don't work,he comes to that.

Yeah.

All right.
What's the most memorable mealyou've ever had and what made it
so special beyond just thefood?

People shouldn't cry at this, but it's dinner
with my wife last night.
Because I've been traveling alot and I got to have proper
dinner with my wife at a lovelyItalian restaurant downtown in
Boston.
It was the best meal I've everhad.

Wow, what a romantic David.
Yeah.

Well, his answers are perfect.
Avi, can you believe it?
I think he's using magpie.
I think he's got magpieembedded in his brain.
Exactly.
There's something, yeah.

You still have to have the competition, uh, you
know, Raji.
We still have to know whoevercan identify what the bank pie
used first in the comments getswhat should we give them?
A thousand dollars of freetokens on scx.ai.
Whoa.

You heard it here.

That's it.
But they're gonna identify whatit is.

Let's do it.
Thank you.
Thank you for all those extracomments, folks.
I appreciate it.

I think for all of my memories of food are actually
tinged with memories of thepeople I've had them with.
So it's really honestly,sometimes I don't even remember
the food.
It's the laughs we had at thetable or the special event is
at.
So I kind of with David on thisone.
I'm not gonna go syrupy and sayit was just with my wife.
But over the years, it's aseries of special experiences to
me.
Meals and people are thingsthat just make life worth it.

I would just say I find street food with people
just joking around, uh standing.
You don't need to even besitting to be quite
entertaining.
And you know, differentcultures with different people
that know those cultures.
So on travels, you you know, Itry and meet with people in
their local countries and tryand indulge in some of their
local street food.
So that gives me joy.

I appreciate that.
Yeah, that's a great way ofputting it because you really
see what's in their heart.

Yeah, yeah, exactly.
You know, how how the taxidrivers or the local people
enjoy, you know, their snacksand their culture.

Amazing.
Okay, Akash, one pickleballfactoid that we didn't know
about.

You can actually go in the kitchen.
How's that?
So, you know, they say youcan't go in the kitchen.
You can go in the kitchen.
You can go in the kitchen whenthe ball has bounced.
You can go in the kitchen bygoing and standing on the side
of the kitchen and putting yourarm out and preventing the ball.
Physically, your legs can't bein the kitchen and pick a ball.
So there's a couple ofworkarounds.

(01:03:18):
One is when the ball's bounced,you can stand near the post and
basically put your arm out andstop the ball.
That's so that's one.
And the other one is that whenyou're serving, you can serve
into the opponent's body andyou'll win the point.
If the ball touches youropponent's body, you win the
point.
So that's a technical rule andnobody can argue against it.
So when you're playing withsomeone and it's 11-0 or it's

(01:03:38):
12-11 and you really want to winthe point, serve the ball into
you didn't hear it from me.
You can serve the ball into therecipient, the person that's
standing near the kitchen, andhe can't.
If basically you're gonna winthe point unless he avoids the
ball.
Because if it touches hispaddle, if it touches him, you
win the point.
That's what happened to me in atournament.
I didn't know this rule.

(01:03:59):
It was 12-11, and the guylooked at me and laughed, and he
said, Clearly, you should haveknown the rules.
I said, You're right.

That's the new Akash.
He actually said, You're right.

My my sense is our car saying something really
rude back to him.
That's what I'm thinking.
I didn't say anything becausethat was the wrong.
There were three other people,so you got to be careful.
I would have gotten anotherpoint taken away if I'd said
something rude.

I want to thank you all for being with us today.
I definitely would love it ifyou guys all go to scx.ai and
learn more about sovereign data.
I think it's more thanAustralia.
I think there's Australia, butit's a great model for what
could be an incrediblephenomenon around the world.
And I think, you know, weshould all go check out
Salmanova's website, which acertain company helped design

(01:04:42):
and develop, you should checkout and learn more about the
multi-layered solution theyhave.
And I thank all three of youfor joining us today.
You guys were really amazing.
This is fast emerging, and soI'm really excited to have all
you guys come together.
Thank you so much.

Thank you for having us.

Thank you so much.
Great event.

Yeah, great event.
Thank you guys.
Bye-bye.

That was really interesting and fun from people
who I've played pickleball with.
What I took away was there'sthree people who've been
executives and entrepreneurswith somewhat similar
backgrounds, but diversebackgrounds, talking about how
they're building this incrediblydisruptive, innovative set of

(01:05:28):
businesses.
And this whole notion ofsovereign AI may sound very
esoteric to you, but is verypersonal.
And I think David really pulledit out when he talked about the
need in Australia to havethings about the way people in
Australia talk and act, theircustoms, their legal language,
their regulations, the way theylook at things, in the way their

(01:05:52):
AI applications, conversations,capabilities are developed.
And then there was the greaternotion of splitting the notion
of training and inferencing.
And Abi explained it so well,kind of like AI 101 for
business.
If you really understandinferencing and training and how
they're two different things,you can turbocharge the way you

(01:06:15):
look at budgets and how youbuild applications and how you
drive them.
And then we learned a lot aboutcompetition and how this AI
world isn't necessarily leadingtowards winner take all, even
though it may seem that waytoday.
There was a time when Yahoo wasdominant.
There was a time where in thebrowser space, Microsoft was
dominant.
Nowadays, there's even newbrowsers coming from the

(01:06:36):
different LLMs.
You know, Chrome was supposedto be dominant.
That's blowing up too.
So it's the amazing nature oftechnology and the sense of
optimism in terms of what ispotentially out there.
This isn't job replacementnecessarily.
This could be uplifting to youin your career and your life.
And that's the kind ofinteresting optimism I saw from

(01:06:58):
all three of them.
And during the MBA portion,wasn't it notable how each one
thought of their MBA experienceand what's notable today about
the new one?
You can relate to what washappening in the past, but it's
changed dramatically of what'shappening today.
But in many ways, these folkshave known each other prior to
this through various events thatsome of which I'm proud of

(01:07:19):
putting them together for.
They've met each other and nowthey're working together and
doing business together andmaking a difference together.
And that's the power of thenetwork.
And that's the power, as wetalked about in our last episode
of going to that extra event,going to meet folks, just
putting yourself out there andnot having any expectation other
than openness to serendipity.

(01:07:40):
Of course, we had a lot of fun,a lot of playfulness.
And I think this is somethingto aspire to if you're out there
and you're steadying yourselfup and saying, wow, you know,
you have David, who's amulti-time entrepreneur, same
thing with Akash, same thingwith Abi.
These guys are amazing folkswho keep hitting different
levels in their life and theynever, they never stop going.
And it's something thatinspires me every single day

(01:08:01):
when I get to hang out withfolks like them and play
pickleball with them.
All right, thanks forlistening.
If you enjoyed the pod, pleasetake a moment to rate it and
comment.
You can find us on Apple,Spotify, YouTube, and everywhere
podcasts can be found.
You saw David's promotion aboutthe $1,000 token credit.
Take advantage of that today.
Go to SCX.ai.
This show is produced by SundayParik and Adam Shah with

(01:08:23):
production assistance by TarynTalley and edited by Laura
Ballant.
I'm your host, Rajiv Parik fromPosition Squared.
We are an AI enabled growthmarketing company based in
Silicon Valley.
Come visit us at position2.com.
This has been an F Funnyproduction, and we'll catch you
next time.
And remember, folks, be evercurious.

All Episodes

Why You Can't Build Sovereign AI Alone/David Keane, Akash Agarwal, Abhi Ingle - SCX.ai, SambaNova, Australia ~ Spark of Ages Ep 50

Episode Transcript

Popular Podcasts

Las Culturistas with Matt Rogers and Bowen Yang

Crime Junkie

Stuff You Should Know

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Why You Can't Build Sovereign AI Alone/David Keane, Akash Agarwal, Abhi Ingle - SCX.ai, SambaNova, Australia ~ Spark of Ages Ep 50