Subscribe
Share
Share
Embed
Around the Desk: This is where we at Avory think out loud, challenge narratives, and look for signal through the noise. Each episode, the Avory & Co. team dives into what’s moving markets, how companies are performing, and where opportunities may be forming. We break down earnings, macro trends, and investor sentiment — all from the lens of a concentrated, high-conviction portfolio.
*** The views expressed on Avory Podcast: Around the Desk are those of the hosts and guests and do not constitute investment advice. This podcast is for informational purposes only and should not be relied upon to make investment decisions. All investments involve risk, including potential loss of capital. Avory & Co. may hold positions in the companies discussed.
For 18 months, I think we've all heard
the same story: AI will kill software.
Klarna is dumping Salesforce.
You know, that's what you listen
to, that's what you heard.
Agents are replacing seats.
SaaS is dead.
You know, then this month, you know,
again, we're, we're sitting here in
August, Salesforce accelerated CRPO,
Current Re-Remaining Performance
Obligations, to fourteen percent.
Then you look at someone like
ServiceNow holding, um, you know,
CRPO at, you know, twenty-one percent.
Okta crushed it, accelerated
from fourteen from twelve.
Palantir's obviously doing its thing.
MongoDB with Atlas grew
twenty-nine percent.
NVIDIA guided, you know, next
quarter, um, to, you know, I think
a hundred and eight billion dollars.
You know, two things everyone
said were opposites are, are
really running at the same time.
And today, you know, we're gonna talk
about why, and why the person, you
know, who was, you know, uh, essentially
looking at the market and, and making
some of these calls, I think, you know,
was very too, you know, significantly,
uh, too linear in their thought process.
Welcome back to Avery Around the Desk.
I am Sean Emery, founder and chief
investment officer here at Avery and Co.
Each episode, we take the
important investment questions,
work through it using data, first
principles, long-term thinking.
We're here around the desk on a Thursday.
You know, reporting season has happened.
Before we get into it,
though, this discussion is
for educational purposes only.
Should not be considered
investment advice.
Opinions reflect our views here at Avery.
As of the recording
date, things may change.
New information may emerge.
Please do your own research.
Consult a professional before making
any investment decisions on your own.
So let's get into it.
You know, today's question is simple:
Is the SaaS apocalypse narrative wrong?
And if it is, what does that tell us
about where we are in that AI cycle
that everyone is trying to figure out
the timing, the duration, the quantity,
the who, what, when, where of who's
going to win, who's going to lose.
The reality, though, is we are seeing
in real time the development of AI on
top of everything that was built before
it, and I think that's super important
to put in the back of your mind.
, in general, look, the consensus I think
that we're rejecting here at Avery, , is,
you know, the SaaS apocalypse narrative.
You know, agents will, you know, replace
seats and, uh, per-seat pricing dies.
Incumbents get, you know, you know,
disrupted, let's say, you know, by many of
the AI narrative or, or native startups.
, I think the, the evidence that people
were, were showing here, at least early
on, was, you know, how Klarna went from,
you know, this idea that, , they're gonna
r- you know, get rid of, , Salesforce and,
you know, cancel that and, you know, they,
they're getting rid of all their customer
support only to bring them right back on.
So again, there was, there's plenty of
tell, plenty of signal, and if you follow
our work on our investingwithdata.com
or Investing With Data
newsletter, , big plug there.
But in general, you know, we've been
articulating this stuff through data,
through signal, through noise, , and
really trying to, , ensure that we have
a good grip of, you know, what's really
happening here underneath the surface and
getting away from some of the narratives.
But, you know, you saw a little bit
in 2024, 2025, you know, some of these
software, , growth stories at least
plateau, and I think a lot of that just
has to do with the duration of software.
You know, it's really been a
20-year, , endeavor, let's say.
, Also, you know, the consensus was You
know that, you know, cloud, or at least
like historically, right, everyone was
using the same, , rhythm of the past.
And it's really,, how cloud killed
hosting and mobile killed desktop.
And, you know, why wouldn't, AI kill SaaS?
And I think the problem is, again,
they were extrapolating, you know,
a lot of this out way too far.
, But when you look at the backlogs,
not only for, , the, , the core AI
players, but, , beyond that, you're
starting to see some of that backlog,
, through, , RPO, remaining performance
obligations, or CRPO, which is current
remaining performance obligations.
Uh, this isn't really a gap number, right?
This isn't something
that hits the headlines.
It's something you have to dig into
a little bit, , to understand that.
So let's kind of like walk
through some of those things.
You know, you look at
Salesforce, revenue 11%, right?
CRPO, current, current remaining
performance obligations, 33.5
billion, up 14%.
So that accelerated
from, , , the prior quarter.
Agentforce, ARR at 1.5
billion, 2,000 additional
paying customers in production.
Agentic workloads at 3.2
billion, , you know, users, and
that's up, , 97% quarter over quarter.
And then they raised, obviously,
their revenue guide for the full
year, which makes a lot of sense.
Not only that, you had
Dario on the right chair.
You had Benioff on the left chair,
which is, you know, the CEO founder of
Salesforce and the CEO founder, founder
of Anthropic sitting together talking
about how they use each other's, you
know, products and solutions, which
again, is such a narrative violation
to the AIs disrupting this or that.
And, and Dario at Anthropic basically
saying that they're massive users
of, of Salesforce, and then they
co-branded, , Claude Force, , as
a solution that they're, you
know, gonna go to market together.
And again, like if there was ever a
ribbon, you know, cutting ceremony
that would suggest that, , there is
this convergence of software and AI,
, vendors, , that, that, that was really it.
, I think more than ever, , as time's
has gone on, it's, it's more and more,
commonplace to think, at least logically
for us, is that, you know, many of these
AI labs actually need these software
vendors because at the end of the day,
, that is your sales force is, , and I don't
mean Salesforce the company, I mean that
is your sales force to sell AI into the
enterprise is more likely to, , embrace
and, and work with the current vendors.
, Second, you know, that was a
lot on Salesforce the business,
but ServiceNow, like same thing.
You know, subscription revenue up
twenty-four point five percent.
CRPO up twenty-one, , twenty-one and a
half in constant currency, which was,
you know, somewhere around two percent
above, , their original guidance.
And ServiceNow AI crossed
a billion in ACV, so annual
contract value, , in the quarter.
So you can just hear how AI is starting to
emerge as, you know, a real, , line item.
Okta, , big owners, , for
us, , it-- for Okta.
You know, CRPO grew at,
you know, fourteen percent.
That's an acceleration, a two percent
acceleration from the prior quarter.
And, , honestly, this is like pre
any of the, their agentic, , AI, uh,
identity solutions actually hitting
, their revenue line and, you know,
they called it the largest, , current
remaining performance obligation
outperformance, you know, in over a year.
So, you know, obviously the, the, you
know, the stock reacted positively,
but in general, like it's, , here's
something where agents, AI agents that
are, you know, scanning the web need
identities or otherwise you have, you
know, rogue, you know, , AI tools kind
of going all across the ecosystem.
You have Palantir, you know,
obviously seeing strong growth across
commercial, um, and again, their,
their RPO, , , growing triple digits.
So, you know, big numbers there.
They're primarily a software vendor.
, MongoDB, you know, revenue up twenty-five
percent, Atlas up twenty-nine percent.
, RPO, remaining performance
obligations up, , somewhere around
eighty-eight percent, which is crazy.
, CRPO up sixty-nine percent.
, So these are big numbers.
, You have Zoom Enterprise,
, accelerating here, , to,
, somewhere around eight percent.
, Their AI, , , tooling, , , monthly
active users was up a hundred and
eighty-four percent year over year.
And the amount of, , deployed, , to--
or they had a customer where, who
deployed, , , their AI tools across,
you know, sixty thousand seats.
So all this stuff, you know, again,
just highlights more and more things,
and then you complement that with the
idea that NVIDIA just reported ninety,
, six, you know, a hundred billion
dollars of revenue growing, , triple
digits, um, with data center, you
know, ninety-two percent of that mix.
So, like you have, , NVIDIA,
, accelerating to some degree, and,
and then you have the rest of, , some
of these, you know, leading software
vendors that are kind of gravity and have
distribution, , following, , behind them.
So again, it's, it's kind of a
narrative viol- you know, violation
when you have, you know, these
two forces working at once.
And ultimately, what it looks like is, you
know, some of these software vendors are
becoming big customers of the AI vendors.
, And these AI vendors, , essentially,
, need these software vendors.
And so you're seeing
more, , , uh, f-friendship.
, Some will say frenemies,
depending if they're competing.
Um, but more and more I actually think,
, , NVIDIA is actually competing against the,
uh, AI native ecosystem, and then the
software vendors will do their thing.
Now- A little bit of context here.
I think there, you know, there's
four layers of the AI stack.
Again, they're all expanding
generally at the same time.
You have the infrastructure
layer, and that's NVIDIA, right?
And that's, you know, , some of the
chips and, , some of, , the hyperscalers
obviously with Google, Amazon,
Meta, , , Microsoft, Oracle's in there.
You have the Neoclouds, which are
AI-specific - , infrastructure.
, You know, there's kind of this
love-hate relationship there.
Those Neoclouds do everything differently.
You know, you have like a Nebius,
you know, trying to, , lean
in on the software side.
You have, , Cerebras, which is, , has, has
created their chip, but then also have,
you know, moved and migrated more of,
, their efforts towards, you know, hosting,
, , their chips in their own cloud.
, You have the Coreweaves of the world,
and then you have kind of the line
of, , the players behind them, , as well.
And so that's, , an interesting space,
but, , it-- there's definitely a lot more,
, divergence between, , how they all operate
obviously, , if you actually
think of all those, , that
chain, the picks and shovels,
NVIDIA's encroaching on all of it.
, , Everything from, you know, GPUs
to CPUs to networking to, , all
the investments they've made, and
I'll talk a second about that.
But then you go to the next, , layer of
the stack, it's that model layer, OpenAI,
Anthropic, Google, uh, you know, Meta,
we-- you know, with their ecosystem.
You have Mistral, you have xAI, you
have, you know, um, , some of the, uh,
the, the open source models that, , many
of them are dominant Chinese open
source models that are catching up.
Then you have the third layer, which,
, again, I've, I've s- mentioned.
This is where a lot of the
magic is happening in that
orchestration harness layer.
, Block came out with Buzz, which
is an open source kind of harness.
, They followed that up with,
, another s- you know, , kind
of open source tool as well.
You have, you know, the
Perplexities of the world.
You have, , , plenty of other,
you know, orchestration platforms.
I mean, you're hearing more and
more, uh, companies come out and
say, , "We wanna be the harness."
I think that is the battleground
that, , is really gonna exude the
value across the chain, just given
that it is the orchestration layer.
And anytime there's complexity
and you can build something, a
solution that, , i-improves the
complexity, that's where the value is.
That's why the cloud, , the hyperscalers
like AWS is so valuable and how they can
extract so much margin is just simply the
idea that, , there's a lot of complexity.
It's hardware, software,
services, uh, uptime.
, , All of that complexity is,
you know, really handled by
AWS, and it's as a service.
It's as a service, you know, , models.
Um, and so if you canâ¦
, once you go up the stack, , the models,
and then you have orchestration of models,
I think, again, you're starting to,.
it's a little bit more abstract because
you're, you have to understand, , why
your orchestration layer is choosing
a certain model over the other.
But, , in general, I do think, um, if
you're creating value and the value
will show up in , the intelligence
per dollar essentially, , that's kind
of like the easy way to say it and
ultimately an orchestration or a harness
is, is, is, , increases usability of
these AI tools, but then it also on
a-another side, it increases, , the
total cost of ownership of that.
The fourth is the application layer, and
that is embedded AI across, , existing
SaaS platforms like Salesforce with their
Agentforce, ServiceNow with Now Assist.
HubSpot has, , their Breeze solution.
Zoom has, you know, their
AI, , solutions as well.
Their-- How you win there and why
they're winning is around distribution.
They have a distribution advantage.
They have enterprises
all across the globe.
They have a large sales force
that can sell and tell the story.
Then you have the next is really
like these purpose-built AI, you
know, native solutions and, , those
could be, , MongoDB Atlas, that could
be, , Cursor and, and some of the
other, , purpose-built AI natives.
And they have a product advantage.
They're winning because they're, you know,
starting from the ground up, purpose-built
for AI, and they win in that way.
And I think all this stuff is,
you know, semi-converging, but,
but, , the, the key takeaway here is
that, , the, , the SaaS apocalypse trade
is, , in theory, , not real, right?
There, there are obviously some areas that
are getting disrupted by AI, for sure.
, But if at it, , Salesforce still
owns, , the customer graph.
ServiceNow still owns the workflow.
Okta still owns identity.
You know, Palantir owns,
um, you know, ontology.
You know, MongoDB owns, , the
data, uh, , kinda plane.
Um, NVIDIA owns the, you know,
the compute floor all the way
up, and they continue to move up.
Now the question I think then
reframes, , around, , you know, who
is going to map out this whole cycle.
So like, you know, just trying
to step back and, and think
about NVIDIA's moves here.
Why is NVIDIA not just selling
chips, but really investing across
every layer of the stack, uh, it
doesn't even like sell into really?
Know, and if, , I went back and looked
at, you know, twenty twenty-six alone
and, you know, I think they made an
investment in CoreWeave in January
of sorts for two billion, , Neocloud.
You have, , two billion into, uh,
you know, Nebius and Neocloud.
, I think that was like
March, April timeframe.
, Lambda, Enscale, Crusoe, Groq,
who they, they brought internally.
, I think it was roughly, , thirty
to forty to fifty billion
dollars in that Neocloud segment.
The model labs, they have a hundred
billion dollar commitment to
OpenAI, ten tranches of ten billion.
I know there's five billion, , to
Anthropic, six billion I think
to xAI, , you know, two billion
to Mistral, Cohere, Reflection,,
Thinking Machines, , and some others.
I think, you know, the deal they
announced the other day with Perplexity.
, So that's the model labs, and then you
have the applications and tooling, you
know, , which is, , everything from
like D-Wave and World Labs, , LangChain.
Um, who else am I thinking about?
Um, you know, other kind of like open
source labs as well, and they have their
own, , , open source model themselves.
And I think, you know, again, Neocloud
to the model labs, the applications
and tooling, all of that, , totaling,
, somewhere around, , fifty, sixty
billion dollars across many rounds.
And you ask yourself like,
"Why, why are they doing this?"
And I don't think it's 'cause of returns,
you know, at least in the short term.
I think they need the gravity.
You know, every dollar NVIDIA invests in
a Neocloud is a dollar that eventually
just cycles back through as GPU orders.
And then every model lab they
seed, you know, becomes a customer
of those, of those, essentially
either the cloud, Neoclouds
or, , directly to the, the chips.
And then, every application company
they back is essentially again
a downstream demand signal, you
know, that keeps the,, that stack
calibrated around, you know, CUDA and
NVLink, which is their networking.
And so look, this is a
playbook we've seen before.
Amazon's kind of done this.
They, , they start here and then they
work their way down, and then all
of a sudden you have, , U- you know,
UPS and, and FedEx on their heels.
, And , again, NVIDIA is not a chip company
as much as it is a full stack company.
And , a lot of people before said
it was more around the, , , CUDA,
which is their software.
Uh, I don't think that's it.
I think, , that original thesis
isn't, , probably doesn't hold up.
, But it's just more the idea
that they're going beyond GPU.
With CPUs, , as well.
, They're also then, , obviously moving
into networking, , aggressively, and
that's, - , again, that's that full stack.
And then, look, there's not a worldâ¦
And I was speaking about this openly
the other day, is there's a world where,
you know, some of these, , , companies
across the stack that are, that are large
in size,- the margins they're getting
today, this is as good as it gets, right?
The, the supply-demand is so imbalanced.
And so there's a world where, , they
need, continued funding when things
when demand starts to, plateau, , those
returns on that CapEx start to, , migrate
lower into, , the s- low single digits.
I think, , most people when they look at
a lot of these deals that are happening
on the NeoCloud space, , the actual
return is mid to high single digits.
Not that much.
, And so that could get squeezed out.
And what does that imply?
What does that suggest to us?
That means potentially there's this
moment where Some of the larger
players, , have the opportunity to
go vertical and essentially create
a, , a, a monster , , fifth cloud,
let's say, , beyond AWS and all this.
And, , again, purpose-built for, , AI.
, This is kind of like, you know, uh,
end-of-cycle scenarios, but I think
they're all positioning for that.
Ultimately, they're, they're the biggest
customers of all these companies, and,
, without them, they can't do anything.
So when, , they pull back and returns here
are, you know, high to mid-single digits
at, at peak demand, , , what does it look
like when, when that demand, , comes down?
So anyways, I think, look, neo
clouds have their, , their place.
, There's real risk in that space.
, They're all doing th-things a
little bit differently, so you
wanna be very specific there.
Some model labs probably won't survive.
, Open source is becoming more and more
powerful, smarter, per token cost.
And then, you know, software names
are, you know, doing their thing.,
Especially the ones with distribution.
Again, if you're a point solution, like
we've always said, you're in trouble.
If you are a platform with
real distribution, with real
enterprise presence, I think
you're, , in a good spot.
If you're serving, customers
from a risk perspective, which is
identity, , which is, , cybersecurity,
there's clearly demand there.
I didn't talk about CrowdStrike, but
they had a really strong quarter.
, So anyways, that's really it.
Things to watch going forward, I think,
, you know, CRPO continuing to watch,
, accelerations there across, different,
you know, th-through the rest of the year.
, I think on the negative side, you wanna
look at, uh, NVIDIA's gross margin.
Obviously, they, they guided to,
low seventies from seventy-five.
, So that's the start of potentially
some margin compression there.
But the demand there on the
revenue side is so large, um, so
it's hard to, criticize too much.
But, , at the end of the day, I
think, , their margins eventually
normalize, you know, probably
closer to, you know, sixty something
percent down the line, not right now.
Um, and then model, pricing, let's say.
, That's a challenge.
I think w-- you're seeing
a lot of, change happening.
We'll have our, our, our, , Investing
with Data newsletter, , and we'll
show, you know, Codex versus, , Claude
in terms of, , demand there.
And y- things change fast.
You're one API call away from,
, getting disrupted, , in general.
So look, I'll close out here.
, Earnings this week.
Strong earnings reports by AI natives AI
infrastructure and software, all of that
is breaking the narrative that there's,
there was or is this SaaS apocalypse.,
We really started with that one question.
, We ended up, , somewhere a
little bit different there.
But ultimately, I think, , as the
SaaS, uh, apocalypse story folds to
the back, I think ultimately,, these
companies, , have the ability to lean
further in as opposed to playing defense,
which is kinda what they've been doing
over the course of the last year.
, And you saw that in some of the
price action as they reported.
I think, Okta was up, I don't
know, close to thirty percent.
You had someone like Salesforce
up, twenty-something percent.
, We happen to own both of those,
you know, full disclosure.
, And, that is that.
So thanks for, , joining us here
Around the Desk here at Avery.
, If you enjoyed the discussion,
you know, make sure you subscribe,
wherever you listen to our podcast.
Follow Avery, , for more
data-driven investment research.
We love to do it, and thanks
for joining us, , today.
See you next time.