Latest / In The Loop / The SaaS Apocalypse: Why $1 Trillion Was Wiped from Software Stocks

Transcript

The SaaS Apocalypse: Why $1 Trillion Was Wiped from Software Stocks · In The Loop

0:00
Over a trillion dollars have been wiped from software stocks.
0:03
Traders are calling it the SAS apocalypse because to them it
0:07
feels like the end of a technology business model,
0:10
software as a service. And we're now weeks into this
0:13
giant sell off and it's only accelerating.
0:16
In this episode I'm going to dig into why markets are panicking,
0:19
why Thomas Reuters can post what seems like such strong earnings
0:23
and yet still lose 20% of their valuation in a single week, and
0:27
what the SAS CE OS themselves are actually saying in response.
0:30
It's important to remember that SAS stocks and technology stocks
0:34
are the only reason really the market hasn't collapsed around
0:37
the world. They are quite literally
0:38
propping up the world economy. I'm going to look at exactly
0:41
what Jensen Huang and Klan as CEO got right and wrong.
0:45
Why I think a per seat pricing model, either business model
0:49
might just be broken and what Sam Altman means when he says
0:53
every company is now an API company.
0:56
And as always in this podcast, I don't think the real answer here
0:59
is in the stock market. And in fact, that's the least
1:02
interesting part of today's discussion.
1:04
I'm going to look at what happened when the creation costs
1:06
collapsed TO0IN music and photography and how the
1:09
industries didn't die, but the business models themselves broke
1:13
and then were rebuilt. And I'm going to argue that I
1:16
think the same is going to happen in SAS.
1:19
This is in the leap for Jack Horton.
1:20
I hope you enjoy the show. I'm
1:36
going to apologize in advance if you hear sipping of coffee.
1:39
I'm absolutely shattered this morning and just trying to keep
1:43
myself awake. I, I think it's important to
1:47
start with, I guess what's actually happening in the market
1:51
now. I'm not going to spend too much
1:52
time here, don't worry. I know some people probably find
1:54
it quite interesting. I do find it interesting from an
1:57
investments perspective, but really it gets quite boring
2:01
quite quickly and I actually really want to jump into the
2:03
history and the interesting parts, to be honest, to help us
2:07
understand where this could probably or likely could go.
2:10
So recently, about 4-3 or four weeks ago, ACNBC reporter, she
2:14
sat down on a news channel and to see if she could build a $5
2:19
billion software company as an application herself.
2:23
Now, Deirdre Bosa had no engineering background, of
2:25
course. She just opened, Clawed,
2:27
described what she wanted in English, and an hour later had a
2:30
working project management tool that was plugged into her
2:33
calendar in a Gmail. And it surfaced kids birthday
2:35
parties that she'd forgotten about.
2:37
The total cost was probably about $20, maybe in compute
2:40
credits. You know, in that same week,
2:42
$830 billion in market value was lost because of software stocks.
2:47
Salesforce, Work Day were both down more than 40% over the last
2:51
12 months. HubSpot fell 39%.
2:54
Atlassian fell by 35% in a single week.
2:58
Unity, the company that allows people to build games, so an
3:02
incredibly important and big platform, lost 1/4 of its entire
3:06
value in a single day because Google released a new model that
3:10
can literally generate playable 3D World from just natural
3:13
language. This has been called by traders
3:15
the SAS apocalypse, with people calling it essentially, I want
3:18
to say SAS, for all those that don't know, it means software as
3:21
a service, but many people are calling it an apocalypse of
3:25
software as a service, as a business model.
3:27
So paying monthly amounts of money for software and you got
3:31
hedge funds now all over the world shorting, so betting
3:33
against giant companies. And in total, they've bet over
3:38
$24 billion that software stocks are going to start to go down in
3:43
2026 alone. I know we've talked about this a
3:46
number of times on this episode, on this show.
3:48
Sorry. And now the catalyst of this
3:50
latest wave of, you know, panic everywhere was of all things
3:54
illegal plug in. So when Anthropic released cloud
3:57
Co work plug insurance. So you, you might have listened
3:59
to the cloud Co work episode. I recommend you to do that.
4:03
Well, since that episode, they launched something called plug
4:06
insurance. Now plug insurance are basically
4:09
a ability for an agent to follow a process in a really simple
4:12
way. Now, this isn't anything new,
4:13
but it's essentially copied what coders have been doing for some
4:17
time and brought it into, I guess, general knowledge work
4:21
accessibility. And suddenly Co work could do
4:24
contract reviews, NDA triages, compliance checks.
4:28
And that means that suddenly a lot of companies that were
4:31
making a lot of money in that sector suddenly lost a lot of
4:35
money. You know, Thomas, Reuters fell
4:37
18%, LegalZoom dropped by 20%. You know, the list goes on.
4:42
And essentially what we're watching or the market is
4:44
watching is AI companies release what are free tools or free
4:50
additional tools to the quite affordable subscription that
4:54
everyone's using and as a result immediately reprice an entire
4:58
software sector. And it just keeps going.
5:00
This wasn't a blip. It's becoming a major trend.
5:04
You know, perhaps the most telling case study is that
5:05
Thomas Reuters that I've mentioned a few times because
5:08
during their sell off their stock had fallen 20% in a single
5:12
week. And yet the actual earnings of
5:14
the revenue that they had actually grown by was really
5:18
strong. 7% revenue growth, 9% growth in EBITDA profits.
5:23
So the actual money they took home margins were at 44%.
5:27
You know, that's these aren't small numbers, these are really
5:29
impressive. And the CEO came to the market
5:33
and said we're seeing massive benefits from our investments in
5:36
AI and they had planned to scale up new agentic features across
5:40
their platform. So yes, strong results right
5:43
now. So strong revenue results from
5:44
big businesses are meaning very little in the market right now
5:48
because people don't have much belief.
5:50
Because ultimately, and I'll explain why this is happening,
5:53
good numbers today still leaves existential questions for
5:57
tomorrow. Now, again, as I said at the
5:59
beginning of this episode, the software level, I guess debate
6:02
and the market debate. Is software dead or you know,
6:05
isn't it isn't something that really interests me that much.
6:09
You know, it's a really important question.
6:11
It's a big question, but it's pretty obvious.
6:13
Of course, software isn't that. The more interesting question, I
6:16
think, is what happens when the cost of creating something
6:20
collapses to almost 0? Because we've seen this before
6:24
several times in history. And I'll tell you about it
6:27
because I think it's really important to help us understand
6:29
what's actually going to happen. And just remember, this has huge
6:43
ramifications for the global market, for the global economy,
6:47
because tech stocks are predominantly propping up
6:49
markets right now. What happens when the cost of
6:52
creation collapses to almost zero?
6:54
Then let's go into it in 1972. You'll tell this is very fun for
6:58
me to research and and read about and get stuck into.
7:01
The Rolling Stones spent about $2,000,000 and over three years
7:05
recording Exile on Main Street. CS studio time was about $800 a
7:10
day and major label album budgets were about half
7:15
$1,000,000 and and that was before any marketing was
7:17
happening. So suddenly, now, today, you can
7:20
record an album on a laptop basically for free.
7:24
An AI tool like SUNO will generate an entire song from a
7:28
single prompt and again for free.
7:30
So the barrier to creating any good music or any music at all
7:33
went from, you know, millions of dollars to basically nothing in
7:36
40 years. And obviously the music industry
7:40
hasn't died. It completely restructured.
7:43
And the story about how it restructured is particularly
7:47
relevant and interesting for this discussion.
7:50
So US recorded music revenue actually peaked at about $14
7:53
billion in 1999. Yet 10 years later, in 2009, it
7:58
it collapsed completely to about 6 1/2 billion.
8:01
You know, CD sales lost 76% of their volume and their revenue.
8:05
The creation side of the business was completely
8:07
destroyed. And obviously everybody was
8:09
panicking, just like what's happening in the technology
8:11
sector today. But then something quite
8:14
interesting happened in 2009, revenue not just recovered, but
8:18
actually exceeded that previous peak.
8:20
By 2024, US music revenues actually hit 17 1/2 billion
8:26
dollars, which is the highest in history.
8:28
And streaming actually had accounted for about 85% of that
8:32
total number. And meanwhile, live events were
8:36
making about $23 billion in revenue.
8:38
So from concerts and from meet ups and clubs with 150 million
8:43
fans attending about 50,000 events in the US alone.
8:47
So yeah, the industry didn't die.
8:48
The value just shifted. It migrated from creation to and
8:53
licensing to distribution and live experiences.
8:58
And Spotify obviously captured a huge amount of the revenue that
9:01
most of the labels lost. And you had Taylor Swift's EUR
9:05
tall last year, gross about 2 1/2 billion dollars.
9:09
Because the one thing you cannot truly digitize is an amazing,
9:12
you know, live experience. And we have the exact same
9:14
happen in publishing. You know, in 2005, about 280,000
9:18
new titles were published in the US Yeah, by 2021, that number
9:22
had crossed about 3 million. So from 282,000 to 3,000,000,
9:27
tenfold increase, yet the actual total revenue from the book
9:32
industry hadn't moved at all. And the only real ones that made
9:36
any money of course was Amazon. And that's because they own
9:39
distribution. Photography was the same stock
9:42
images made a huge amount of money from owning images and
9:46
then licensing that to companies and photographers for money.
9:50
And obviously when AI image generation arrived, everybody
9:53
completely panicked. But actually Shutterstocking
9:56
out, earning record revenue from not selling photos, but
10:00
basically licensing it's imagery for AI companies to train their
10:04
AI language models on. Suddenly it's training data that
10:07
is the value again, value is shifting.
10:10
So yeah, every time the cost of creation collapses to 0, we see
10:14
a very similar pattern play out. You know, the volume of that
10:18
thing explodes per unit economics.
10:20
So the value of selling each unit collapses because the thing
10:24
isn't scarce anymore, it's everywhere and therefore the
10:27
industry must completely restructure around whoever
10:30
controls distribution and what can't be commoditized.
10:33
The thing itself never dies, just the value people place on
10:38
it economically changes. And this is mirrored in a over I
10:42
think 160 year old economic theory which is the Gevons
10:45
paradox. And essentially, this is when a
10:48
thing is made massively cheaper and accessible, the number of
10:51
applications built with that thing increases.
10:55
So for example, we saw this when coal became affordable.
10:58
Suddenly there was an explosion of factories in the industrial
11:01
era. And now we can apply this to
11:02
software. If AI can make it 10 times
11:05
cheaper to create technology, we won't use a 10th as much.
11:08
We used 100 times more of it. However, the Chevron's paradox
11:12
also tells us what's going to happen to many companies in this
11:15
space. Because more total consumption
11:18
didn't mean that every existing producer will suddenly get
11:22
richer and be successful. When coal got cheaper, Although
11:25
total consumption of coal went through the roof, many, many
11:30
individual coal producers at that time completely went out of
11:33
business. And so when recordings, we saw
11:36
this in the music industry when recording costs collapse.
11:39
Although more music existed than ever and this is still the rule
11:43
today, most musicians earn less than ever before.
11:47
You know, the pie gets bigger, but the slices are also much
11:49
thinner. And Ben Thompson put it well in
11:51
his structure E piece, which is a a fantastic place to go learn
11:55
about what's happening in the stock market.
11:58
He said that the real risk I see for software companies is the
12:01
fact that while they can write infinite software for much
12:04
cheaper thanks to AI, so can every other software company.
12:08
The growth story for all these companies is in serious question
12:11
and the industry wide RE rating seems completely justified to
12:16
me. And essentially what he's saying
12:18
is that because every company can produce lots of software for
12:21
the same affordable price, everybody's narrative around
12:25
we're building XYZ to have differentiation that other
12:30
people can't replicate and that we're providing and delivering
12:33
value that no one else could beat us on.
12:35
Suddenly that story feels less compelling because it is
12:39
actually true that a startup or another company that's just as
12:42
big could replicate exactly what you're investing in and building
12:48
really easily. And suddenly the things that
12:51
used to feel less important, so maybe relationships you have
12:55
with vendors or good marketing, good branding, maybe that's
12:59
going to become more important whilst the technology
13:02
technically becomes less important.
13:17
Now I don't want to just, you know, present A1 sided
13:19
discussion here. There are many interesting
13:21
counter arguments to this and I do want to give weight to the
13:24
other side. So yeah, on February 4th, Jensen
13:26
Huang, the for anyone that doesn't know is the founder and
13:30
CEO of NVIDIA. He stood up at the Cisco AI
13:32
Summit and called the software sell off the most illogical
13:35
thing in the world. His thought experiment was quite
13:37
simple. Suppose we are the ultimate AI,
13:42
you know, artificial general robots.
13:44
And if you were a human or a robot, would you use a
13:47
screwdriver or invent a new screwdriver?
13:50
Most of the time you would just use the screwdriver that already
13:52
existed. Would you use a hammer or invent
13:55
a new hammer? And so then you just extended
13:57
that to the digital realm. If we're an artificial general
14:00
intelligence, would you use tools like ServiceNow and SAP
14:03
and big companies that exist, or would you try and reinvent them?
14:07
It's a genuinely good argument. I'm not sure I'm fully persuaded
14:10
by it, but the idea is that AGI would just rebuild every digital
14:14
tool from scratch rather than use the best of what already
14:17
exists. Sounds quite silly when you put
14:20
it like that. And it's not just Jensen Huang
14:23
that says this. The SUS CEOs themselves aren't
14:25
panicking, or or at least they're not publicly panicking.
14:28
So Bock's CEO Aaron Levie argued that the narrative and his I
14:32
quote sometimes and somewhat misunderstands this idea of
14:36
where companies tend to spend their resources and their time
14:40
and energy. His pitch basically is that, you
14:43
know, SAS companies are going to build agents on top of their
14:46
existing products and make their SAS subscriptions just as
14:51
valuable. Therefore, AI is an opportunity,
14:53
not a replacement. And he called it the most
14:55
exciting moment in the last 20 years of technology and cloud
14:59
software. You know, Salesforce CEO Marc
15:02
Benioff said the exact same. He said the agent force, so
15:06
their product for building agents is the fastest growing
15:09
product in their history. And he points to the fact that
15:12
they own customer data and that is their Moat.
15:15
And then there's also the enterprise argument, which is
15:17
that large organizations don't run on small little apps.
15:21
They run on decades of layered systems and you know, data
15:26
warehouses, custom services, very shaky integrations,
15:30
compliance being really important and everyone being
15:33
very worried about compliance. And you know, markets can move
15:36
on what they think is true, but that doesn't mean enterprise
15:41
businesses and architecture can move that fast.
15:44
You know, they are going to be quite risk averse to just use
15:48
small little apps spin up by Claude in a minute.
15:51
Those timelines for change is very different.
15:53
And finally there's also the market based counter argument,
15:58
because actually you can kind of point to this almost logical
16:01
contradiction with everybody getting very scared because for
16:06
the technology sector to collapse it requires 2 mutually
16:10
exclusive things to be true. At the same time, the AI CapEx
16:14
is essentially deteriorating, the companies are spending a
16:18
huge amount of money on AI and the profits aren't clear, which
16:21
is what a lot of people say. Whilst at the exact same time AI
16:26
being so effective and successful that it's spreading
16:29
everywhere and reshaping how technology works and therefore
16:32
makes technology worthless. And you could argue that two of
16:36
those things at the same time make no sense.
16:39
So therefore, this kind of negative bear case that we
16:43
should all take money out of the stock market of technology
16:45
stocks is a little bit illogical now anyway, I think where I land
16:49
on this is every single company is at risk of being replaced by
16:55
an AI native competitor who offers an infinitely better
16:59
experience, a faster experience, and one that solves problems and
17:04
for a much more affordable price.
17:06
Because ultimately, yes, things take a long time to change.
17:10
I think it's worth remembering, you know, the BlackBerry wasn't
17:14
protected, Kodak wasn't protected, Blockbuster wasn't
17:18
protected, or shall I say it was protected until it suddenly
17:22
wasn't. The timeline might have been
17:24
longer than, you know, maybe people like myself or others
17:27
predict. It's almost always the case, but
17:30
it's always much shorter than the big incumbents, those that
17:34
are at the top and have all the money and power.
17:37
It's always shorter than they assume.
17:39
And then suddenly all this delayed disruption tends to
17:42
arrive at once rather than suddenly, just gradually,
17:45
because by the time everybody realizes it's already changed,
17:49
it's kind of too late. Anyway, so I think from here I
18:04
think it's going to be really interesting.
18:05
You know what, what therefore does software become?
18:08
Because this is the really interesting question, not
18:11
whether software stocks recover or survive.
18:13
I think what software becomes on the other side of this massive
18:18
change is something that I'm really interested in.
18:20
So first, the business model is completely breaking.
18:23
So traditionally most technology companies sell on users, or a
18:27
lot of them do. Suddenly you are buying software
18:30
for humans to use. But if you're buying it for
18:32
agents to do the work, therefore if your product is charging by
18:36
the user, you're selling essentially a tax on
18:38
productivity. And even if companies keep their
18:41
SAS subscription, and many, many will, if they're slashing
18:45
headcount because AI let's 10s of people, if they're slashing
18:49
headcount because let's say AI might let just 10 people do the
18:52
work of 100, they don't need 100 seats on an application anymore.
18:57
So that per user pricing model that a lot of the SAS industry
19:00
runs on faces a massive existential question.
19:04
And when CEO of Open Eye Sam Altman was asked, you know, is
19:08
software dead? He responded a little bit
19:10
differently, which is every company is now an API company
19:14
whether they want to be or not. And what he's basically saying
19:18
is that interfaces are no longer important.
19:21
It's the outcomes an agent can perform for people as a result
19:25
of the APIs companies make available.
19:28
So essentially letting an agent use their system regardless of
19:33
what interface they're in. And most likely he wants
19:35
everybody to be in the ChatGPT open AI interface and every
19:39
other company in the world becomes I guess a back end
19:42
system. So you can speak to your Spotify
19:44
for example through open AI. But if that becomes true, the
19:48
value is not the interface and mindset.
19:49
We've been saying this for quite a long time.
19:52
Every chat bot looks the same, you should do a good job of it.
19:56
But actually what's really important, it's actually the
19:58
actions that an agent can perform on behalf of users using
20:02
your APIs. And APIs compete on price and
20:05
functionality, not on complex UI.
20:08
Now this connects to the second thread.
20:09
What happens when AI chooses the tools?
20:11
Jensen Huang's screwdriver argument assumes that a human is
20:15
picking a tool off the shelf based on some form of brand
20:18
familiarity and trust. But that's just not how our AI
20:21
agents will work. They'll select the optimal tool
20:23
based on, you know, measurable data costs, speed,
20:26
functionality. And those criteria are changing
20:29
all the time. Doesn't care about UI, doesn't
20:31
feel the switching cost friction.
20:33
There's no brand loyalty. If a better option just shows up
20:36
tomorrow, it'll just switch. Now if this becomes true and we
20:40
actually delegate tool choice to an agent, which I'm unconvinced
20:44
by, but it's a potential interesting discussion.
20:46
Certainly SAS business models again face the potential
20:50
collapse. Third, the quality question.
20:52
One of the comments that hits hardest in this whole debate is
20:56
essentially most of these SAS products weren't that good.
20:59
It's hideously broken a lot of the time, like really bad UX and
21:03
bugs and functionality that's obviously missing.
21:05
And, and he's right. If anyone's used enterprise
21:07
software on a day in, day out basis, they'll probably all
21:11
share the same thought, which is I want to RIP my hair out.
21:14
The bar for quality has already been low.
21:16
So when suddenly they face this existential threat, maybe we'll
21:21
finally get software that's really good because they have to
21:23
compete on quality for the first time in history.
21:26
But therefore, it might not be the death of software.
21:27
It might be the birth of software that is actually good
21:31
and doesn't make me want to throw a laptop out the window.
21:33
And 4th, and this is something I'm I'm seeing consistently as a
21:37
pattern, temporary software. So software that doesn't exist
21:40
as a permanent product that you'd subscribe to.
21:42
It's software that's generated for specific tasks, used once,
21:45
twice, maybe three times, and it's thrown away.
21:48
Software as a verb, not a noun. You know, developers and
21:51
analysts are already seeing this pattern everywhere.
21:53
You you don't buy a project management tool.
21:56
You might describe what you need for a specific project and an
21:58
agent assembles it. You don't subscribe to Analytics
22:02
dashboard. You ask a question and the
22:04
dashboard materializes around them.
22:06
This is what Mindset enables companies to offer to their
22:09
customers, essentially a chat interface with loads of
22:13
components or widgets that an agent can pull together for a
22:17
person and then they could go on with their day once they've used
22:19
it. You know, even our VP of sales,
22:22
non developer uses this for key deal review.
22:25
So instead of using Hoopspot to go through which deals are we
22:29
looking to close, he creates a app which is essentially a
22:33
report that's interactive and can leave notes on it.
22:35
Now, it wasn't connected to a back end, so it was quite
22:37
temporary, but that's a problem that's going to be fixed by
22:40
someone like Claude very soon. But everyone could just describe
22:43
the task they're trying to complete, and it's done.
22:59
To conclude, I think the mistakes many people are making
23:03
here when we say soft is software dead is treating this
23:06
as some binary thing. You know, software obviously
23:09
isn't dead. Software as we know it might be
23:12
dying. We're going to be surrounded by
23:14
a dramatically different landscape with a lot more
23:18
software than ever before. You know, the Chevron's Paradox
23:21
guarantees that when something gets cheaper and easier to
23:24
create, we don't use less of it, we use more of it.
23:27
But the software era of the last 10 years, which is that specific
23:32
economic arrangement of packaged products as technology charging,
23:38
you know, an amount of money per user per seat with annual 12
23:44
month contracts that you have to sign with Moat built in to the
23:49
fact that their UI is complex and difficult to navigate might
23:52
just be over. Software has been valued
23:55
typically at 30 to 35 times profit for the last 15 years.
23:59
And a lot of this is because the market assumes that these big
24:02
companies have 10/20/30 potential years of being market
24:08
leaders. But if AI contracts that
24:12
horizon, so makes that period instead of 30 years suddenly
24:15
five years or eight years or less than 10 years, then we'll
24:19
reprice those technology companies.
24:22
The valuations won't be 30 to 35 times, they might be 510 or 15
24:26
times. But remember, the music industry
24:29
in 2024 is still bigger than it was in 1999.
24:33
Now if you're ACD manufacturer in 1999, you know that start
24:37
means very little because you're out of business.
24:39
But the software industry in 2035 will I guarantee almost
24:43
certainly be bigger than it is today.
24:46
The question is really whether it's bigger in a way that
24:50
companies currently selling can capture, or whether the value is
24:54
already migrated and changed and most of those companies are
24:58
going to die out anyway. I hope you enjoyed today's
25:02
episode. Thanks for listening and I hope
25:05
to see you next week.