AI App Crisis, OpenAI Does Math, Big Nvidia Deal

Welcome to the podcast.

I'm your host, Jaden Schaefer.

Guys, today is my 30th birthday,

but I had to record a podcast because there

was some crazy stuff happening.

Number one, there's a bunch of research and data

coming that is showing AI-powered apps

are really struggling with long-term retention.

Also, Chat Gbt can now create interactive visuals

that are going to help you understand math and science,

which Google was kind of doing something similar.

It's going to be really cool to see Chat Gbt do this.

And third, thinking machine labs

has just created a massive compute deal within video,

which is pretty exciting for a company

that has such a legendary background

and has raised so much money.

So we're going to get all of these stories today,

but before we do, I have to say a huge shout out

in the last couple of days.

I've asked people for my birthday.

If you could leave a rating and review, if you haven't already,

I want to read the most recent review that someone dropped.

This is from Eating Crab yesterday, he said,

just wanted to say thank you for the podcast.

I don't have very much time in my day being a full-time student.

I'm working full-time, but I have a huge passion

for AI being able to keep up with your podcast,

helps me keep in the loop.

I appreciate it, keep it up, and happy birthday.

A huge shout out to Eating Crab.

Thank you so much for the review.

Guys, today is my birthday.

I'm turning 30 before I go through a midlife crisis.

If you guys could do me a massive favor for today,

and please leave a rating review if you haven't already.

It would be the greatest birthday present of all time.

I will be eternally grateful over on Apple or Spotify

or wherever you get your podcast.

I know it's usually annoying,

but today's my birthday, so if you've ever appreciated

that podcast in the past or today,

it would be greatly appreciated to drop a review.

All right, let's get into the episode today.

So the thing that I think is really interesting

is this idea right now that all of the AI-powered apps

are really struggling to keep long-term,

people engaged, long-term retention on the app.

So I think there's a couple of problems with this,

as someone that has built AI-powered apps in the past,

and as someone that is actively working in AI startup

and AI box and a company,

I can understand where a lot of this challenge is,

and that is I think with AI coming out

in the power of AI being so incredible,

I think we definitely had a really big wave,

especially in the last couple of years

where there was a lot of concepts of what AI could do

and would be able to do,

and a lot of people I think overhyped

or oversold their apps,

and I think that's gonna be the primary driver

of lower tension.

In addition, I do think that right now,

I try probably 10 times as much software

as I have over the last five, 10 years

working in the industry,

and so I think right now we just try so much more

and then we kind of settle on what works best.

I think if you're a developer

and you're creating a tool with AI in it,

you have one shot really for someone to go try your tool

and for it to wow them and for them to be impressed

and be like, okay, I will keep this

as part of my long-term tool belt of the tools I use.

If they try it and it flops,

there's a bunch of tools from big companies

that I've tried in the past,

they've flopped and I haven't gone back.

I think one of those examples would be something

like Runway for Video.

This is a platform that I tried a lot.

In the early days, it wasn't that great.

And I mean, you have to give them a huge kudo

for being first,

but I never really got back that platform

and then Soono came out

and a lot of these other video generations

that you have Higgs Field

which has a whole bunch of models on there.

And I tend to just use more of those types of tools today

than going back to some of the OG video tools.

I think this is kind of a trend you'll see with a lot of,

I know it's kind of like a random story

from my experience,

but I think you're gonna see that a lot.

So there was a recent report that came out of Revenue Cat

and they showed that the subscription infrastructure

for more than 75,000 different developers

and they kind of analyze it

because they power all that subscriptions.

And by the way,

these are kind of my favorite reports.

Mercury, a SaaS kind of bank, SaaS focused bank

is an awesome one.

They do kind of a state of AI every year

where they show the top AI companies

that people are actually using

and actually have subscriptions to.

So it's kind of cool to see Revenue Cat do something similar.

What they found though,

these are just awesome reliable places

because you know exactly where money's actually being spent.

But in any case, they found that while AI apps monetize

really quickly,

they really struggle to keep users around.

And I'll even say for my own startup, AIbox.ai,

when we first launched,

there was a lot more bugs at the beginning.

I mean, with anything and a lot of different features

that we didn't have.

And I think our churn was pretty high.

When we first launched,

I'm pretty proud of being able to pull that churn rate down

and get people to stick around a lot longer.

But it was a lot of work for us to achieve that

and to be able to bring that down.

I think a lot of other people struggle with that too.

According to their 2026 state of subscription apps report,

AI apps experienced significantly higher churn

compared to traditional apps.

So this is interesting, right?

Like it's not just like,

oh, people try more software than they use less today.

Well, people actually still keep a lot of their OG apps

that they have and OG subscriptions.

But it's the new AI ones

that are just trying out this new buzz thing.

I'll look at you do this crazy cool thing

and make an image of me looking like XYZ,

you go try it and then you kind of dump it.

So according to the report,

they analyze more than one billion

in app subscription transactions.

And that was about $11 billion

in annual developer revenue.

I mean, based off of their scale and their ecosystem,

I think this is a pretty good indicator.

They have like iOS, Android, and web apps.

And what's interesting to me is that, you know,

despite like, I think a lot of the hype around AI,

most subscription apps are not built around it.

Only 27% of apps analyzed according to their report

were categorized as AI powered.

About 72% of those are just non-AI apps.

So a majority of apps, and this is interesting

because I think we see a lot of the bigger players

like immediately injecting AI in.

But I think a lot of the smaller apps

and companies are like, well, if we don't need it,

maybe there's not a reason to just, you know,

bolt something on that's not necessary.

And that's making up, you know, almost 73% of all apps.

I think what is obviously pretty clear

is how fast AI adoption is accelerating

about one in four apps now marketing themselves,

are marketing themselves as AI driven.

So a lot of those are not necessarily going to be,

you know, chats you to your Gemini,

but they might just be, you know, an app

that has some AI features in their product experience.

Which I think a lot of people should be

and could be experimenting with.

And there's a lot of things that AI can do

just inside of all traditional SaaS.

I do think this is a smart idea for most any apps.

I think definitely the categories that are adopting AI

faster than others would be things like photo and video apps.

Those are kind of the top according to what they're seeing

in their analytics.

They say 61% of photo and video apps

are incorporating AI features.

They say gaming is on the complete opposite extreme.

Only 6.2% of gaming apps are using AI in their code offering.

There's a bunch of other like pretty low adoption categories,

including travel, which is only 12.3%,

which is hilarious because I swear every single demo

we see from chat, you could be in Gemini.

It's like, check out the new advancements we made.

Like, it's gonna make planning your travel itinerary

like a thousand times faster.

Like, just say you wanna go to Greece

and it's gonna give you like a 15 day itinerary

with everything you need to do every hour.

Now, I don't know how often people

are planning the travel itinerary.

I don't know why this is the one demo we get stuck on

with everyone and I apologize for my voice

on the demo voice there.

But it is just one of the things that drives me crazy.

And it's so wild to see

that only 12% of the travel apps are actually using this.

Well, every single AI company is like basically using

this category as the main demo of a use case

and it just doesn't turn out to be that useful

or I guess the demand isn't there.

Business also has a 19.1% rate of having AI inside

of the apps that the business category.

I think where things get a lot more surprising

is customer retention.

So across both monthly and annual subscription plans,

AI powered apps consistently were underperforming.

Just regular non-AI apps.

After 12 months, AI apps had about a 21% retention rate

for their subscriber.

So if 100 people subscribe on day one, 12 months later,

only 21 of those will still be subscribed compared

to over 30%, 30.7% for non-AI apps.

So 10% higher if you don't have AI embedded in there.

Now, I think there's going to be some things that obviously

skew that which is that a lot of these AI apps

are kind of a new interesting use case

and people are going to be trying them out

for the first time.

I also think that there's a lot of hype

and if the AI can't do exactly what you want perfectly,

you're going to move on.

And I think those capabilities will come back in the future.

So people may retry those same apps in the future

and they'll stop being called just like an AI app.

Like there's not going to be a buzzword.

It's just going to be like an app that does XYZ,

does use AI, but people don't really care.

It just does it correctly.

And I think then you'll see the retention rates higher.

On a monthly basis, I think there's also a pretty big gap

between these two categories.

AI apps have a 6.1% retention.

Non AI apps have a 9.5% retention rate just month to month.

I think one of the other areas where AI apps

are performing a little bit better

is in weekly subscriptions.

Retention is at 2.5% compared to 1.7%.

I mean, really that's just showing you that I think

a lot of people are trying these apps.

Weekly subscriptions is not a very common thing.

And I have actually seen this with a bunch of AI apps.

I saw I see weekly subscriptions with people

that have a tool that you don't really need it

for a super long period of time.

So they try to hit you multiple times in a week.

It's basically my least favorite subscription amount

of time to re-subscribe.

And it's when people are like, look, it's only $2 a week.

It's like just say $10 a month or something like so annoying.

In any case, part of the churn that we're seeing right now

I think is going to be obviously just how fast AI is going.

All of the experimentation happening in the industry,

I think you're going to see metrics like AI apps

have 20% higher refund rates than non-AI apps.

The median refund of 4.2% is, you know,

you can compare that to 5.3%.

At the high end, I think the difference is even more pronounced

with AI apps seeing refund rates as high as 15.6%

compared to 12.5% for non-AI apps.

And according to what revenue cat is saying,

basically volatility is coming from some big issues

about value product experience, long-term utility.

And I really just think a lot of this comes down

to over-promising and under-delivering

what the AI is capable of doing.

And I see this in so many areas

because I'm in marketing them in AI.

So clearly this is a problem I think we see in the industry.

I think we should probably normalize, you know,

being able to under-hype your app,

but it being super useful and people just use it

without, you know, having to oversell all of its capabilities.

I think AI apps right now also monetize downloads

significantly more effective overall.

Median download monetization is at about 2.4%

for AI apps versus 2% for non-AI apps.

So that is interesting.

And I think AI apps also generate higher realized lifetime value.

So here we are, you know, talking about,

oh look, regular apps versus AI apps,

AI apps aren't able to keep people subscribed as long,

but they're getting a lot more money out of people, right?

On a monthly basis, AI apps produce a median real lifetime value

of $18 per user compared to $13 for non-AI apps.

It's actually closer to $19 and like $13.50 for non-AI apps.

So I mean, that's a pretty big step up.

People are paying more for AI apps.

Obviously the costs of those are higher on an annual basis

though, I think it gets even bigger.

So for annual AI apps are reaching $30 versus $20

of real lifetime value for their users.

I think if you look at all of this together,

basically the data, like the pattern that I see in this

is that AI features are going to help apps monetize

really quickly, but sustaining that long term

is going to be the challenge and making sure your product

is actually useful, delivers on all the promises is harder.

It's totally possible, but I think there's also

a lot of competition and even for an app like,

you know, ChatGPT for example,

that was like the bell of the ball for forever for years.

Number one, and then all of a sudden,

Gemini comes out and Claude comes out

and they have kind of some new capabilities

and move reasons why you'd want them.

So I think that we're just going to see a lot of competition

and it's going to be an interesting space.

It's no one's, you know, it's not like anyone

has this completely cornered.

Okay, switching gears for a second,

I want to talk specifically speaking of ChatGPT,

a new feature that is going to support

inside of ChatGPT and inside of the app,

obviously we can see that we got to try to make

these things more useful for users.

And so this is the latest thing ChatGPT is doing,

obviously to try to keep their turn down.

So this week ChatGPT has just introduced

dynamic visual explanations inside of ChatGP.

This is basically a feature that is going to let you

have make mathematical and scientific concepts

a lot more interactive.

So rather than just, you know,

getting like some sort of text explanation

or maybe like a static diagram,

this new feature is going to let you manipulate variables

directly and then you can watch equations update

in real time inside of ChatGPT.

This is very cool.

So an example of this is like if you were exploring,

you know, the Pythagorean theorem or something,

users could basically you could go

and adjust the sides of a triangle

and then immediately you'll see how the hypotenuse

is changing.

The feature right now is supported

in more than 70 different math and science concepts.

It includes compound interest, exponential decay,

linear equations, columns law,

ohms law, kinetic energy, hooks law.

It's, honestly, it's pretty cool if I'm not,

if I'm telling the truth, I think that right now

if you're able to kind of turn an explanation

into this sort of like interactive module,

the feature is going to shift from maybe just the tool

giving you these really simple answers

to actually helping users.

And I mean, hopefully students and others

explore how the concept actually works

and get kind of a deeper insight

and understanding of the problem.

So honestly, you know, I remember when AI came out

and everyone said it's gonna make everybody dumber

and we're gonna just, you know, outsource our brains to AI.

But I actually think it's an incredible tool for education

that's going to make us smarter

and we're going to be able to learn more.

Open AI says that more than 140 million people

already use ChatGPT every single week

for math and science help.

I think it's over 900 million people

weekly just for general use.

But, you know, 140 million just for math and science

is a huge chunk of that.

And so I think this is obviously something

that's been very tricky for a lot of people.

It's hard to get access to good tutors

and this is an awesome opportunity, I think,

for a lot of people.

Other companies are definitely experimenting

with some similar approaches in 2025

kind of at the end of last year.

Gemini introduced some interactive diagrams

within their own AI assistant as part of kind of an effort

to get more into education.

I think I did a podcast on it back at the time.

I think the race right now to build the next generation

of AI infrastructure is going to be interesting.

We have all these new features,

but all of these new features have to be powered

by infrastructure.

They've got to be powered by more compute

as these tools just get more and more intense.

And on that note, Miriam Maraudi's startup

Thinking Machine Labs has just announced

a multi-year strategic partnership with Nvidia

to deploy large-scale computing systems.

And this is actually going to start in 2027.

So not this year, it's interesting.

I mean, it's kind of crazy.

Miriam Maraudi and then we have,

super-safe intelligence as well,

kind of these spin-offs from some of the top brass

over open AI when they came out

and they raised like a billion dollars.

Their startups didn't launch something immediately

although I do believe that Miriam Maraudi

Thinking Machine Labs does have tinker,

I think is their product.

They do have a product out there.

But I think they got a lot more on the pipe

and they're building all of these kind of compute

partnerships that are starting, you know,

not this, like they got out maybe last year.

They didn't really put out a ton last year

and they have all of this year, they're still working.

And these compute deals aren't rolling out until next year.

So you could expect that whenever their products

are scaling is going to be in the future.

This agreement in particular includes deploying

at least one gigawatt of Nvidia's Vera Rubin AI systems.

I mean, honestly, even just going to sign in a gigawatt deal,

it's a lot of confidence that their product

is going to be incredibly useful.

I think this is one of the company's newest architectures

and Nvidia is also making a strategic investment

in Thinking Machine Labs,

which has already raised more than $2 billion

since it was founded last year.

It's valued at over $12 billion.

Thinking Machine Labs is focused on building AI models

that are designed to produce more reputable

and reliable outputs.

And they released their first API product tinker last year.

So right now, this partnership is showing,

I think basically a bigger trend in the industry,

AI companies are really having to be very aggressive

in how they compete for access to this computing power.

Jensen Lang, CEO of Nvidia, he predicted

that the industry would spend $3 trillion to $4 trillion

on AI infrastructure by the end of the decade.

So so much money is getting put into this industry.

I think the exact value of Thinking Machine Labs

and their deal that they're doing with Nvidia here,

that wasn't all disclosed.

But I think the massive compute agreements

are just becoming more and more common

for some of these big companies.

Last year, for example,

Open AI struck a $300 billion compute partnership

with Oracle.

And so yeah, obviously this is something

that's not slowing down.

If you look at all of these together,

I think the industry right now is experimenting rapidly.

They're just trying to put out tons of different products

and we see that from kind of the state of AI apps

I was talking about earlier.

But I also think the AI is starting

to reshape how apps monetize, how people learn,

how infrastructure is built, how long-term winners

are likely going to be, you know,

people that are moving beyond just the novelty

and they're actually delivering a really durable value

to the users, right?

And I think it's important when even a company

like Thinking Machine Labs and a lot of these others

has to think like when you build a product

in Open AI with their latest, you know,

kind of math and science feature,

like when you build a tool, when you build a product,

make sure it works really good on launch

and then you're gonna be able to keep your turn up

and then all of these long-term infrastructure deals

you have are not gonna be wasted

because you're gonna be able to keep all of your users

using the product.

So this is an interesting time in the industry.

A lot is going on.

I'll definitely keep you up to date on all of it.

If you wanna try all of the AI models I talked about

on the show, make sure you go check out

AIbox.ai, you can access to over 40 of the top AI models.

All in one place, you can chat with them.

We have some exciting new features dropping soon.

You can check it out.

Link is in the description to AIbox.ai.

And everyone, remember, today is my birthday.

I turn 30 today, the number one present

and thing in the entire world I would ask for my birthday

is if you could leave a rating review on the podcast,

it would mean the world to me, I would be so thrilled.

So if you haven't already, leave a review

and I will be eternally grateful on my birthday.

All right, hope you guys all have a fantastic rest of your day.