July 10, 2025 • 10 mins
In this week’s monologue, Ed Zitron walks you through how Anthropic jacked up prices on its largest customer Cursor - and how it’s a sign that the Subprime AI Crisis has begun.
Premium Newsletter covering this:
https://www.wheresyoured.at/anthropic-and-openai-have-begun-the-subprime-ai-crisis/
Bluesky thread summarizing: https://bsky.app/profile/edzitron.com/post/3ltfabnj3ec23
YOU CAN NOW BUY BETTER OFFLINE MERCH! Go to https://cottonbureau.com/people/better-offline and use code FREE99 for free shipping on orders of $99 or more.
---
LINKS: https://www.tinyurl.com/betterofflinelinks
Newsletter: https://www.wheresyoured.at/
Reddit: https://www.reddit.com/r/BetterOffline/
Discord: chat.wheresyoured.at
Ed's Socials:
https://www.instagram.com/edzitron
https://bsky.app/profile/edzitron.com
https://www.threads.net/@edzitron
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:02):
Zone Media. Hello and welcome to this week's Better Offline Monologue.
I'm your host ed zetron over on my premium newsletter.
This week I reported out a troubling situation with AI
powered coding environment Cursor maybe, a company called any Sphere
(00:23):
who unexpectedly changed their pricing in the middle of June
as well as adding a two hundred dollars a month
ultra subscription that well pissed everyone off and pissed off
a lot of people for many different reasons. In fact,
for some history, Cursor is or was the bell of
the ball of the AI industry. They had five hundred
million dollars of annualized recurring revenue, which means their highest
(00:43):
month times twelve, making them the first high revenue startup
outside of anthropic and open AI to actually show growth potential.
There really is no one else like them out there
at this number. I mean as Alpha sense four hundred
million according to the AI data based from the information,
but everyone else is powering around one hundred two hundred
million arr and raising far more money. Everyone was really
(01:05):
excited about Cursor. They were growing fast. Everyone loved them,
and it was because their product allowed you to just
build software just by typing stuff, and engineers liked it
and regular people liked it. Not saying it was good
or anything, just that people liked it, and they liked
it because you gave you kind of unrestricted access to
ward these models to build all this stuff. Now, what
(01:25):
if I told you that perhaps they had grown to
that five hundred million ARR number using techniques like allowing
people to spend way more than they were paying for
the service. Yeah, that's exactly what happened. Previously, Cursor's pricing
was pretty simple. You had five hundred fast completions, meaning
that you could ask Cursor to do something using model
like Claudes sonnet for or Opus four, and it deduct
(01:47):
requests from your allowance. The amount deductive was based on
the model, with some models costing less than others, such
as Anthropics claud for Opus costing more. Once you were
out of those fast requests, Cursor would then put you
in the slow lane, the unlimited slow lane, I should add,
which nevertheless would let you use premium models, but you'd
be put in the back of the queue and at
peak times could be waiting some time before you get
to your completion. You could also select auto and Cursor
(02:09):
would select whatever was available the best model for the
job at the time, never want to turn the best
over to the company though. But on June seventeenth, twenty
twenty five, Cursor launched that two hundred dollars a month
ultraplan along with a new and confusing pro package It's
very muddied. A new wording around said twenty dollars a
month plan and how it would work going forward, claiming
(02:31):
that they would be rolling out changes to make it
more generous, but actually changing pricing to reflect new wording,
offering unlimited agent requests and no real explanation as to
what that meant, and no further mention of any request
limits or what those requests or rate limits might be.
The reality was, of course, far grimmer. Cursor users have
found themselves heavily rate limited, especially on Anthropic models, but
(02:51):
the so called unlimited agent requests mostly pushing towards Cursor's
on frontier model, which users claim is nowhere near as
effective as model SMOP and AI Andthropic. On June thirtieth,
Curser made another change to their product features, changing unlimited
agents to extended limits on agents, further money in pricing
on a product that was once renowned for its simplicity
and lack of limits. So this isn't good, is it.
(03:15):
Silicon Valley's favorite coding startup has in the last few
weeks completely changed how its customers interact with its product,
both degrading the service and making it far more expensive
in the process. They also allow you now to use
usage based pricing. Now, this is the funny one. If
you pay twenty dollars a month on Cursor, they will
guarantee you twenty dollars a month of AI compute at least,
(03:35):
so they're literally giving money away. They add a twenty
percent fee on top of compute. But wow, so you're
making twenty percent morgin. Now you're fucking not. They're losing
money on everything. I'm sure of them. But it's strange.
It's strange, and it's bad. But this is also textbook
and shitification. Corey doctor O's term for when platforms offer
a high quality product to gain a large user base,
usually through convenience or great value. Then the grade the
(03:57):
service over time to make more money as a means
of maximizing value shareholders or making money themselves. It's also
part of my rot economy. Thesis. The growth of all
costs thinking has dominated the tech industry, and I'd argue,
thanks to the proliferation of business idi it's controlling everything
that they've drained some of the logic behind in certification away,
because very few of these companies actually have a plan
for sustainability, let alone profitability. Generally, in certification gets people
(04:20):
through the door and makes the service totally impossible to avoid,
makes it essential. That's really difficult to do with a
paid software product like Cursor, because AI compute is so
fucking expensive. It's so expensive, so Cursor has decided to
en shittify without making sure that they have something essential.
(04:40):
But the thing is, I had texted Cory about this.
I think this actually is something quite different. I think
this is the world's first chain in shitification. My belief
is simple. Any Sphere, which is the company that makes Cursor,
by the way, is despite getting nine hundred million dollars
in funding in early May, running out of money, or
at least believes that continue to operate its business in
(05:01):
the way it did less than a month ago would
cause it to do so. But why the sudden changes,
Why the knife in the heart of their customers just
after raising nearly a billion dollars. It's simple, it's a
chain and citification. Anthropic jacked up their prices, and so
did open Ai. On May twenty second, twenty twenty five,
a few weeks after Cursor raised nine hundred million dollars
and Thropic launched both claud Sonet and claud Opus four.
(05:23):
You might say Claude four Opus of Claude force Sonet.
I don't care. But these were two new powerful as
judged by benchmarks made specifically for large language models. And
I should add that the expensive claud Opus four was
more focused on coding benchmarks. This is an important detail now.
Eight days later, on May thirtieth, twenty twenty five, patron
(05:44):
Anthropics API documentation appeared for the first time called service tiers,
adding priority tiers for enterprise stats that didn't want to
and I quote sorry, I mean did want to and
I quote provide a guarantee around the infrequency of server
overloaded errors even during peak time. Hmm, that's not good.
Anthropic service tiers require a multi month upfront commitment on
(06:06):
how many tokens per minute your startup will use, but
also add an insidious charge around prompt casing. Now Prompt
cashing is when when you put something into a model,
such as a code base in the case of a
coding startup, or I don't know a great deal of
stuff about how a model might want to act, you
put it in the cash so that it kind of
like RAM, it just kind of reads off of it
instead of rewriting and rewriting and rewriting. So they've added
(06:30):
a vig to it or attacks whatever you call it,
a toll perhaps, and they're now charging and this is
really fucking insidious, either one hundred and twenty five percent
or two hundred percent of the cost of casing information
that you need to access it more readily. And I
should add this is only if you want the Priority tier.
And another thing, coding startups are extremely prompt cash heavy,
(06:51):
kind of fucked up right, very fucked up, and they're
not the only ones doing it. On June twenty fifth,
twenty twenty five, open Ai also launched Priority processing for
API customers. They claim it's the payers you go to
where you pay higher API prices for and I quote
predictable low latency. However, despite this being payers you go,
the service is only available for their enterprise customers who
(07:12):
have made an up prompt commitment much like Anthropic does,
the difference being they are not trying to tax you
on prompt caching now. Cursor CEO Michael Truell also said
when announcing their agree just two hundred dollars a month
plan that it was only made possible by multi year
partnerships with OpenAI, Anthropic, Google, and Xai. In short, I
think that nine hundred million dollars that Cursor got may
(07:32):
have immediately been handed to or committed to the major
model developers. They got looted baby. And now what makes
this weirder is that Cursor is also Anthropic's largest customer,
with sources claiming that they're such a large customer that
they're taxing anthropics infrastructure and making them run out of GPUs. Now.
I should also add that Cursor isn't the only one
(07:54):
that's recently dramatically changed their prices. Vibe coding startups like
replt and Lovable Bits both have had to do so
in the last month. And I think this is just
the beginning of something really, really, really bad. I believe
that Anthropic did this either as a deliberate attempt to
price gouge its largest customers, and or as a means
of increasing revenue and its money losing software. These changes
(08:15):
that deliberate aggressive and targeted price increases, and they were
timed with the launches of claud for Opus and Sonnet,
which suggests that Anthropics costs have dramatically increased with these models,
so much that simply increase in the cost per million
tokens is in sufficient. It's my belief that the launchers
of both claud Opus and to a lesser extent, son
It have caused en up hevil in Anthropics costs and
compute demands, which in turn forced them to start increasing
(08:37):
costs on their customers. However, Anthropic has I believe, realized
that there is no real way to just increase the
cost of Opus and Sonnet for further on just the
amount of tokens that the customer might use, and that
doing so might push away smaller developers, which wouldn't make
them more money. So they decided to find a way
to specifically exploit the finances of their largest customers coding startups,
(08:58):
in a way that won't be immediately obvious or that
would spook non coding assistant customers. Except I'm a crafty
little fuck and I look at everything all the time.
I do this for fun. I don't do this because
I have to listen to me. Wario Amadee. If you're
listening to this, I'm watching you. I have archive dot
org and a million diet cokes. Anyway, last year I
talked about the subprime AI crisis, where almost the entire
(09:21):
tech industry is bought in on a technology sold at
this vastly discounted rate, and they've heavily centralized and subsidized
it too. And I predicted one major thing that these
model developers would eventually have to find a way to
make their costs work, that they would have to find
a way to crank up costs on their customers, because
otherwise they would just continue burning money. I still think
(09:42):
they are doing so, and I think that's what's happening here.
I think anthropic and OpenAI to a lesser extent of
realized that they need to start making money back on
these fuckers, and they're doing it. Except the fact that
the companies of immediate companies like Cursor and Repler have
immediately had to change their prices suggest that maybe not
one of these businesses makes sense. Maybe this whole time
(10:03):
this thing was unsustainable. Maybe it turns out the general
if AI doesn't have the kind of business returns you
need to run a startup. If only someone had said something. Remember,
look see how fucked up saying Remember but I'm not
editing it out. This is an honest podcast. But seriously, everyone,
remember the pale horses. Rate limits, service interruptions, price increases,
(10:28):
trouble raising funding, and trouble with money. The horses are
drawing nearer and telling you we're coming to the end
of this. I don't know if it will be soon,
I don't know if it'll be next year. But nothing
about this suggests that things are going well.
Zone Media. Hello and welcome to this week's Better Offline Monologue.
I'm your host ed zetron over on my premium newsletter.
This week I reported out a troubling situation with AI
powered coding environment Cursor maybe, a company called any Sphere
(00:23):
who unexpectedly changed their pricing in the middle of June
as well as adding a two hundred dollars a month
ultra subscription that well pissed everyone off and pissed off
a lot of people for many different reasons. In fact,
for some history, Cursor is or was the bell of
the ball of the AI industry. They had five hundred
million dollars of annualized recurring revenue, which means their highest
(00:43):
month times twelve, making them the first high revenue startup
outside of anthropic and open AI to actually show growth potential.
There really is no one else like them out there
at this number. I mean as Alpha sense four hundred
million according to the AI data based from the information,
but everyone else is powering around one hundred two hundred
million arr and raising far more money. Everyone was really
(01:05):
excited about Cursor. They were growing fast. Everyone loved them,
and it was because their product allowed you to just
build software just by typing stuff, and engineers liked it
and regular people liked it. Not saying it was good
or anything, just that people liked it, and they liked
it because you gave you kind of unrestricted access to
ward these models to build all this stuff. Now, what
(01:25):
if I told you that perhaps they had grown to
that five hundred million ARR number using techniques like allowing
people to spend way more than they were paying for
the service. Yeah, that's exactly what happened. Previously, Cursor's pricing
was pretty simple. You had five hundred fast completions, meaning
that you could ask Cursor to do something using model
like Claudes sonnet for or Opus four, and it deduct
(01:47):
requests from your allowance. The amount deductive was based on
the model, with some models costing less than others, such
as Anthropics claud for Opus costing more. Once you were
out of those fast requests, Cursor would then put you
in the slow lane, the unlimited slow lane, I should add,
which nevertheless would let you use premium models, but you'd
be put in the back of the queue and at
peak times could be waiting some time before you get
to your completion. You could also select auto and Cursor
(02:09):
would select whatever was available the best model for the
job at the time, never want to turn the best
over to the company though. But on June seventeenth, twenty
twenty five, Cursor launched that two hundred dollars a month
ultraplan along with a new and confusing pro package It's
very muddied. A new wording around said twenty dollars a
month plan and how it would work going forward, claiming
(02:31):
that they would be rolling out changes to make it
more generous, but actually changing pricing to reflect new wording,
offering unlimited agent requests and no real explanation as to
what that meant, and no further mention of any request
limits or what those requests or rate limits might be.
The reality was, of course, far grimmer. Cursor users have
found themselves heavily rate limited, especially on Anthropic models, but
(02:51):
the so called unlimited agent requests mostly pushing towards Cursor's
on frontier model, which users claim is nowhere near as
effective as model SMOP and AI Andthropic. On June thirtieth,
Curser made another change to their product features, changing unlimited
agents to extended limits on agents, further money in pricing
on a product that was once renowned for its simplicity
and lack of limits. So this isn't good, is it.
(03:15):
Silicon Valley's favorite coding startup has in the last few
weeks completely changed how its customers interact with its product,
both degrading the service and making it far more expensive
in the process. They also allow you now to use
usage based pricing. Now, this is the funny one. If
you pay twenty dollars a month on Cursor, they will
guarantee you twenty dollars a month of AI compute at least,
(03:35):
so they're literally giving money away. They add a twenty
percent fee on top of compute. But wow, so you're
making twenty percent morgin. Now you're fucking not. They're losing
money on everything. I'm sure of them. But it's strange.
It's strange, and it's bad. But this is also textbook
and shitification. Corey doctor O's term for when platforms offer
a high quality product to gain a large user base,
usually through convenience or great value. Then the grade the
(03:57):
service over time to make more money as a means
of maximizing value shareholders or making money themselves. It's also
part of my rot economy. Thesis. The growth of all
costs thinking has dominated the tech industry, and I'd argue,
thanks to the proliferation of business idi it's controlling everything
that they've drained some of the logic behind in certification away,
because very few of these companies actually have a plan
for sustainability, let alone profitability. Generally, in certification gets people
(04:20):
through the door and makes the service totally impossible to avoid,
makes it essential. That's really difficult to do with a
paid software product like Cursor, because AI compute is so
fucking expensive. It's so expensive, so Cursor has decided to
en shittify without making sure that they have something essential.
(04:40):
But the thing is, I had texted Cory about this.
I think this actually is something quite different. I think
this is the world's first chain in shitification. My belief
is simple. Any Sphere, which is the company that makes Cursor,
by the way, is despite getting nine hundred million dollars
in funding in early May, running out of money, or
at least believes that continue to operate its business in
(05:01):
the way it did less than a month ago would
cause it to do so. But why the sudden changes,
Why the knife in the heart of their customers just
after raising nearly a billion dollars. It's simple, it's a
chain and citification. Anthropic jacked up their prices, and so
did open Ai. On May twenty second, twenty twenty five,
a few weeks after Cursor raised nine hundred million dollars
and Thropic launched both claud Sonet and claud Opus four.
(05:23):
You might say Claude four Opus of Claude force Sonet.
I don't care. But these were two new powerful as
judged by benchmarks made specifically for large language models. And
I should add that the expensive claud Opus four was
more focused on coding benchmarks. This is an important detail now.
Eight days later, on May thirtieth, twenty twenty five, patron
(05:44):
Anthropics API documentation appeared for the first time called service tiers,
adding priority tiers for enterprise stats that didn't want to
and I quote sorry, I mean did want to and
I quote provide a guarantee around the infrequency of server
overloaded errors even during peak time. Hmm, that's not good.
Anthropic service tiers require a multi month upfront commitment on
(06:06):
how many tokens per minute your startup will use, but
also add an insidious charge around prompt casing. Now Prompt
cashing is when when you put something into a model,
such as a code base in the case of a
coding startup, or I don't know a great deal of
stuff about how a model might want to act, you
put it in the cash so that it kind of
like RAM, it just kind of reads off of it
instead of rewriting and rewriting and rewriting. So they've added
(06:30):
a vig to it or attacks whatever you call it,
a toll perhaps, and they're now charging and this is
really fucking insidious, either one hundred and twenty five percent
or two hundred percent of the cost of casing information
that you need to access it more readily. And I
should add this is only if you want the Priority tier.
And another thing, coding startups are extremely prompt cash heavy,
(06:51):
kind of fucked up right, very fucked up, and they're
not the only ones doing it. On June twenty fifth,
twenty twenty five, open Ai also launched Priority processing for
API customers. They claim it's the payers you go to
where you pay higher API prices for and I quote
predictable low latency. However, despite this being payers you go,
the service is only available for their enterprise customers who
(07:12):
have made an up prompt commitment much like Anthropic does,
the difference being they are not trying to tax you
on prompt caching now. Cursor CEO Michael Truell also said
when announcing their agree just two hundred dollars a month
plan that it was only made possible by multi year
partnerships with OpenAI, Anthropic, Google, and Xai. In short, I
think that nine hundred million dollars that Cursor got may
(07:32):
have immediately been handed to or committed to the major
model developers. They got looted baby. And now what makes
this weirder is that Cursor is also Anthropic's largest customer,
with sources claiming that they're such a large customer that
they're taxing anthropics infrastructure and making them run out of GPUs. Now.
I should also add that Cursor isn't the only one
(07:54):
that's recently dramatically changed their prices. Vibe coding startups like
replt and Lovable Bits both have had to do so
in the last month. And I think this is just
the beginning of something really, really, really bad. I believe
that Anthropic did this either as a deliberate attempt to
price gouge its largest customers, and or as a means
of increasing revenue and its money losing software. These changes
(08:15):
that deliberate aggressive and targeted price increases, and they were
timed with the launches of claud for Opus and Sonnet,
which suggests that Anthropics costs have dramatically increased with these models,
so much that simply increase in the cost per million
tokens is in sufficient. It's my belief that the launchers
of both claud Opus and to a lesser extent, son
It have caused en up hevil in Anthropics costs and
compute demands, which in turn forced them to start increasing
(08:37):
costs on their customers. However, Anthropic has I believe, realized
that there is no real way to just increase the
cost of Opus and Sonnet for further on just the
amount of tokens that the customer might use, and that
doing so might push away smaller developers, which wouldn't make
them more money. So they decided to find a way
to specifically exploit the finances of their largest customers coding startups,
(08:58):
in a way that won't be immediately obvious or that
would spook non coding assistant customers. Except I'm a crafty
little fuck and I look at everything all the time.
I do this for fun. I don't do this because
I have to listen to me. Wario Amadee. If you're
listening to this, I'm watching you. I have archive dot
org and a million diet cokes. Anyway, last year I
talked about the subprime AI crisis, where almost the entire
(09:21):
tech industry is bought in on a technology sold at
this vastly discounted rate, and they've heavily centralized and subsidized
it too. And I predicted one major thing that these
model developers would eventually have to find a way to
make their costs work, that they would have to find
a way to crank up costs on their customers, because
otherwise they would just continue burning money. I still think
(09:42):
they are doing so, and I think that's what's happening here.
I think anthropic and OpenAI to a lesser extent of
realized that they need to start making money back on
these fuckers, and they're doing it. Except the fact that
the companies of immediate companies like Cursor and Repler have
immediately had to change their prices suggest that maybe not
one of these businesses makes sense. Maybe this whole time
(10:03):
this thing was unsustainable. Maybe it turns out the general
if AI doesn't have the kind of business returns you
need to run a startup. If only someone had said something. Remember,
look see how fucked up saying Remember but I'm not
editing it out. This is an honest podcast. But seriously, everyone,
remember the pale horses. Rate limits, service interruptions, price increases,
(10:28):
trouble raising funding, and trouble with money. The horses are
drawing nearer and telling you we're coming to the end
of this. I don't know if it will be soon,
I don't know if it'll be next year. But nothing
about this suggests that things are going well.
Host
Ed Zitron
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.