Today I want to talk about what the 
compute situation for the labs will
look like over the next few years.
For the last three consecutive years,
Anthropic's revenue has 10x'd year over year, and 
it's likely to do so again this year. They ended
last year with nine billion in revenue. I think 
they'll probably end this year with somewhere
between one hundred billion to one hundred 
and fifty billion dollars in revenue. Now,
for this trend to continue, Anthropic would 
need to make one trillion dollars in revenue
by the end of next year. Of course, 
there's no deep reason why this has
to be true. It's a very wild conclusion, and 
it's ultimately a question of AI capabilities.
Does AI get that useful by the end of next 
year? But suppose the trend does continue. I
want to think through what happens in that world.
The other big trend in AI is that lab compute only
3x's year over year. For a lab to keep 10x'ing 
revenue year over year while compute only 3x's,
one of the following three things needs to 
happen, or some combination of the three:
One, lab margins have to increase. Two, the 
price of compute has to increase. Or three,
the percentage of compute that labs spend on 
inference rather than training has to increase.
My understanding is that basically all three 
of these things are already happening. With
regards to the margins, Anthropic's inference 
margins reportedly went from forty percent in
the middle of last year to upwards of eighty 
percent now for Fable. With regards to compute,
the spot prices for compute are more than forty 
percent higher than they were in the February
trough that we had earlier this year. And with 
regards to the share of compute that goes to
training versus inference, in 2024, according to 
Epoch, OpenAI was spending just a quarter of its
compute on inference, and that number is likely 
closer to fifty percent, if not higher, now.
Now, labs would prefer not to do this final 
thing of increasing the share of compute they
spend on inference. The way the labs see the 
world, the whole point of inference revenue is
to help convince investors to give you more money 
in order to train the next bigger, better model.
And if you're spending most of your compute 
on inference, then you're basically declaring
that AI progress has stalled and you're just now 
in the business of being a cloud provider. This
is a less compelling business than building AGI, 
so the labs do not want to be in this business,
nor do they think they are in this world. They 
think that within a year, they'll have built
models that make the current ones look extremely 
shitty. But they need to invest a lot of their
compute — the majority of their compute 
— into doing the training and experiments
that are necessary to build the next model.
So that leaves only two options for how you
can get out of this gap between the fact that 
lab compute only increases 3x year over year,
but revenue increases 10x. Either 
the lab's margins have to increase
so that they get the surplus, or the price 
of compute has to increase so that everybody
in the stack below the lab gets the surplus.
It's not clear to me which world we end up in.
Do we end up in a world where we go from 80% for 
some of the top models to greater than 90% margins
if the lab margin effect dominates? That would 
require the leading model to be so far ahead of
the competition, because the nature of margins 
— why they exist in a market economy — is that
the thing you are serving is so much better than 
what somebody else could go get and replace you
on the market. But it's just really wild for me 
to consider that the margins for something like
intelligence will be greater than 90% and 
they don't get competed away at that level.
So that leaves only one other possibility of 
this escape valve between these two trends,
which is that the price of compute has to 
increase. As I mentioned, this is already starting
to happen. And the effect is even stronger when 
you look at the tranche of compute that the
frontier labs actually need to accumulate, because 
they can't just go out and buy a spot instance.
They need to make sure that they get enough scale 
to get really good efficiency and flexibility,
and also that they have the kind of compute that 
lends itself to the security they need for their
own weights and for their customers' information.
I think a relevant case study here is to look at
the compute that Google and Anthropic are renting 
from SpaceX. Google, for example, is paying
nine hundred million dollars a month for a hundred 
and ten thousand GPUs that are a blend of GB200s
and GB300s. The price that Google is paying here 
is 2x the spot price per hour for those GPUs. And
that spot price itself is more than forty percent 
higher than it would have been in February.
I want to emphasize a key conclusion here: as 
AI models get smarter, they will be better able
to monetize the same amount of compute. If a true 
human-level software engineer could run on an H100
equivalent, then at today's prices for software 
engineers, that H100 should rent for over 250K a
year. That's over 15x the current spot price for 
an H100. And this is not even accounting for the
fact that your AI can work nights and weekends.
Of course, you might expect that if we had ten
million extra software engineers suddenly appear 
in the economy, the marginal value of a software
engineer would decrease, and thus the revenue 
that that H100 would be able to generate would
not be 15x higher than it is right now. But 
I actually don't know if this is true. If we
apply this argument to people instead of AIs, then 
this would be the classic lump of labor fallacy.
For example, economists generally believe that 
high-skill immigration does not decrease wages
in the long run because of how innovation and 
specialization increase the value of labor.
Maybe this labor supply shock will be so big 
and so fast that we can't count on this general
heuristic anymore. But if you believe what 
standard economics says, then the marginal
value of labor, and thus the marginal value 
of compute, should stay astonishingly high.
So let's think about what changes in such a world. 
One of the things that would happen is that as the
top labs get better and better at monetizing 
compute, and the cost of compute increases,
it becomes harder for anybody else to 
compete against them, because they have
to bid for this resource against somebody who 
is basically able to make better use of it.
Another thing that will happen — and I think this 
is actually the most interesting implication of
this whole thought exercise — is that if you 
can train the best, most efficient model, then
you'll be able to charge much higher margins than 
you can today. This is the Alchian-Allen effect
in economics, and what it's basically saying is 
that if it costs twenty dollars an hour to rent
an H100, then it would be extremely stupid to use 
a weaker, less efficient model, because it's gonna
burn more tokens on your expensive compute to 
get the exact same result. So labs will be able
to charge a much larger premium if they can train 
a model that better economizes this scarce input.
Basically, if you have a model that can get the 
same result by using less compute, then you've,
in some sense, created more compute, and 
the value of compute is gonna increase.
Another thing that will happen is that a lot of 
current popular applications of AI will probably
get priced out. The reason AI is relatively cheap 
right now is that AI just can't do a lot of things
that top humans can do. But this, at some point, 
will no longer be the case. And at that point,
Google or Anthropic or OpenAI will be 
willing to pay more for the tokens to
automate AI research than you or I will be 
willing to pay to make more AI slop talk.
I'm a bit worried that this kind of analysis 
honestly pattern matches a lot onto the ways
that people in the past have been wrong 
about scarcity. I'm thinking, for example,
of the famous Simon-Ehrlich bet. Paul Ehrlich 
was this famous doomer about population growth,
and he made this bet that a basket of commodities 
would increase in price rather than decrease in
the decade preceding 1990. This is a very famous 
bet because it's supposed to illustrate how
Ehrlich's Malthusian worldview was wrong, 
and how he did not anticipate the way in
which market signals and human ingenuity can 
find better ways to economize scarce inputs.
I'm guessing that the analogy to this bet is 
probably wrong. Other analysis has shown that
if that bet had been made in a different decade, 
Ehrlich might well have won. But more generally,
I think the supply of compute is much less 
elastic, much less capable of absorbing large
demand shocks, and much less capable of being 
accommodated by using different substitutes
than the extraction of different metals is.
To illustrate why I think this 3x in compute
capacity year over year is hard to 
budge or potentially even sustain:
I don't see how any of the three elements that 
constitute that 3x can be much accelerated.
1.4x of that is coming from Moore's Law. Far 
from increasing it, I think it'll be a miracle
if we can just keep it going for a few more 
years. 1.2x is coming from building new fabs.
This process is ultimately gonna be bottlenecked 
up to 2030 and potentially even beyond by just
building new ASML EUV machines. Dylan, when he was 
on the podcast a few months ago, talked about this
in great detail. And 1.8x comes from the fact that 
AI is absorbing a lot of wafer allocation that was
previously going to smartphones and PCs. This is 
probably gonna hit a wall by the end of next year,
when at the leading edge N3 nodes at TSMC, AI 
will have gone from 60% to 86%. At some point,
you have just absorbed all leading-edge wafer 
capacity for AI, and you can't keep increasing
this number. So I don't know how we even continue 
to do 3x compute scaling year over year for the
next few years, much less go beyond that.
At the end of the month, I go through the
time-honored tradition of closing my 
books. I start by opening Mercury,
which is my banking platform, to make sure that 
all my transactions are properly categorized.
Auto-categorization rules handle the predictable 
stuff pretty well. But I'm constantly working
with new contractors—tutors, researchers, 
and videographers—and I'm also trying new
tools. Manually categorizing all of 
these transactions would add a couple
of hours of overhead every single month.
So instead of going through them one by one,
I have Command, Mercury's built-in AI, 
take a stab at all of them at once.
Command proposes a category for each transaction 
and provides its rationale. I just review,
fix anything that's off, and approve.
Once all this work is done in Mercury,
it syncs everything with QuickBooks. And 
Command's judgment calls are genuinely
good. It does the obvious things, like looking 
at the vendor, but it also investigates who on
my team made the purchase and looks at notes and 
memos to build up as much context as possible.
This is just one of the ways you can use Command 
to automate the back end of your business.
To learn more, go to mercury.com/command.
Mercury is a fintech company, not an
FDIC-insured bank. Banking services are provided 
through Choice Financial Group and Column N.A.,
Members FDIC. AI-generated responses and 
suggested actions may vary and are not guaranteed.
Now, I wanna clarify that at some point in 
the future, compute will get cheap again. At
some point, we'll just have robots that can 
convert shores of silica sand and mines of
copper into new computer chips, and then the 
price of compute is basically the raw inputs
and the tools required to do this processing. I'm 
just talking about this current pre-singularity
regime where AI compute merely 3x's year 
over year, which is not enough to offset
how much more valuable AI is becoming over time.
By the way, the fact that Anthropic's revenue has
been 10x'ing year over year, whereas their 
compute has only been 3x'ing year over year,
I think illustrates how strong the economies of 
scale are in the model business. And logically,
this makes sense. When you train a model, 
you just have to spend this one-time cost
to learn all these different skills that then 
get to be shared across all your users. This is
very unlike human labor, where each instance has 
to be retrained from scratch. I wish we didn't
live in a world with such strong economies of 
scale for intelligence, because I'm worried
about power concentration, but it seems we do.
Okay, this was a narration of a blog post that I
also released on my website at dwarkesh.com. Check 
it out for other posts or to be notified when I
release a post in the future. Otherwise, 
I'll see you for the next full episode.