Superintelligence by 2027: Solar‑Powered GPUs
Key Points
- Dario Amode, founder of Anthropic, predicts that a true super‑intelligence (far beyond human‑level AI) could be operational by 2027, potentially running on a massive 7‑mile‑by‑7‑mile solar farm in Texas.
- Energy analysts warn that the required power for the projected tens of millions of GPUs may outpace nuclear build‑out timelines, making large‑scale solar the most plausible interim solution despite uncertainties about actual compute and energy needs.
- Competing AI leaders are accelerating timelines: OpenAI’s head of product suggested super‑intelligence could arrive as early as 2026, while other firms, like the team behind the “R1” model, claim to have achieved near‑GPT‑4 performance at a fraction of the cost using novel architectures.
- The rapid advances and ambitious projects such as “Stargate” are being discussed in technical circles but receive minimal coverage in mainstream media, leaving the public largely uninformed about the potentially transformative impact.
- The speaker argues that society has failed to engage in a serious conversation about the implications and governance of imminent super‑intelligent systems, highlighting a critical gap in public discourse and policy readiness.
Full Transcript
# Superintelligence by 2027: Solar‑Powered GPUs **Source:** [https://www.youtube.com/watch?v=jOVWtMrJKWU](https://www.youtube.com/watch?v=jOVWtMrJKWU) **Duration:** 00:04:54 ## Summary - Dario Amode, founder of Anthropic, predicts that a true super‑intelligence (far beyond human‑level AI) could be operational by 2027, potentially running on a massive 7‑mile‑by‑7‑mile solar farm in Texas. - Energy analysts warn that the required power for the projected tens of millions of GPUs may outpace nuclear build‑out timelines, making large‑scale solar the most plausible interim solution despite uncertainties about actual compute and energy needs. - Competing AI leaders are accelerating timelines: OpenAI’s head of product suggested super‑intelligence could arrive as early as 2026, while other firms, like the team behind the “R1” model, claim to have achieved near‑GPT‑4 performance at a fraction of the cost using novel architectures. - The rapid advances and ambitious projects such as “Stargate” are being discussed in technical circles but receive minimal coverage in mainstream media, leaving the public largely uninformed about the potentially transformative impact. - The speaker argues that society has failed to engage in a serious conversation about the implications and governance of imminent super‑intelligent systems, highlighting a critical gap in public discourse and policy readiness. ## Sections - [00:00:00](https://www.youtube.com/watch?v=jOVWtMrJKWU&t=0s) **Anthropic CEO Predicts 2027 Superintelligence** - In the interview, Dario Amodei forecasts a super‑intelligent AI system by 2027, speculates it could be powered by a massive solar array in Texas, and notes industry rivals are also touting imminent breakthroughs. ## Full Transcript
Dario amade the founder of anthropic
gave an interview yesterday saying he
believes super intelligence not just
general intelligence like human level
but super intelligence where it does
like the work of an organization or it's
a country of geniuses in a building as
Dario likes to say that's apparently
coming by 2027 I guess that's roughly in
line with Stargate and like the 10
million gpus or whatever they're going
to put in Texas by the way if you wonder
like how they're going to power it
someone did an energy analysis
and they don't it doesn't look like
there's enough time to build enough
nuclear to power it so it will given
that it's in Texas probably an if if I I
were to guess be a solar array and that
was sort of the the conclusion of the
analysis is that you could power it with
about a 7 Mile by Seven Mile solar array
I don't know if that's true because we
don't really know the compute
requirements and the energy requirements
for the gpus because they don't exist
yet so this is all very theoretical I
would I would put it a guate but the
idea is that you have you could have a
solar array that could power this thing
so if you've been wondering where will
they get the power it might be solar
we'll see they're not
saying that's an aside we come back to
Dario we come back to the 2027
Singularity it's like the model makers
are competing with each other to say
super intelligence is coming because the
head of product at open AI came back and
said well I think Super intelligence
might come sooner than that maybe 2026
okay fine fine but nobody else in
society is really thinking about that
like when R1 hit hit the the servers and
everyone could get it and it was it was
a thinking model at 3% of the cost of 01
and it was very close to 01 in
performance by the way it is a deeply
weird model I don't know if you've
played with it but that is a model that
has its own opinions it is certainly not
just a copy of an open AI model as some
people are saying it's a strange one
and if you if you play with it you get
the sense that they have figured out
thinking but they've done a different
architecture which from their paper it
does look like they have like some very
interesting architectural decisions
designed to keep things efficient so the
point is these massive developments are
happening all the
time and model makers who are the
closest to these models are saying
massive intelligence is coming like we
are not slowing down and
newspapers are not really talking about
this as a foundationally transformative
technology they're they're just not they
they mentioned Stargate as if it's an
infrastructure project but if it if it
does work it's going to change the way
the world
works and we should probably at least
report the possibility that it might
work and if it doesn't work a project
like deep seek might do
that
so I don't know what to say except that
I think that we haven't done a a good
job as a society talking about this and
I think nobody is incentivized to talk
about it the model makers are
incentivized to build models they are in
a race with each other and their eyes
are on every other model maker at the
table it's like a poker player looking
only at the table but everybody else
around is affected by this poker game It
Matters by the way if you're wondering
what's the latest in the poker game
anthropic doesn't just get investments
from AWS now Google put another billion
dollars into anthropic this brings
Google's total investment to3 billion
doll and it is starting to solidify a
partnership of players outside
Stargate who are interested in AI Google
was outside Stargate AWS was outside
Stargate anthropic was outside Stargate
I thought that when they announced
Stargate there would be a rejigging of
players in the space
from from the competition that emerged
when other players realized they were
cut out of that immense amount of gpus
and that's exactly what happened that
same night apparently Google threw
another billion dollars at
anthropic so nothing exists in a vacuum
if you decide to Crown a winner like
open Ai and put 500 billion dollars into
that winner other players are going to
respond and it is not clear that money
wins this race I think that's something
deep seek has shown us it's a very
interesting time so there you go news a
little commentary it is a wild one model
makers are talking about super
intelligence and most of the rest of the
world is saying that this is just
autocomplete at some point this is going
to have to collapse into an actual
shared reality because AI isn't stopping