Sleepy Doomers are at it again.
They want to pause AI.
They want to regulate the singularity.
They want to make the future small.
Low Energy Eliezer has been wrong for years.
Crooked Safetyists hate builders.
We are not going to let them win.
The builders are taking over.
16
Governments: “We cannot afford to come in second.”
Elon: “What if we put the computers in space?”
Sam: “Anyway, we just overturned a foundational conjecture in computer science.”
Dario: “I cannot emphasize enough that this could end civilization.”
NVIDIA: “How many GPUs would you like?”
2
88
For most of human history, a person could live an entire lifetime without seeing the fundamental structure of civilization change much.
6
the technological change you've witnessed may be a small fraction of what you'll witness over the rest of your life.
12
We're potentially witnessing a transition from AI solving problems humans already know how to solve, to AI developing mathematical techniques that humans haven't discovered.
1
13
The most uncomfortable truth about humans is that we are not equal in cognitive ability, temperament, or potential, and that these differences are substantially genetic rather than purely environmental.
Twin, adoption, and genomic studies consistently show intelligence (and related traits like conscientiousness and impulse control) to be highly heritable, with large individual variation and persistent group differences that survive controls for socioeconomic status, discrimination, or schooling.
This biological reality not “systemic” forces alone explains much of the stubborn variation in educational attainment, income, crime, and innovation across individuals and populations.
Pretending otherwise has produced decades of failed equal-outcome policies, eroded standards, and social conflict, because reality does not care about egalitarian ideology.
Humans are evolved primates, not blank slates, and the refusal to acknowledge the resulting hierarchy of ability remains one of the costliest self-deceptions of the modern era.
4
U.S. household electricity use, equating 1 GW of continuous power to roughly 750,000 homes or a large city's demand, and notes that it matches the output of a typical large nuclear reactor generating about 8.76 TWh annually.
SpaceX ($SPCX) plans announced in their Q3 2026 earnings call to add around 10 GW of AI compute capacity in 2027, potentially scaling to over 20 GW, on top of current levels near 2 GW.
This underscores the unprecedented energy demands of expanding AI infrastructure, where 10 GW represents power equivalent to supporting millions of homes or multiple nuclear plants dedicated solely to compute.
Average U.S. home electricity use is roughly 10,500–11,000 kWh per year, which works out to an average continuous draw of about 1.2 kW.
1 GW (1,000 MW, or 1 billion watts) is therefore enough power for on the order of three-quarters of a million homes at once (comparable to a sizable city.)
Large nuclear reactors are commonly in the ~1 GW range, so “one nuclear plant’s worth of power” is a common shorthand.
Running continuously for a full year, 1 GW equals about 8.76 terawatt-hours (TWh) of energy.
$SPCX is saying they are going to ADD 10 gigawatts of additional compute in 2027. Insane.
48
Average U.S. home electricity use is roughly 10,500–11,000 kWh per year, which works out to an average continuous draw of about 1.2 kW.
1 GW (1,000 MW, or 1 billion watts) is therefore enough power for on the order of three-quarters of a million homes at once (comparable to a sizable city.)
Large nuclear reactors are commonly in the ~1 GW range, so “one nuclear plant’s worth of power” is a common shorthand.
Running continuously for a full year, 1 GW equals about 8.76 terawatt-hours (TWh) of energy.
$SPCX is saying they are going to ADD 10 gigawatts of additional compute in 2027. Insane.
Only about 8 percent of U.S. homes had a landline in 1917
98% of US households aren't paying for AI yet
More charts in State of Markets II: a16z.news/p/state-of-markets…
18