Technological Singularity Is Here · A Biographical Brief with Interwoven Concepts in Science · By Ashok Mehan
This Week · Energy, Power & the Scale of Everything
The Universe Built You From Stardust. China Is Building the Future From Gigawatts. America Is Still Arguing About the Permit.
A Carl Sagan calculation, a power grid that tells you everything about who will win the AI race, and a memoir passage about what it means to build something from nothing — twice.
By Ashok Mehan · Washington, D.C.
Carl Sagan famously said that if you wish to make an apple pie from scratch, you must first invent the universe. I have been thinking about that line for months, ever since I began writing the chapters on energy and compute in this book. Because Sagan wasn't being whimsical. He was being precise. The atoms in that apple pie — the carbon, oxygen, hydrogen, nitrogen — were forged in the cores of stars that exploded billions of years before our sun existed. Every particle in your body has a cosmic address. And every watt of electricity powering the datacenters racing to build artificial superintelligence has the same lineage. It all started with the Big Bang. It all ends, if we are not careful, with a civilisation that ran out of power before it ran out of ideas.
Energy is not a supporting character in the AI story. It is the protagonist. Without power, there are no datacenters. Without datacenters, there are no frontier models. Without frontier models, there is no AGI. Without AGI, the singularity remains a line on a graph rather than a lived reality. The nations that control energy will control intelligence. And the numbers, when you look at them honestly, are not encouraging for the United States.
"Compute is the new oil. Energy is the new territory. Datacenters are the new factories. And GPUs are the new weapons."— Technological Singularity Is Here, Chapter 6
In 2024, China added 546 gigawatts of new power capacity. The United States added 51 gigawatts. That is not a gap. That is a chasm. China added more power in a single year than the United States has built in the last five combined. And America's stated need — roughly one million gigawatts by 2030 to meet AI datacenter demand — is being addressed primarily through nuclear plants that produce between one and one-and-a-half gigawatts each, take a decade to permit and construct, and face fierce local opposition at every stage. The math does not work. The timeline does not work. And yet the conversation in Washington continues to unfold as if this is a problem that can be solved by the next committee.
China has no such committee. It has a plan, a grid, and the political will to execute both at a pace that democratic systems structurally cannot match. This is not a criticism of democracy. It is a description of physics meeting politics. And physics, as always, is indifferent to the outcome.
✦ From the Desk
A reader wrote to me this week and asked: "Ashok, do you actually believe America can lose this race?" My answer is that America is not losing the race. It is running a different race — one defined by innovation at the frontier, the best researchers, the most capable models. What it is losing is the infrastructure race. And without infrastructure, frontier innovation eventually stalls.
I have built companies from nothing. I know what happens when you have the idea but not the power supply. You stall. You improvise. And sometimes — if you are lucky — you find a different way. America needs to find that way. Quickly.
Power Capacity Added in 2024 — U.S. vs. China
China's 2024 additions = 10× the US. The 2030 US target is ~1,832× China's entire 2024 output.
The Nuclear Math Problem
- US target ~1,000,000 GW by 2030 for AI demand
- Nuclear output 1–1.5 GW per plant
- Plants needed ~667,000–1,000,000 new plants
- Build time 10–15 years per plant, minimum
- Ireland has already halted all new datacenter construction
- N. Virginia utilities warn demand now outpaces capacity
- Conclusion The grid cannot scale at the pace AI requires on current policy
546 GWChina Power Added, 2024
51 GWUS Power Added, 2024
10×China's Advantage in New Grid Capacity
10%Global Electricity for Datacenters by 2030
400BStars in the Milky Way — Sagan's Cosmic Scale
The Sagan Calculation · From Stardust to Gigawatts
How Big Is the Universe? A Calculation for Human Scale
Carl Sagan liked to put the cosmos in human terms. I want to do the same — connecting the scale of the universe to the scale of what we are now trying to build with artificial intelligence. These numbers are not decorative. They are the point.
The observable universe contains roughly 2 trillion galaxies. Our own Milky Way holds approximately 400 billion stars. The average star has roughly 10 trillion trillion atoms — that is 10 followed by 22 zeros, or 10²². Multiply those together and you approach the total number of atoms in all the stars of the observable universe: a number so large it makes the US national debt look like a rounding error on a napkin.
Now consider the human brain. It contains roughly 100 billion neurons — 10¹¹ — each connected to up to 10,000 others, yielding somewhere in the range of 10¹⁵ synaptic connections. That is one quadrillion connections inside a 3-pound organ burning 20% of your resting energy. Sagan's point was that the universe is not abstract. It is made of the same stuff as you. And you are, by any cosmic measure, extraordinarily improbable.
"The cosmos is within us. We are made of star-stuff. We are a way for the universe to know itself."
— Carl Sagan, Cosmos (1980)
Now extend this to AI. A frontier model like GPT-5.2 trains on roughly 10²⁷ floating-point operations — FLOPs. That number is in the same order of magnitude as the number of atoms in several thousand human brains. We are not building software. We are building something whose computational substrate begins to rival the physical complexity of biology. And we are doing it on a power grid that, in the United States, grew by 51 gigawatts last year.
The Sagan Numbers: A Working Table
| Scale | What It Represents | Number |
| Galaxies (observable universe) | Each a city of stars | ~2 × 10¹² |
| Stars in Milky Way | Each a potential sun | ~4 × 10¹¹ |
| Atoms per star (avg.) | Mostly hydrogen & helium | ~10²²–10⁵⁷ |
| Atoms in human body | You, entirely | ~7 × 10²⁷ |
| Neurons in human brain | The seat of your intelligence | ~10¹¹ |
| Synaptic connections | Where memory & thought live | ~10¹⁵ |
| FLOPs to train GPT-5.2 | One frontier AI training run | ~10²⁷ |
| FLOPs for next-gen ASI (est.) | What comes after GPT-5.2 | ~10³⁰+ |
| Watts — one datacenter (large) | Power for one training cluster | ~10⁹ W (1 GW) |
| Total US grid added (2024) | New capacity, entire nation | 51 × 10⁹ W |
| Total China grid added (2024) | New capacity, entire nation | 546 × 10⁹ W |
The Sagan analogy: training a next-generation ASI system requires compute approaching the number of atoms in a human body. The power to run it requires more grid capacity than the US added in all of 2024. These are not rhetorical numbers. They are engineering constraints.
From the Memoir · Building From Nothing — Twice
SMAC, the Lawsuit & the Second Beginning
In December 1990, I founded SMAC Data Systems on $5,000 spread across four credit cards. No office, no team, no blueprint. Just a gap in the market I had spotted while lugging boxes in a computer warehouse and an obstinate belief that someone had to fill it. The company grew to serve the computer manufacturing and reseller ecosystem in the Washington metro area, and for a while — a good while — it worked. Not smoothly. Not without setbacks. But it worked.
Then came a lawsuit from a disgruntled employee. I didn't know anyone who could help me navigate it. I didn't have the legal infrastructure. I folded the business in 2001. Every cent gone. The company I had built from four credit cards and a warehouse pallet was over in a way that felt both sudden and inevitable, the way all long-building disasters do when they finally arrive.
What I didn't know in 2001 was that in 2004 I would build something even more unlikely. FEDMINE.US was born on a Mac computer, teaching myself MySQL from an O'Reilly subscription, hiring a programmer 4,000 miles away to access data I had loaded myself without knowing how to write a single line of code. The US government spends roughly $800 billion a year. At the time, fewer than a handful of firms were tracking it with any sophistication. That gap was a market. I walked into it alone.
I tell you this because the energy chapter in this book is not abstract to me either. I know what it means to try to build something when the infrastructure isn't there. When the power supply — metaphorical or otherwise — isn't keeping pace with the ambition. The United States is in exactly that position today. It has the ambition. It has the models. It does not yet have the grid.
"Every failure carved a new shape. Every collapse forced a reinvention. Every reinvention taught me resilience — the one trait without which sentient life on Earth would be unrecognisable."— Technological Singularity Is Here, Introduction
✦ On Building Without a Blueprint
People sometimes ask me how I built FEDMINE without knowing how to code. The honest answer is that I didn't build it. I assembled it — piece by piece, failure by failure, late night by late night. I found the gaps and filled them with whatever I had available. A Mac with open-source tools. A programmer in a different time zone. An O'Reilly subscription and stubbornness that my family would have diagnosed as a clinical condition.
The US needs to do the same thing with its energy grid. Stop waiting for the perfect nuclear plant. Find the gaps. Fill them. Quickly. The window is not wide.
The DeepSeek Problem
- Method Fake accounts + stealth routers to access US frontier models
- Technique Chain-of-Thought distillation at industrial scale
- Result Capable models built at 20% of normal R&D cost and time
- Risk Distilled models strip safety guardrails — dangerous capabilities proliferate
- Anthropic "Models built through illicit distillation are unlikely to retain safeguards"
- Pattern Same state-backed IP theft China has practised for decades in software
Space Is the Only Answer
- On Earth Solar is intermittent; nuclear takes decades; grids are saturated
- In orbit Sun never sets in sun-synchronous orbit — constant, uninterrupted power
- Google Actively discussing TPU clusters in orbit
- Next step Swarms of compute satellites → early Dyson swarm architecture
- Material Asteroid mining for aluminium, nickel, rare earths — no longer speculative
- Timeline First generation orbital datacenters within this decade
Reading, Thinking, Writing
What I'm Reading This Week
1Cosmos — Carl Sagan (1980). I return to this every few years. It remains the most humane book ever written about scale. Sagan understood that the point of making you feel small was not to diminish you — it was to make you understand that you are, improbably and magnificently, part of all of it.
2The Coming Wave — Mustafa Suleyman. The co-founder of DeepMind on why AI and synthetic biology represent a containment problem unlike any humanity has previously faced. His chapter on energy infrastructure reads like a companion piece to this issue.
3Power, Sex, Suicide: Mitochondria and the Meaning of Life — Nick Lane. A strange and wonderful book about energy at the cellular level. If you want to understand why power is the foundation of everything — from the first living cell to the first datacenter — this is the place to start.
4The Precipice — Toby Ord. On existential risk and humanity's long-run future. Ord's chapter on the risks from misaligned AI is, in my view, the most sober and rigorous treatment of the subject available to the general reader. Required reading before Issue No. 4.
The Cosmic Ledger: A Final Note
Sagan's calculation — that the number of stars in the observable universe exceeds the number of grains of sand on all of Earth's beaches — is not a party trick. It is a calibration device. It tells you where you are in the scheme of things.
Here is my version of that calculation for the AI era. The number of floating-point operations required to train a next-generation ASI system is approaching the number of atoms in a human body: roughly 10²⁷ to 10³⁰. The power required to execute those operations in a reasonable timeframe — months, not centuries — is measured in gigawatts. And the gap between the gigawatts America added to its grid in 2024 and the gigawatts it needs by 2030 is not a policy problem. It is a physics problem dressed in a policy problem's clothing.
The universe took 13.8 billion years to produce a species capable of asking these questions. It would be a remarkable waste to let the answer be: we ran out of electricity.
Next Issue Preview
Issue No. 4 takes up the collapse of the professions — law, medicine, media, finance — and the strange new cultural reality of synthetic personas. Plus more from the memoir: the FEDMINE years, 18-hour days, and what it means to retire from something you built alone and then watch a machine do it in seconds.
Next Issue
The Collapse of the Professions: Law, Medicine, Media
Synthetic Personas & the New Cultural Reality
From the Memoir: The FEDMINE Years
The Escape Risk: When AI Slips Beyond Oversight