Has the singularity begun?
It sure seems like it…
In today’s Diary:
Cheaper than sunshine
It will be illegal to drive
Singularity saved my life
Drones that make it rain
The fastest $100M ever
Hey Rational Optimist,
Ever hear of the Singularity?
In the early 1990s futurist Ray Kurzweil predicted superhuman intelligence would trigger a point where technological progress goes exponential, rapidly and irreversibly transforming human life.
It appears he was right.
“We’re in the Singularity,” Elon Musk said recently. Stripe founder Patrick Collison put it more precisely: “I think Q1 2026 will be looked back upon as the first quarter of the Singularity.”
If you feel like the world is moving faster than ever, you’re not wrong. Many important technologies are going exponential. They all look the same: a long, flat line… then a sharp “whoosh.” We’re living in the whoosh.
Take AI. The original ChatGPT could work independently for 36 seconds before getting confused and outputting nonsense.
Anthropic’s latest Claude model can now reliably complete tasks that take a human expert 14.5 hours. Look at the progress on this important measure. Slow, steady growth. Then whoosh:
AI is doubling the length of tasks it can complete roughly every four months. By the end of 2026 the most powerful AI models will be able to finish 80+ hour tasks while you sleep.
Each leap makes AI exponentially more useful. At 30 minutes AI can write a block of code. At a few hours it can hunt down and fix bugs. At multiple days it can handle full audits.
What this means for you. Whatever field you work in, become the person who knows AI best. I’ve written two guides on AI agents: How to use AI agents and what I’m telling my kids about AI. These are great places to start.
The leap in AI capability is only possible thanks to the leap in raw computing power. Training today’s best AI models takes roughly 65 million times more computing power than it did in 2016:
Whoosh. Moore’s Law predicted chip density would double every two years. AI is lapping Moore’s Law, with “compute” doubling roughly every six months.
What this means for you. Computing’s exponential growth already transformed NVIDIA into the world’s most valuable company. But to feed AI chips we also need networking gear, liquid cooling systems, transformers and an enormous amount of power. Invest accordingly.
Intelligence is now too cheap to meter…
Three years ago it cost $37.50 to generate a million AI “tokens.” Today you can buy the same level of intelligence for $0.14. That’s a 99.6% price collapse.
This chart plots the cost of AI against its capability. Every new model is both smarter and cheaper than the one before:
Electricity was so expensive at first only factories and rich people could afford it. Then prices collapsed, and we put electricity in everything, even teddy bears.
This is now happening with intelligence. Soon a student in Nigeria will be able to afford an AI-personalized tutor once reserved for trust fund kids in New York. I have a feeling AI will be the great equalizer.
I’ve been a huge winner from the collapse in AI costs. I’ve never coded in my life and built an entire startup database for ROS with Claude Code. Two years ago this would’ve cost me thousands of dollars and weeks—if it were even possible.
What this means for you. Pick a project. Something you’ve wanted to build for years but never got around to doing. Put Claude Code to work. You can have a world-class personal AI agent for $100/month. There are no excuses. This is the worst AI will ever be.
Solar’s going exponential too
A solar module that cost $2/watt in 2010 costs $0.10 today. Prices are down more than 99.9% since 1975. We can extract roughly twice the power from the same roof space as you could 20 years ago.
Look at that whoosh through the roof.
The cost of storing a kilowatt-hour of energy has also collapsed 99% over 30 years. We deployed more battery storage in the last two years than in all of prior history combined.
I don’t see whole countries running on sun power anytime soon. But given how fast solar costs are falling we’d be foolish to ignore this trend.
What this means for you. The most interesting opportunities aren’t the solar panels themselves. It’s what dirt-cheap solar enables. My friend Casey Handmer at Terraform Industries is taking advantage of rapidly falling solar prices to make natural gas from sunshine, water, and air.
Deep in the sun’s core billions of atoms smash together each second…
Every time atoms “fuse” they unleash mind-boggling amounts of energy. Just 1 gram of fusion fuel the size of a crumb could power your home for over 50 years.
Fusion is energy’s holy grail. Problem is, atoms naturally repel each other. To overcome this you must heat a puff of gas to about six times hotter than the sun’s core. And that’s only the first step.
We’re finally making big leaps in our quest to create a star on Earth.
In 2022 the National Ignition Facility (NIF) in California crossed the breakeven point where a fusion reactor produced more energy than it consumed. Last April the NIF achieved 4X energy out versus energy in. Whoosh:
We still don’t have a fusion reactor that can deliver power to a grid. But two forces are converging that make me think we’ll get there this decade.
The great handoff. Fusion was long the exclusive domain of massive, slow government projects. Now there are dozens of well-funded innovators with startup urgency competing to recreate the sun’s fusion power here on Earth.
AI. Now AI can simulate plasma behavior in minutes. What used to take years of physical experiments and millions of dollars now takes an afternoon.
What makes the Singularity different from other innovation booms is how these trends feed each other. Better AI can bring us nuclear fusion faster. And then fusion can power AI.
What this means for you. It’s time to take fusion seriously. Dan and I visited Pacific Fusion and Fuse in California last month. I hear Proxima Fusion out of Europe is a leader too.
In 2003 scientists read the whole DNA of one person.
It took 13 years and $2.7 billion dollars.
A scrappy San Diego startup named Element Biosciences just did it for 100 bucks!
The cost of reading DNA has plunged a millionfold in the past 25 years. It’s turned a monumental scientific undertaking into a test you can do at home:
A few years ago I spat into a tube and mailed it to 23andMe. I got back roughly 700,000 data points about my genes. I recently took that raw data file and fed it into Claude.
Claude told me I carry the gene most strongly associated with Alzheimer’s risk. It also told me the best thing to reduce that risk is two hours of light aerobic exercise per week.
Cheap DNA reading + AI = data that could add years to our lives. It’s the Singularity showing up in a spit tube.
I think the most underrated genomics use case is embryo selection. Startups like Herasight and Orchid offer parents pursuing IVF the ability to screen embryos. Families are already sidestepping hereditary cancers and severe disorders that were previously invisible.
What this means for you. 23andMe only maps 0.1% of your genes. Companies like Nebula Genomics and Sequencing.com read your whole genome. Go get your DNA mapped for a few hundred bucks. Once you get the results feed it into Claude and start asking questions.
It will be illegal for humans to drive by 2050.
That’s my unpopular take after seeing the latest self-driving car data.
Waymo’s robotaxis have driven over 200 million fully autonomous miles on US roads. It took Waymo 16 years to reach the first 100 million mark—then only 7 months to double that.
Its fleet now logs nearly half a million fully autonomous miles every day. Whoosh:
Robotaxis are getting exponentially safer too.
In 2022 Tesla’s “full self-drive” mode needed a human to intervene roughly every 15 miles. The latest version travels 2,800 miles between critical interventions. That’s a nearly 200X improvement in four years.
The moment a politician can say, “human drivers are now the leading cause of preventable road deaths” is the moment we start seeing driving bans. You may be the last generation of human drivers. I say “you” because I don’t drive!
What this means for you. Carl Sagan famously said, “It was easy to predict mass car ownership, but hard to predict Walmart.” We can all imagine a city filled with robotaxis. But what kind of businesses spring up when you reclaim 10+ hours a week of commute time?
Every 30 seconds a Zipline drone makes a delivery.
We crowned Zipline “innovator of the year” in our inaugural ROS awards.
Zipline took 7.5 years to reach 1 million deliveries—and less than two years to cross 2 million:
Whoosh.
Zipline has flown 140 million autonomous miles with zero safety incidents.
In fact, it’s saved countless lives by dropping blood and medicine into remote Rwandan villages by drone. Now it’s coming to America.
Driving robots (robotaxis) and flying robots (drones) are like powerful longevity drugs. Get ready for drones that stamp out wildfires (Seneca Systems), drones that kill bad guys (Neros) and drones that make it rain (Rainmaker).
What this means for you. The use cases for drones are endless. I recently chatted with a founder whose drones are used to inspect rollercoasters. Think about how you can use drones to solve a pain point.
Imagine if every time you flew, the airline had to build a new plane.
That’s how space travel used to work.
Before SpaceX rockets were “one-and-done.” SpaceX figured out how to “catch” rockets to reuse them—and slashed the cost of a rocket launch by over 95%:
We know what happens when costs fall that fast.
Last year SpaceX completed more missions than NASA’s Space Shuttle program did in its entire 30-year history. The number of objects launched into space is going vertical:
This whoosh is the foundation for a trillion-dollar orbital economy. Starship will be able to launch more mass in a single month than the entire Space Shuttle era.
What this means for you. The most interesting opportunities aren’t in rockets, but what cheap space travel unlocks. Two startups we met in California have me giddy: Varda is making drugs in orbit, and Apex Space is building a high-speed assembly line for a massive missile defense shield.
How fast can a startup reach $100 million in annual revenue?
Microsoft—over a decade. Facebook and Google—four years. Slack, a hypergrowth darling of the last tech wave, took a little over three years.
AI startups are leaving them in the dust.
Coding tool Cursor reached $100 million in annual revenue in under two years. AI voice company ElevenLabs did it in roughly the same time:
This is the economic proof of the Singularity. It shows the real-world wealth creation velocity is itself going exponential.
What this means for you. The barrier between an idea and a successful business has never been lower. Cursor was two dudes who knew AI would change coding. It hit $100 million in sales with fewer than 20 employees. For investors, understand the old valuation frameworks are dead.
The AI era’s winners will look absurd by historical standards.
Scared of the Singularity?
Most people are. But not us.
I think the Singularity will be among the greatest wealth creation events in history. The barriers between ordinary people and extraordinary outcomes have never been lower.
The Singularity rewards agency, curiosity and the willingness to try new tools—not credentials.
By being a ROS member and paying attention you are already ahead of the vast majority of people.
Please hit “like” and “restack” to help us bring new Rational Optimists into the fold.
—Stephen McBride
The future looks brilliant. The present is filled with fear and rage. We have to get through a few years first. CMac
You didn't mention robots making robots (which, for industrial robots, is already the case). Now extend that to pretty much everything. We are headed for an age of abundance.