Hi, I'm Euwyn.
Going to start posting here again, so a quick reintroduction.
I'm the founder of Orbital. We're building AI data centers in space.
A few things about me:
In 2010, I moved to Silicon Valley to join YC (S10) to build my first startup.
In 2017, I co-founded Spin, the first micromobility company in the US.
In 2018, Ford acquired Spin for close to $100M. Over the following years, we scaled it to 1,800 employees, deployed 250,000+ vehicles and expanded into 100+ markets.
In 2026, I founded Orbital. We raised funding from a16z speedrun, Human Interest, Basis Set, Anti Fund, and others. The team has decades of experience building satellites at Amazon Leo, Lockheed Martin, Northrop Grumman and other aerospace companies.
Our mission is to solve one of AI's biggest problems: the electricity shortage. AI demand is growing exponentially and we can't build enough power plants and grid capacity to match it. Data centers in space is the solution.
I'll be sharing what we're building, what we're learning and why I believe space will play a huge role in the future of AI.
If you're interested in AI, space, data centers and the future of computing, you'll probably enjoy what I post here.
Follow along for the journey.
Elon Musk says 100x100 miles of solar is enough to power the entire US.
That sounds crazy, right?
But at the World Economic Forum, Musk explained something fascinating.
A solar installation covering just 10,000 square miles could theoretically generate enough electricity to meet America's needs.
That's a relatively SMALL fraction of the country's total land area.
And Musk says similar potential exists in Europe.
So why aren't we doing this?
According to Musk, one major obstacle in the US is the high tariffs on imported solar equipment.
China manufactures enormous quantities of inexpensive solar panels, but trade barriers make them more expensive to deploy in America.
— Elon Musk (@elonmusk) at the World Economic Forum (.@wef) 9 months ago
Larry Page reportedly said 99.999% of computing will eventually be in orbit.
And Elon Musk publicly agreed with this prediction.
Sounds crazy, right?
But there's a reason why this idea is getting serious attention.
AI data centers require enormous amounts of electricity.
And solar power is one of the most promising ways to meet that demand.
But on Earth, solar panels are limited by day/night cycles, weather and seasonal changes.
Then there's the challenge of securing land, building grid connections, installing cooling infrastructure and obtaining regulatory approvals.
All of this making it difficult to build them at scale to keep up with AI's electricity demand.
But in space, satellites in certain orbits could harness near-constant sunlight 24/7 to power AI chips directly.
Without relying on Earth's electricity grids.
Without competing for valuable land or navigating local data center zoning approvals.
Without the limitations of nighttime, weather, and seasonal changes.
They could then transmit the computing results back to Earth.
Of course, there are still MAJOR challenges.
Launch costs, cooling, radiation and reliable communications all need to be solved at commercial scale.
— Martin Casado (.@martin_casado), GP at A16z (.@a16z), with Patrick O'Shaughnessy (.@patrick_oshag) on the ILTB (.@InvestLikeBest) podcast
This is exactly why we're building AI data centers in space.
According to Business Insider, Amazon expected to receive roughly $5 billion in Nvidia AI server racks in August, followed by another $3 billion per month.
Some of its data centers reportedly have enough physical space for these GPUs, but not enough electricity to run them.
So some of the equipment is being stored in warehouses until the facilities are ready.
By now, it's no surprise we're manufacturing AI chips faster than we can expand the electricity infrastructure needed to power them.
Building new power plants takes years. Connecting data centers to the grid can take even longer. and even large natural gas turbines have multiyear order backlogs.
So how do we solve this?
Orbital compute is the solution.
Amazon is parking billions of dollars of brand-new Nvidia computers in warehouses because there isn't enough electricity to turn them on.
And it plans to take delivery of about $3 billion more every month.
Elon called this back in February. He told Dwarkesh that by the end of this year, AI chips would start piling up because companies couldn't get the power to run them. There are about eight weeks left in the year, and it's already happening.
Business Insider got hold of Amazon's internal emails. The program is called Early Rack Receipt, and it means AWS takes ownership of the server racks before the data centers they're headed to can actually power them. One August email projected roughly $5 billion of Nvidia GB300 racks that month alone. Amazon says this is routine, so the racks can go live the moment the power shows up.
People in the industry call them dark GPUs.
Satya admitted a year ago that Microsoft had chips sitting in inventory with nowhere to plug them in, and the problem has spread a lot since then. Sightline Climate counted about 16 GW of US data center capacity that was supposed to come online in 2026, and by spring only around 5 GW of it had even started construction.
Building your own power plant sounds like the fix until you call GE Vernova. They're the biggest gas turbine maker in the world, they have 122 GW of turbines under contract, and their factories build about 20 GW a year. That's six years of output already spoken for. Management says 2029 and 2030 are close to sold out.
Elon piled on this week with the rest of the shopping list. Electrical equipment, cooling, networking, batteries. Every one of those has a backlog of its own, and a big power transformer alone takes around 128 weeks to show up.
Nuclear is where a lot of people's minds go, and it'll help in the 2030s. The last reactors America actually finished, Vogtle 3 and 4 in Georgia, got their construction license in 2012 and didn't both run until 2024.
This is why xAI's Memphis setup keeps looking smarter. xAI went around the utility entirely. It bought an old power plant across the state line in Southaven, Mississippi, filled it with mobile gas turbines, and used Tesla Megapacks to smooth out the power swings from training runs. Satellite images from May showed roughly 1.3 GW of temporary turbines on that one site.
Meanwhile, Amazon is on pace to take about $100 million of Nvidia hardware a day, and some of it rolls straight into a warehouse to wait for electricity.
Most satellites lose the sun for part of every orbit.
One orbit almost never does.
It’s called the dawn-dusk sun-synchronous orbit (SSO).
Here’s how it works.
By 2029, solar is projected to become the world's #1 renewable electricity source, overtaking hydropower and wind for the first time.
According to the IEA's forecasts, its estimated share of global electricity generation will be:
- Solar: ~14%
- Hydropower: ~12%
- Wind: ~11%
By 2030, the IEA projects solar will generate 15% of the world's electricity, nearly DOUBLING its share in just 5 years.
Just a decade ago, solar supplied only 1% of global electricity.
So what's driving this massive shift?
COST.
Building traditional power plants is incredibly expensive.
New nuclear plants require billions of dollars in upfront investment, lengthy regulatory approvals and construction that often takes many years.
Solar is much cheaper.
China has scaled solar panel manufacturing to unprecedented levels, driving prices down dramatically.
Today, some solar panels sell for under $100 and are designed to generate electricity for DECADES. and solar farms are being deployed much faster than new nuclear plants.
This chart is fascinating.
It shows a country's economic prosperity is closely linked to its ability to generate and access electricity.
And it's not hard to wrap your head around why. Almost EVERYTHING required to build a modern economy depends on electricity.
Factories need it to manufacture goods. Hospitals need it to operate. Businesses need it to run their operations.
Without sufficient electricity, countries struggle to develop industries, increase productivity and improve living standards.
Take the 3 wealthy countries:
1. Norway: ~20,000 kWh consumed per person annually, with GDP per capita of roughly $70,000.
2. United States: ~12,000 kWh per person, with GDP per capita around $65,000.
3. Japan: ~7,000 kWh per person, with GDP per capita around $40,000.
Now compare that with 3 of the poorest countries:
1. Burundi: ~30 kWh per person, with GDP per capita around $200.
2. Somalia: ~30 kWh per person, with GDP per capita around $400.
3. Chad: ~20 kWh per person, with GDP per capita around $600.
The difference is staggering.
Of course, electricity alone doesn't make a country wealthy. Wealthier economies also have more resources to build electricity infrastructure.
4 months ago at the SAIR conference, I was asked what was the number one piece of advice i'd give to my younger self.
My answer was simple.
Besides wishing I'd been born a little later to take advantage of everything happening in AI now.
Dive headfirst into AI.
Elon Musk says solar panels in SPACE will generate 5X more electricity than on Earth.
AI chips require enormous amounts of electricity to operate 24/7, but Earth's power grids can't expand fast enough to keep up with SKYROCKETING demand.
At the current pace, projections suggest a 100 GW electricity capacity shortfall by 2030.
So what are our options?
1) Connect to the GRID: Some data centers face 5–7 YEAR waits for electricity connections.
2) Build GAS plants: They need specialized turbine blades, vanes, generators and transformers. Large turbines face 5+ YEAR delivery queues, with some manufacturers booked through 2030.
3) Build NUCLEAR plants: Permitting and construction take YEARS. America's Vogtle reactors took roughly 14–15 YEARS from early construction to operation.
What about solar?
On Earth, panels lose sunlight to clouds, the atmosphere, and nighttime. That means data centers need expensive batteries or backup power to operate 24/7.
That's why Musk believes SPACE is the answer.
In the right orbit, solar panels avoid clouds and most nighttime interruptions, harvesting roughly 5X more energy over time.
Instead of beaming electricity to Earth, he'd power AI chips directly in orbit and transmit the results back as DATA.
— Elon Musk (@elonmusk) on Dwarkesh Patel's (@dwarkesh_sp) podcast
Elon Musk once said: “Earth only receives about half a billionth of the Sun’s energy... if you’re going to climb the Kardashev scale and harness some nontrivial percentage of the Sun’s energy... obviously, the only way to scale is to go to space with solar.”
Let me explain what he’s talking about.
The Kardashev scale is a way of measuring how advanced a civilization is, based on how much energy it can use.
Type I: uses all the energy available on its home planet
Type II: uses all the energy of its star
Type III: uses all the energy of its galaxy
The more energy a civilization can harness, the more it can build, and advance.
Humanity isn’t even at Type I yet.
Here’s why:
Earth only receives about half a billionth of the Sun’s energy.
So even if we wanted to harness just a millionth of the Sun’s output, that would mean roughly 100,000X more electricity than our entire civilization generates today.
No amount of power plants, nuclear reactors or fusion on Earth gets us anywhere close to that.
And AI is about to need more electricity than ever before.
That’s why Elon says the only way to scale is to go to space.
In the right orbit, solar panels can capture near-constant sunlight with no night, no clouds and no atmosphere in the way.
— Elon Musk (@elonmusk) on Dwarkesh Patel’s (@dwarkesh_sp) podcast
Larry Page reportedly has a number for how much computing will end up in space: five nines. As in 99.999%.
@patrick_oshag heard it secondhand: "Someone asked Larry Page how much of compute will be orbital, and he said five-nines."
@martin_casado's reason it might happen: "It's just basically regulatory." He thinks there are 500 proposed bills to stop data center building in the US, Texas included.
From @InvestLikeBest.