A company in every pocket

Mars is impossible, and that is the point

This week SpaceX goes public at around $1.75 trillion, the largest IPO in history.

Morningstar ran the numbers on the actual business and got $780 billion. Less than half.

That roughly trillion-dollar gap is the dream. It's Mars.

Nobody is going to live there anytime soon. The economics are absurd, and even a routine human trip will stay out of reach for years.

Yet "make humanity multiplanetary" may turn out to be the most lucrative corporate mission of the century.

Because chasing it forced SpaceX to build reusable rockets and a global satellite network.

The destination matters less than what gets built on the way.

And the dream is most of what investors are buying. SpaceX is priced on where it's going, not where it is.

When building Voltaire, I wanted to figure out what our "Mars" would be.

The famed "one-person billion-dollar company".

But above all, making life better for a billion people.

I dream of a world where a billion "ordinary people" each run a real company staffed by AI agents they own and control.

And the economy grows at rates that look unhinged from where we sit in 2026.

Science fiction, today.

A mission like that isn't just inspiration. It's a filter. It keeps you building the right thing, right now.

AGI much?

The reason to take this seriously isn't just based on anyone's hypotheses. It's actually been studied.

METR, an evaluations lab, measures how long an AI agent can keep going on a task independently.

That length has doubled about every seven months for six years. And from 2024 to 2025, it doubled every four months. In 2019, agents handled tasks of about a minute. By 2025, they were past the hour mark.

Run that line forward, and somewhere around 2030, an agent reliably performs work that would take a skilled person a month (METR, "Measuring AI Ability to Complete Long Tasks," 2025).

A month of unattended work signals "autonomous worker".

Demis Hassabis gives roughly even odds on AGI by 2030, with 2029 in play. Dario Amodei calls the target a "country of geniuses in a datacenter": millions of AIs, each sharper than our best people, running 10 to 100 times faster.

DeepMind's own safety paper treats human-level AI emerging by 2030 as plausible enough to plan around.

The skeptic worth respecting, Daron Acemoglu, puts the gain at under 2% of US GDP over a decade (2024).

The people actually building the models think he's low by orders of magnitude. That's the camp we build for.

2030: cognitive labor gets nearly free

Say AGI, or something close enough that the difference stops mattering for business, arrives around 2030.

One economic fact follows, and everything else hangs off it: the price of competent thinking work collapses toward the cost of the electricity to run it.

Expertise has always been scarce and expensive. A good lawyer, a good marketer, a sharp analyst, a real engineer: each costs a salary, and you could only afford a handful.

In this scenario, you rent any of them, all of them, by the minute, by the thousand.

Epoch AI's GATE model finds that AI-driven growth kicks in more easily than some economists have assumed; that the economically optimal AI investment could hit $25 trillion in a single year; that compute spending could pass 10% of world GDP; and that, under heavy automation, global output might grow somewhere between 30 and 100% a year.

For comparison, the postwar economic miracle was about 10% a year, in a couple of countries, for a couple of decades. GATE is putting something like that on the table globally every year as a floor.

Dario Amodei reaches the same neighborhood through a different door: 10-20% annual growth once cognitive work is automated at scale.

I won't pretend these numbers are settled. They're the bull case from credible people, backed by actual models.

Our bet is simple: the bull case is worth building for, because if it's even half right, 2040 will be unrecognizable.

2040: the end of large enterprises

Picture 2040, a decade into cheap thinking work. What changes for regular people?

The big company stops making as much sense. Firms got large mostly because coordinating expertise was expensive. You hired people, managed them, and put them under one roof because that was the cheapest way to assemble the ten or fifty or ten thousand kinds of skill a business needs. Make skills available on tap for cents, and the reason to bundle hundreds of humans into one entity gets weaker.

Years ago, Sam Altman predicted the first one-person billion-dollar company would emerge soon. In this scenario, by 2040, that isn't a stunt. It's a category.

Work flips too. For two centuries, the move was to sell your hours to whoever owned the machine.

When the machine is a model anyone can rent, you can own the judgment calls instead and point its AI workforce at something people will pay for.

The IMF already estimates that AI will affect 40% of jobs worldwide and 60% in rich countries, and its managing director has repeatedly warned that the most likely result is greater inequality. Hold onto that warning. It's the whole game.

Because here's what actually gets scarce. When intelligence is everywhere and cheap, intelligence stops being the bottleneck. Agency takes its place. Knowing what to aim all that capability at. Being allowed to act. Owning the result.

A million genius-equivalents in a datacenter do nothing until someone points them at a real market and signs their name to the outcome.

In 2040, the scarce thing isn't the genius. It's the owner.

Two very different new world orders

Cheap cognitive labor doesn't come with instructions for who gets it.

The same technology builds two completely different worlds, and the variable is who holds the controls.

One world: AI workforces go to whoever already has the capital, the customers, and the legal team. Big firms get wildly more productive with a fraction of the staff. A few thousand people pocket the trillions.

Which is exactly what Amodei means when he admits his own optimistic story risks "a few individuals capturing trillions," and exactly the inequality the IMF is bracing for.

That's AGI driving the largest concentration of wealth in history, and the headline of the 2030s is: the machines got good and the owners kept the money.

Now, the other world: the AI workforce is something an ordinary person owns, like a utility.

Someone in any country runs a company's worth of AI labor and keeps what it makes. Same productivity explosion, but the ownership spreads across billions instead of pooling in thousands.

That's AGI as the largest spread of opportunity in history, and the headline is: AI gave everyone a company.

Same models, same growth, opposite civilizations. What decides it isn't the technology.

It's whether anyone focuses on building the layer that puts the controls in normal hands.

Right now, the industry is overwhelmingly building copilots for people who already have jobs and tools for companies that already exist.

The first world on autopilot. The second world is mostly a product nobody has shipped yet.

That's the one Voltaire aims to be building.

Why the poorest countries are the front line, not the afterthought

The lazy assumption is that this lands in San Francisco first and trickles down.

The data says the opposite, and it might be the most underrated fact in the whole conversation.

Anthropic's Economic Index reads anonymized usage at real scale, and it found that people in lower-income countries already lean on AI more for full automation (hand it the task and take the output) while people in rich countries lean on it for augmentation, working alongside it to get better at the job they already have.

Think about why. If you've got a team, a title, and a paycheck, you want a copilot that makes you sharper.

If you've got none of that, you don't want a copilot. You want a worker. You want the team you never had.

So the people with the least are the ones who most want to hand over the whole thing, because they're starting from zero; they have no existing job for AI to politely assist with. That's the leapfrog.

Africa skipped landlines and went straight to mobile money; the next jump skips "AI that helps you at work" and goes straight to "AI that runs the business you could never have afforded to start."

Around 49% of occupations already have at least a quarter of their tasks done with one AI assistant (Anthropic, 2026)…. and that's with today's still rather clumsy models. The raw capability is showing up everywhere all at once.

What's missing most where people have the least is the layer that turns that capability into an owned, running, paying business.

What we're actually building

One layer that gives a single ordinary person command of an AI workforce and ownership of the company it runs. Not a copilot stapled to a job. Not software sold to firms that already exist.

A company in your pocket, for the billions who never had one.

Up close, today, it looks almost modest, and it should.

We run real businesses for real people right now: a side hustle for a teacher, a little shop for a retiree.

We take a cut of what they earn, so we only win when they win. That's the launch business that funds the moonshot.

And every business we run, start to finish, teaches our AI workforce what actually makes money out in the messy real economy.

That hard-won knowledge is the rocket we're really building.

The teacher's side hustle is the reusable booster.

The destination is a billion people owning a company.

The destination also drives our near-term calls. It's why we won't be a copilot for existing firms even though that's the more obvious sale, because that road just builds the existing concentration world.

It's why the individual owns the product, and Voltaire is paid on their revenue instead of renting it to their boss.

A different mission picks different fights. Mars makes you build reusable rockets.

This one makes you build the controls to put businesses in everyone's hands.

Controlling this

A bold bet only deserves attention if it's clear about how it stalls.

The fork could close the wrong way. Regulation, payment rails, and platform power could trap AI workforces within big incumbents and keep them out of ordinary hands. The concentration world, arriving by policy rather than physics.

That's the outcome Voltaire exists to fight, and the fight is mostly a race to build the distributed version before people accept the concentrated one as the norm.

And the dark mirror is real.

The same tool that lets someone in Lagos run a legitimate global company lets them run a scam, and the criminal economy is already the best-funded customer for exactly this.

Which is the whole reason the controls have to come from someone whose model is legitimate, aligned, and truthful, instead of being left to drift wherever the market points.

The guardrails aren't a tax on the mission. They are the mission. A billion "owned businesses" are only worth building if they're real… and legal.

Our bet

We're picking an optimistic future for AI-run businesses.

Where the most powerful productive force ever made doesn't pool in a few thousand hands but lands in a few billion.

Where a kid in Nairobi and Bezos's "nurse in Queens" command the kind of workforce that used to cost a fortune.

Where the question stops being "can I get a job" and turns into "what do I want to build."

That world isn't guaranteed, and it isn't close.

It's Mars.

And you get to Mars by launching now.

I'm Fabrice, and I'm building Voltaire - your own business, built and run by AI.