A Billion New Jobs.

In 1956, shipping containers cut cargo loading costs 36x. The shipping industry didn't die. It exploded. The bottleneck just moved to ports, rail, and warehouses. AI is doing the same thing to software right now. The constraint is shifting from code to chips (NVIDIA, $4T), to electricity (Musk says space in 36 months), to a thousand places nobody's watching yet: biosecurity, cow collars, cement, weed-pulling robots. Human demand is infinite. Solve one problem, ten more appear. That's a billion new jobs.

In 1956, loading cargo onto a ship cost $5.86 per ton. Ships sat in port for weeks while dockworkers moved crates by hand. Then a trucking entrepreneur named Malcolm McLean had a simple idea: put everything in a standard-sized metal box. The cost dropped to 16 cents per ton. A 36x reduction.

Did the shipping industry collapse? No. The opposite happened. Containerization blew open a bottleneck that had constrained global trade for centuries. And when that bottleneck broke, the constraint didn’t vanish. It moved. Suddenly the world needed bigger ports, deeper harbors, inland rail networks, massive warehouses, inventory management systems, customs software, cold chain logistics. Entirely new industries appeared. Millions of jobs were created that couldn’t have existed before, because they were hiding behind the old bottleneck.

This is always how it works. You remove a constraint in one part of a system, and the bottleneck shifts somewhere else. Wherever it lands, that’s where the new value concentrates.

We’re living through this right now with AI. And I don’t think most people grasp the scale of what’s coming.

One of our clients at TabAI is a $10 billion revenue manufacturer. Their product? Car parts like steering wheels.

Most people hear that and think: okay, a factory makes a round thing that turns a car. But that $10 billion should stop you. Steering wheels are one component of one part of one vehicle, and the pipeline behind them is enormous.

Before a steering wheel exists, it’s a problem in material science. The right polymer blend, the right steel alloy, the right foam density. That’s chemistry, testing, sourcing raw materials from mines and chemical plants across three continents.

Then it’s a manufacturing problem. Injection molding, CNC machining, quality assurance. Thousands of decisions on a factory floor.

Then logistics. Packaging, freight, customs, warehousing, just-in-time delivery to an assembly line in another country. The global logistics industry is worth over $11 trillion.

Then software. The steering wheel talks to the car now. Sensors, haptic feedback, heated elements, cruise control, lane assist. Firmware, integration testing, OTA updates.

Then sales. Installation. Warranty. Spare parts (the global auto parts market is over $2 trillion). Recycling. Waste management.

One steering wheel. Dozens of pipelines. Each pipeline is its own universe. Material science alone breaks into polymer research, supplier negotiations, environmental compliance, lab testing, patent strategy. The advanced materials market is $90+ billion and growing 6% a year. Global manufacturing output is approaching $15 trillion. That’s not software. That’s atoms.

They look at one slice of the pipeline, usually the software slice, and say AI is going to replace it. But that’s not the end of the story. That’s the beginning.

As Marc Andreessen has pointed out: ask any friend who’s a lawyer, accountant, coder, or finance person with access to AI tools, and ask if they’re working more or less. Everyone’s working more. That’s not a paradox. That’s the pattern.

People who say AI will replace all jobs forget that the world isn’t running on productivity B2B SaaS. The core lesson of economics is that human demand is effectively infinite. That’s the whole reason supply and demand curves exist. There is always more to want, more to build, more to solve. When you remove a bottleneck, you don’t shrink the total work. You expand it. You make visible the problems that were hiding behind the old constraint.

Look at NVIDIA. They control over 80% of the AI chip market and were the first company to hit a $4 trillion market cap. AI blew open the software bottleneck, and the constraint moved to hardware. The scarce resource stopped being code and started being compute. NVIDIA was sitting on the new constraint.

But the constraint will move again. Just like it did after the shipping container.

Everyone’s focused on Harvey, Rogo, Databricks. Those are real companies. But they’re all software eating software. The much bigger story is what happens when AI hits the other 90% of the economy.

It’s already happening. Valthos is a nine-person biosecurity startup that raised $30 million from OpenAI and Founders Fund to build AI that scans biological data and detects engineered pathogens before outbreaks spread. The founding team came from Palantir, DeepMind, and the Broad Institute. That company couldn’t have existed five years ago. The AI wasn’t good enough. Now that the software constraint has eased, the bottleneck moved to biosurveillance, and Valthos is sitting right on top of it.

Arena Physica is building what they call electromagnetic superintelligence. They help hardware engineers figure out why training runs fail by combining LLM reasoning with physics-based simulation. They’re backed by RRE, Eniac, and RTP Global, and already collaborating with AMD. Their market? The trillion-dollar semiconductor supply chain. Not SaaS. Physics.

Then there’s Halter, a New Zealand company making AI-powered collars for cows. That sounds absurd until you learn they’re valued at $2 billion and backed by Peter Thiel’s Founders Fund. Their solar-powered collars create virtual fences, monitor herd health, and let ranchers move cattle from a phone. Five to eight dollars per animal per month. There are roughly a billion cattle on earth. That’s the kind of market that opens up when the constraint shifts from “can we build the software?” to “can we deploy it in the field?”

And it goes deeper. Red Barn Robotics (YC W25) is building autonomous robots that travel through crop rows and mechanically pull weeds while leaving crops untouched. Bindwell (also YC) is mapping pest proteins to design better pesticides at the molecular level. Concrete.ai is using AI to optimize cement manufacturing, one of the dirtiest industries on the planet. Hauler Hero just raised $16 million to bring AI to waste hauling logistics. CattleEye uses a single camera above a walkway to monitor the health of 200,000+ cows by analyzing how they walk.

None of these are productivity SaaS. They’re AI applied to atoms, animals, chemistry, dirt, concrete, and electromagnetic fields. Each one sits on a different part of a different pipeline where the constraint just shifted.

Software used to be the bottleneck. AI is blowing it open. So now the constraint moves to hardware deployment, logistics capacity, regulatory infrastructure, human expertise, physical systems that don’t exist yet. There are a trillion problems we haven’t begun to fathom, sitting in every part of these pipelines.

This is what happened with the internet. Social media manager wasn’t a job in 1995. Neither was cloud architect, UX researcher, or growth engineer. AI will do the same, except the non-software economy it’s about to transform is vastly larger. Manufacturing is $15 trillion. Logistics is $11 trillion. These are massive sectors with deep pipelines only now starting to feel AI-driven acceleration.

So when someone says AI will eliminate a billion jobs, ask them which part of the pipeline they’re looking at. Because I see a billion new ones opening up and a trillion new problems to solve.

Constraints always move. So does opportunity.