Software Doesn't Ship Through the Strait of Hormuz
AI makes domain expertise exportable at scale, without shipping anything at all.
Until the Iran war, most New Zealanders probably had never heard of the Strait of Hormuz.
Now it is in the news every day.
Not because we're close to it. Because too much of our economy depends on physical flows we don't control. Fuel, shipping, fertiliser inputs, export routes. When a distant chokepoint seizes up, the effects show up here quickly.
Every small, trade-exposed economy has a version of this problem. New Zealand's version is sharper than most. Our export economy runs on things that grow, things that graze, and people who visit. That built a lot of the country. It also leaves us exposed.
A drought hits dairy. A cyclone batters the upper North Island. A political shock in Beijing hits demand. A war in the Middle East pushes up energy and freight costs through a waterway on the other side of the world. Too much of New Zealand's export income depends on atoms moving through vulnerable systems.
Software does not. That is not an argument against farming, tourism, or any of the industries that built the country. It is an argument for building something beside them that is exposed to different risks.
In the first post I wrote for this publication, I argued that AI gives small countries a short window to produce many more founders. That was the founder argument. This is the export version. If New Zealand is serious about diversification, the interesting thing about AI isn't just that existing companies can use it to trim headcount or add features. It's that AI makes a second export pillar possible, one that doesn't depend on weather, shipping lanes, or the political mood in a handful of foreign capitals.
Why previous attempts didn't stick
New Zealand has produced globally successful technology companies. Xero, Rocket Lab, Halter. But they are outliers, not a dense base of exporters.
The standard knowledge economy thesis was not crazy. It was just built on economics that didn't fit us. To build a serious software company from New Zealand, you usually needed offshore venture capital, a meaningful engineering team, and eventually some move toward larger markets. Some good companies came out of that model. A lot of the value didn't stay here.
Paul Callaghan, the late physicist and probably the most influential voice New Zealand has had on science-led economic development, had the better framing. He argued the country needed a hundred high-tech companies each doing serious export revenue. He wasn't trying to copy Silicon Valley. He was right about the direction. The problem was that the cost structure still assumed capital and headcount we didn't have at scale.
What's structurally different now
AI doesn't just make programmers faster. It cuts the formation cost of a company to the bone.
A product that once needed a meaningful team and millions of dollars can now be built by a few people with APIs, AI tools, and almost no external capital. Just as important, AI compresses the operating overhead that used to force early hiring. Support, bookkeeping, back-office admin, internal tooling, compliance grunt work, first-draft sales material. The stuff that used to require a company before you had much of a business.
For New Zealand, that matters because our barriers were stacked. Not enough capital. Not enough engineering depth. Too much operating overhead too early. Too much need to assemble a conventional company before the business had really earned it. Previous technology waves lowered one or two of those barriers. AI hits all four at once.
That does not guarantee anything. It does change the export maths. The old model could produce a few outliers. The new model can produce a much denser base of small and mid-sized exporters underneath them. And because these companies sell software, services, and systems rather than physical goods, they are much less correlated with the export risks we already have. No harvest risk. No container bottleneck. No dependence on a port.
This also goes wider than software founders in the traditional sense. The pattern is not just "learn to code and build an app." It is: pair real domain expertise with AI that handles a lot of the operating overhead, then sell globally from New Zealand. A team of two in a niche category can now look more like a small export business than a side project.
This time really might be different
"This time is different" is usually the most expensive sentence in investing. The claim fails because people project sentiment rather than trace mechanics. They feel like something is big, so they assume the outcomes will be big. When the excitement fades, the underlying economics haven't moved.
I want to be careful here, because on the surface this argument sounds exactly like that. So it's worth checking it against what actually changed in the cost structure, not the excitement.
The dot-com wave lowered distribution costs. You could reach customers online for the first time. But building the product still took teams of engineers and millions in capital. New Zealand founders still had to fly to Sand Hill Road. Cloud computing lowered infrastructure costs. You didn't need to buy servers. But you still needed a team of twenty to build anything serious. Mobile lowered access costs. Everyone had a computer in their pocket. But the cost of building, hiring, and scaling didn't move.
Each wave removed one barrier and left the others intact. That's why the knowledge economy prescription kept failing here. The advice was right. The preconditions didn't exist. NZ founders still needed offshore VC, still needed to hire engineers they couldn't find locally, still needed more organisational mass than a small country could easily produce. The barriers that specifically disadvantaged New Zealand survived every previous technology wave intact.
AI is the first wave that changes all four of the formation constraints that mattered most here. The capital required to start drops sharply. The amount of specialised engineering labour required to build drops too. The operating headcount required to run the business drops because AI absorbs work that used to force early hiring. And the pool of viable founders widens: domain experts can now turn judgment and workflow knowledge into products and services without first assembling a conventional software company.
That does not eliminate distance. It does make distance less punishing, because founders need less offshore capital, less concentrated talent, and less organisational mass before they can sell something globally. It also changes investor fit. New Zealand's angel-heavy, VC-light landscape used to be a weakness. Now it looks much more like a match for capital-light companies that do not need multiple rounds of funding before they become real businesses.
I'm not asking anyone to take "this time is different" on faith. I'm asking them to check the arithmetic. If the cost of building has dropped by an order of magnitude, if the team needed to build and run the company is much smaller, and if the set of people who can credibly found one is much wider, does the old pattern hold? Or did the thing that was blocking it actually move?
Maybe it didn't. Maybe there's a barrier I'm not seeing, and the honest problems section below lays out several candidates. But the objection has to be specific. It can't just be "we've heard this before," because what we heard before was a different argument about a different cost structure, and the barriers that killed it are demonstrably lower now.
What a second export pillar actually looks like
The right ambition isn't one giant national champion and a ministerial speech about it. It's a ladder. Roughly:
- 3,000 companies at $500k in export revenue - 1,000 companies at $2 million - 250 companies at $10 million - 50 companies at $50 million - 15 companies at $100 million
That lands in the same territory as Callaghan's $10 billion target, reached through a different shape of economy. To be clear about what this refers to: new companies, started from now, built on AI from day one. The existing tech sector, the companies that built New Zealand's tech scene and proved it could be done from here, will keep growing and adapting in their own right. That matters, and nothing in this argument diminishes it. But the ladder is measuring something additional. A new generation of AI-native companies that don't exist yet, built clean from the ground up, without the legacy architecture that makes retrofitting so hard. That is where the structural export opportunity sits. Don't read the specific numbers as a forecast. Read them as a way of seeing the shape of the opportunity. The point is the distribution, not whether any individual rung is exactly right.
A ladder beats a lottery. If ten small exporters fail, nothing breaks nationally. If the base is broad enough, the category compounds anyway. Stats NZ already counts thousands of small tech firms in New Zealand. This ladder doesn't require all of them to become major exporters. It only requires a minority to turn outward, and a handful to become very large.
The second part matters more than the headline number: what stays here. A company doing $10 million in exports isn't the same national outcome if most of the cap table is offshore, the IP lives elsewhere, and the eventual profits are extracted. The real prize is local capture. Founder ownership. Local payroll. Tax residency. Profits that compound inside the country. AI-native companies have a better shot at this than the old VC-heavy model because the capital requirements are lower. That's not a guarantee. It's a better starting position.
What this looks like in practice is many quiet export businesses. A founder in Tauranga selling compliance tooling into American healthcare. A team in Wellington building workflow software for European fintechs. A finance operator in Christchurch turning years of ugly back-office knowledge into a product sold offshore. A domain expert in a narrow category building a global business with two people and AI. A few dozen loud wins, and underneath them, a much larger base.
That model is less visible than a single IPO. It's also more resilient. The point of a second export pillar is not symbolism. It's density.
The honest problems
Even if the arithmetic is right, several things could still break this.
Employment. This does not replace dairy-scale employment. It creates high-value jobs, founder income, and tax revenue, but with far fewer people attached than the sectors it would sit beside. Export diversification is not the same thing as employment diversification.
I flagged that tension in the first post and I still don't have a clean answer. What I do know is that the entry-level knowledge work ladder is already under pressure. This model does not solve that. The problem is not just fewer jobs. It is fewer first jobs, fewer supervised reps, and fewer places for younger people to build judgment under someone else's roof.
It is also not an argument against the first pillar. Dairy, meat, horticulture, and tourism still matter. They support whole towns in a way a three-person software exporter does not. The case here is narrower: New Zealand needs a second pillar with different exposures, sitting beside the first one rather than pretending to replace it.
Capture. Software is lighter than commodities, but it is not sovereign by default. The infrastructure is mostly American, the pricing is mostly in USD, and the default gravity still pulls successful companies offshore: capital, ownership, and holding structures that shift the value away from New Zealand. If the IP, profits, and tax base leave too, the country can still do the hard part and miss a lot of the upside.
Go-to-market. AI has changed product formation faster than it has changed trust formation. A founder in Auckland can have a product working by Friday and still be months away from the first serious offshore customer. That is why domain expertise matters so much here. The operator with fifteen years in a niche starts with language, trust, and judgment that a generic tech founder does not.
Scale is aspirational. A $10 billion ladder is a compelling target. It is not today's reality. Right now, we are nowhere near three thousand meaningful exporters, and the $100 million rung is still tiny. The category also will not appear on its own. It needs more founders thinking globally from day one, more operators treating AI as business infrastructure rather than a productivity toy, and more local capital that understands what a capital-light export business looks like.
My usual disclosure: I'm building exactly this kind of company. Dolla is an AI-native accounts payable product with zero employees, global ambitions, and the kind of margin structure I've described. So this argument is self-serving. But the macro case does not depend on my company specifically. It depends on whether the conditions exist for many more companies like it. I think they do. My own part in this is narrow: build the first few companies, see what holds up, and make the pattern visible if it does.
If this becomes real, it will be because a lot of people push one small part of it forward from where they already stand. An investor backs a capital-light exporter. A policymaker removes one stupid friction. An operator turns fifteen years of niche knowledge into something worth selling worldwide.
If you are that operator, here's what I'd suggest. Over the next week, pay attention to the work around you differently. Notice the places where people spend their days pulling context out of messy inputs like emails, PDFs, phone calls, and scribbled notes, making sense of it, and pushing it into a system of record. That translation work, the gap between unstructured reality and structured software, is where most of the next wave of companies will come from. Dolla, the company I run, does exactly this for accounts payable: invoices arrive as messy emails, AI extracts the context, learns each customer's patterns over time, and codes them into accounting software. That pattern, extracting context from chaos, building memory, and acting on structured systems, repeats across hundreds of industries. If you start noticing it in yours, you're probably looking at a company. The $10 billion number earlier in this post is built from thousands of problems exactly that small.
For forty years we've said we should diversify into higher-value exports. Fair enough. But for most of that time the machinery required to do it at scale did not exist.
Now it does. And the next time a strait closes or a drought hits or a trading partner changes its mind, the question will be whether we built anything that kept earning while the physical economy absorbed the shock.
If you are looking for the most important export opportunity in front of New Zealand, start with the one that does not need a port.
If you're a domain expert and you keep seeing a painful, repetitive, judgment-heavy workflow that nobody outside your industry understands, tell me about it.
You don't need a polished startup idea. A few paragraphs or a bullet list is plenty.
Tell me what the work is, who does it today, why it's painful, and what your relationship to it is. This is especially true if you're in regional New Zealand, or you're a Kiwi overseas wondering whether the economics of coming home have changed.
Send it to ben@thinkdorepeat.ai - If I can be useful, I'll come back with an honest view: what I see, what I'd pressure-test, and what I'd do first.
I'm Ben Lynch. I write about founders, AI, and what happens next from New Zealand. Say hello at ben@thinkdorepeat.ai.
If you're new here, Start Here is the best place to begin.
If this frame helps you see one piece you can move, send it to someone else who might move theirs.