jump to...

Dependency is not a feeling. It’s an investment decision.

After the frontier-access essay went out, the responses kept circling one feeling.

Not a technical objection. A feeling. If someone else’s hand is on the tap, then we should own a tap of our own.

It sounds like prudence. It is also the most expensive feeling a company, or a country, can act on.

A familiar reflex

I have seen this feeling before, in older costumes. For two decades I watched governments announce national operating systems so they would never depend on foreign software. Step outside software and the list writes itself: energy independence, food security, oil. The reflex is always the same. Needing someone is dangerous, so do everything yourself.

And let’s be fair: the anxiety has receipts. The last essay described a government switching off model access by passport. Nobody gets to call this fear irrational anymore.

But being right to worry and letting the worry decide are two different things. The first is unavoidable. The second is a choice.

The breakfast test

Scale it down from nation-states to a breakfast table. Eggs, milk, bread. All of it arrives from outside: the eggs from the corner store, the milk from the market, the bread from the bakery. If dependency itself is the threat, the response is obvious. Chickens in the yard, a cow beside them, a mill in the garage. While you’re raising chickens, your neighbor buys eggs in two minutes and spends the morning on work that pays for a month of breakfasts. Your egg is independent. It is also the most expensive egg in the world.

So why does nobody fear the corner store? The dependency is right there. Because three more stores sit down the street, an egg is an egg everywhere, and switching costs nothing. Which reveals the actual fear: single-source dependency, with no substitute, that can be cut overnight.

The test comes down to three questions. How many suppliers exist? Do they all answer to the same authority? What does switching cost? Eggs pass easily. Frontier models don’t. A handful of labs, concentrated under two governments’ law and, on the evidence of June, two governments’ mood.

What the math says

Anxiety’s formula is short: dependency is bad, build everything. Strategy’s formula is longer and, I admit, more boring. What is the probability of a cutoff? What does it cost me if it happens? What does building my own cost, and over how many years does it pay back? And the question everyone skips: what would the same capital and the same talent produce somewhere else?

Economists call that last one opportunity cost. I prefer to ask it plainly: what didn’t get built during the three years we spent building the mill? A fully self-sufficient operation can be assembled, in theory. Whether it can be built was never the hard part. The hard part is the invoice for everything you gave up along the way.

The SpaceX objection

I can hear the first objection. Buying everything is a dogma too. Look at Tesla. Look at SpaceX. Fair. Business schools taught a generation to focus on the core and outsource the rest; then SpaceX built rockets down to the screws and collapsed launch costs. Vertical integration wins too. Both models have famous victories and quiet wreckage.

But notice the detail that gets skipped. SpaceX didn’t integrate because it feared suppliers. It ran the numbers; the aerospace supply chain could not hit the cost or the iteration speed the plan required. The analysis said build, so they built. The same analysis tells another company to buy, and it buys. Both are legitimate. What separates them is who made the decision: the fear or the arithmetic.

The drone objection

If you grew up where I did, you have been holding a counterexample for three paragraphs now. Turkey spent decades buying critical defense systems from allies, got embargoed at the worst possible moments, built its own drone industry, and turned it into both strategic autonomy and an export business. I know. I watched it happen.

That story doesn’t break the framework. It passes it, with distinction. The probability of cutoff wasn’t a hypothesis; it had been demonstrated, repeatedly, by actual embargoes. Substitutes weren’t available at any price when it mattered most. And the investment returned, strategically and commercially. That decision was right because it came from the arithmetic, not in spite of it.

This is what a framework worth keeping looks like: the same math sometimes says build and sometimes says buy. Fear only ever gives one answer. Trust no one, make everything. One of these is a compass. The other is a reflex.

A dial, not a switch

Back to AI. The choice was never binary. There’s a dial. At one end, you wire everything to a single provider: the cheapest position and the most exposed. One click over, you spread the load across providers and keep your switching costs deliberately low. Another click, you run open-weight models on hardware you own, which is exactly the floor the last essay was about: capable intelligence you can hold in your own hands. At the far end sits training your own frontier model, a position that demands enormous compute, enormous capital, and the scarcest input of all, talent.

Every position on the dial charges a premium and pays out insurance. Strategy means setting the dial by risk and budget. Setting it by identity is how the egg ends up costing a fortune. For most companies, and frankly for most countries, the right answer sits somewhere in the middle. Sovereignty is a portfolio position, not a personality.

Naming it

When national identity enters the decision, the arithmetic quietly leaves the room. What settles into its place deserves a name. I call it trust poverty: a scarcity of trust in the future, the lived sense that nobody’s word can be treated as collateral, so the only safe posture is to hold everything yourself.

If you grew up in the United States, that sense can be hard to reach, so try it through your family. Anyone raised by people who lived through the Depression has seen trust poverty at the kitchen table: cash kept in envelopes, a distrust of banks that outlived the bank failures by fifty years. Or reach back to March 2020, when the shelves emptied and the world’s richest country discovered it could not buy masks at any price. The semiconductor push that followed, tens of billions of public money to bring chipmaking home, is the same feeling in American clothes. Now imagine it not as one bad year but as the background weather of your whole economic life: a currency that thins between paychecks, savings that vanish in a bank crisis, contracts that quietly stop meaning things. That is the soil this feeling grows in. Once you have lived in it, holding everything yourself stops looking paranoid and starts looking like hygiene.

Companies run a smaller version of the same deficit. Teams that rewrite every library in-house. Platforms rebuilt from scratch because the vendor “can’t be trusted.” The syndrome even wears a name badge: not invented here. The costume changes. The feeling underneath doesn’t.

The interest rate on fear

Now the bitter part. The side that feels this anxiety most intensely is usually the side with the least room for error. A wealthy economy can absorb a vanity project; it buries a few billion in prestige, shakes itself off, moves on. A resource-constrained economy gets no such grace. Every misallocated billion there is a school that didn’t open, a factory that didn’t get built, a generation of engineers who didn’t get trained. The thinner the wallet, the higher the interest rate on emotional decisions.

Which is why the sentence I would frame on the wall is this one: strategic thinking is not a luxury of the rich. It is an obligation of the resource-constrained.

The last essay ended by telling you to always know whose hand is on the tap. Let me add the second half. Knowing whose hand is on the tap is strategy. Insisting on casting every tap yourself is anxiety. Between the two stands nothing but arithmetic.

Where is your dial set right now? And what set it there: the analysis, or the feeling?

Daron Yondem advises senior technology leaders on AI-driven organizational transformation. Learn more →