Every board asks the wrong question: how confident are you?
Confidence is a terrible investment criterion. History is full of executives who were catastrophically wrong with total conviction. Confidence isn’t insight. It’s charisma, or a good run, or enough distance from the downside that you stop feeling it.
The question that actually separates enduring companies from transient ones is different.
What must be true, regardless of which future arrives?
Leaders who consistently arrive early aren’t better forecasters. They’re better students of what the present can no longer sustain.
The Bankruptcy of Point-Forecasting
Corporate strategy is addicted to prediction. Five-year models. TAM slides. Roadmaps. Projected IRR. All useful, most of the time. At the exact moments that matter, all of it fails.
Forecasts fail because they extend yesterday’s line into a future that doesn’t move in lines. Roadmaps fail because what you ship looks nothing like what you pitched. TAM estimates fail because you can’t size a market that doesn’t exist yet with data from the market that did.
Rejecting the forecast doesn’t mean rejecting prediction. It means moving the bet from the product to the economics underneath it.
The trap is mistaking a constraint that’s about to be engineered away for one that can’t be.
Skeptics will say this is post-hoc rationalization — that invariants only look obvious in the rearview mirror, and every failed bet once called itself a permanent law too. Fair challenge. So here’s a test you can run before the market resolves it, not after.
The Veto Test
Every durable constraint survives because something outside the technology itself keeps it alive — physics, institutions, or coordination. The three questions below are just a way to find that force before the market prices it in.
Ask them of any constraint you’re tempted to call permanent.
How many independent parties would have to move together to dissolve it? One — a single engineering team ships a better algorithm — and assume it’s temporary. Several, none of whom report to each other — a regulator, an insurer, a court, a standards body all have to align — and you’re looking at something with real durability, at least on a multi-year clock.
Does anyone with power over the constraint actually benefit from keeping it? Bandwidth scarcity had no defender. Carriers wanted more usage, not less; nothing with leverage over the constraint was rooting for it to survive, so once the engineering existed, it collapsed fast. Contrast that with liability. Insurers, plaintiffs’ bars, and licensing boards all have a business model that depends on accountability staying somewhere identifiable. A constraint with an entrenched defender has inertia a pure engineering gap never does.
Is the fix an improvement in the technology that’s becoming abundant, or does dissolving the constraint require a resource that sits off that improvement curve entirely? Things on the curve get cheaper on schedule. Physical infrastructure and institutional coordination usually sit off it.
Run bandwidth through this and it fails all three: one veto-holder, no defender, pure engineering fix. Run data sovereignty through it around 2010 and it’s closer — several parties, real friction, but still no defender fighting to keep data landlocked once the engineering existed elsewhere. It was never going to hold forever; the test tells you how much scrutiny a claim deserves before you commit capital.
Real conviction doesn’t come from guessing which limitation feels permanent today. It comes from finding what’s still standing after engineers have had a decade to attack it.
Containerization makes the same point from the other side, and it’s worth running the test forward. Ship size has one veto-holder — whichever yard builds the next hull — no entrenched defender, and it sits on shipbuilding’s own improvement curve: a false invariant, visible as one before the fact. Ports, rail, and customs scheduling have several veto-holders who don’t report to each other, an entrenched defender in every port authority protecting its place in the trade lanes, and infrastructure that sits off that curve entirely. Run the test in 1965 and it points at the chokepoint, not the hull, years before the outcome was obvious.
The Law of Migrating Scarcity
Markets don’t pay for abundance. They pay for whatever abundance can’t eliminate. Technology only destroys the constraints that sit on its own improvement curve — everything that endures sits somewhere else. Migrating Scarcity tells you where value goes next. The veto test is just the instrument that tells you whether the scarcity you’re eyeing is actually durable enough to bet on.
The claim is narrower than “value migrates to scarcity.” Durable bets are the ones where the surviving scarcity can’t be eliminated by improvements in the technology that’s becoming abundant. Sometimes that surviving scarcity is institutional — liability, licensing, permission. Sometimes it’s physical — a port, a fab, a grid. Both are outside the reach of the thing getting cheaper.
AWS wasn’t a bet on hosting. It was a bet that software velocity would stay bottlenecked on physical server provisioning regardless of what got built on top — land, power, permitting, capital, none of which gets easier just because code does. The constraint sat off the curve, so the value moved there. NVIDIA’s early push into CUDA wasn’t a bet on deep learning. It was a bet that an expanding class of computational problems would keep outgrowing sequential architectures, regardless of which application eventually created the demand — a hardware limit no amount of better software was going to route around. Same question, both times: once this gets cheap, what still sits off the curve? AI is the live version of that question.
Run AI through that filter and be honest with yourself. Raw intelligence is collapsing in cost. So are generic guardrails and compliance copilots — commoditizing on the same schedule as the models. None of that is the invariant. Novelty is often mistaken for durability.
What doesn’t collapse is who’s on the hook when an autonomous system gets a regulated decision wrong — credit, clinical triage, capital allocation. Nobody sues the model; the regulator and the court go after whoever was running the process.
Software can produce a verifiable, immutable log of what happened. It cannot absorb the consequence of being wrong. That has to sit somewhere, with someone — and that’s where the value lands.
Run this through the same test rather than just asserting it. Veto-holders: regulators, insurers, courts, and licensing bodies all have to move together — several, not one. Defenders: insurers and licensing regimes profit from accountability staying identifiable. Fix type: legal reallocation, not a technology improvement. It clears all three, which is a different claim than “trust me, it feels permanent.”
This framework can fail. If regulation follows the path of Section 230, liability could dissipate rather than concentrate — and institutional accountability would be a far weaker moat than this essay argues. Watch legislatures and insurance filings, not court dockets: a blanket safe harbor shows up as a statute, a concentrating one shows up as a license.
Staging Capital Behind Inevitability
Identifying an invariant tells you where to bet. It doesn’t tell you when to size the bet.
Markets pay you for prediction only in hindsight. In real time, they price inevitability the moment it becomes obvious — and by the time it’s obvious to everyone, you’re paying retail for something you should have owned early.
But finding the right invariant doesn’t remove the timing problem. Being five years early looks identical to being wrong, at least on the P&L. Conviction tells you where the value lands. It doesn’t tell you when to deploy — and that needs an actual mechanism, not a hope.
Two signals, staged:
Deploy exploratory capital once a core cost — inference per complex workflow, say — crosses a threshold where mass deployment stops being a maybe and starts being arithmetic.
Deploy at scale only when clients start actually moving liability onto you in the contract. Real indemnification. Outcome-based pricing. Not a service-credit slap on the wrist for downtime. Watch who holds the leverage here — clients often demand indemnification before a vendor can actually price and carry it, and taking on liability you can’t afford is the same mistake as deploying too early.
Move before the second signal, and you’re funding the market’s growing pains on your own balance sheet. Wait past it, and you’re buying in at the price everyone else already figured out.
Strategy isn’t calling the future correctly. It’s being honest about which parts of today are already unsustainable, finding what survives the shakeout, and having the discipline to stage the bet instead of dumping it in all at once.
Boards keep asking the wrong question. Next time someone asks how confident you are, ask instead: what must still be true if every prediction on this slide turns out to be wrong? That’s where durable strategy begins.
Before you approve the next strategic investment, delete the forecast slide and ask one question instead: if this technology gets 100x cheaper, what constraint still controls the outcome?
The market doesn’t reward the people who guessed the future. It rewards the people who identified the constraints the future couldn’t escape.
And Vijay Says... 