If you value AI upstream, judge how geopolitical risk discounts into compute assets, or are eyeing the sovereign-AI budget — here’s how politics reprices upstream: it doesn’t change the direction profit flees toward scarcity, but it installs a gate on who is eligible to own the scarcity.

Physics sets how deep upstream scarcity runs (last chapter); politics sets who gets to put a shovel in. This is industry-and-capital analysis only — no judgment on any country’s policy, just politics as a repricing mechanism rewriting the profit map.

Export Controls: Split Into Two Disconnected Pricing Systems

Export controls are easiest to misread as “embargoing a product.” The real economic effect is far bigger: they split one global market into two or more disconnected pricing systems.

On one side of the wall, the most advanced compute runs at one price, one supply curve; on the other, the controlled party either pays a steep “geopolitical premium” (acquiring restricted compute through indirect channels) or gets locked into sub-optimal, last-gen, or domestic-substitute compute. The same chip is not the same product on the two sides of the wall — its availability, price, and the competitiveness it buys are set by a political boundary.

The implication for upstream profit runs both ways: for suppliers inside the wall, controls are a moat (artificially walling the strongest rival out of the most advanced market, extending pricing power — engineering scarcity reinforced by politics); for upstream as a whole, controls are market fragmentation (global compute is no longer one uniformly-valued pool but several camp-divided pools). So you can no longer price upstream on a “single global market” assumption — that market has been politically cut. Model it by camp.

Chokepoints: A Tail Risk You Can’t Probability-Price

The AI hardware chain has a few chokepoints: EUV lithography (one global supplier), advanced packaging, leading-edge foundry capacity — and one critical link is geographically concentrated in a geopolitically sensitive zone. A chokepoint’s trait: normally just a supply-chain link, but once a geopolitical conflict triggers, it can drive one side’s supply toward zero in short order.

This risk has a property that troubles every valuation model: its probability distribution contains a discontinuous cliff. Normal business risk is continuous — demand swings 10%, price drops 20%, all fit a discount model. But “chokepoint severed” isn’t continuous: near-zero risk most of the time, catastrophic and nearly binary once it happens. You can’t give it a clean “probability × loss” number, because the loss magnitude at the tail is a cliff.

So chokepoint risk should be handled mainly by exposure management and stress testing: you manage not the expected value of “will it happen” but the survivability of “if it happens, can the portfolio live.” For the upstream names most dependent on a single chokepoint, however tempting the scarcity premium, apply a tail discount that’s unquantifiable but must be managed.

Sovereign AI: A Buyer That Doesn’t Look at ROI

Politics’ third repricing mechanism creates a wholly new buyer: the state. More and more states treat “owning your own AI compute and models” as a strategic necessity — sovereign AI. Its logic isn’t commercial, it’s national-power: in an era where intelligence amplifies economic, military, intelligence, and governance capability, compute is seen as a 21st-century state capability. A state stockpiling GPUs isn’t stockpiling chips — it’s stockpiling national capability.

This buyer is price-insensitive, ROI-insensitive. A company buying compute computes a return; a state buying compute computes strategic autonomy and buys even when it doesn’t pencil out commercially. Three effects:

• A non-economic floor under the upstream profit pool: when commercial buyers pull back because ROI won’t pencil, sovereign buyers may still be buying, holding up part of upstream’s downside — a realer, steadier pool.
• A prolonger of the capital cycle: and more stubborn than circular financing — it doesn’t decide on price signals, so it won’t stop when price discovery breaks. Upstream’s boom may last longer than pure commercial logic could support.
• But also a geopolitical landmine: a sovereign buyer’s demand and supply both hinge on political relations; today’s big customer can vanish or be cut off on a geopolitical turn.

Bifurcation, and Open Source as Escape Hatch

Put together: the world is splitting from “one global tech stack” into “multiple camp-divided stacks.” On this map, open source plays an underrated role — it’s the escape hatch through the wall. When the most advanced closed compute is walled out by controls, the controlled party’s reliance on open source rises systematically (open weights are harder for traditional export controls to fully stop than controlled hardware — though whether weights and model-serving get pulled into controls remains an open policy variable). This adds a geopolitical motive to the “open source depresses model-layer profit” thread: open source is not only a commercial deflationary force but a geopolitical escape force — the controlled side has a strong, politically-driven motive to push and adopt it, adding to model-layer commoditization a political push immune to the commercial cycle.

Three closing lines. One: don’t value upstream on a “single global market” — model supply and pricing by camp; the two sides of the control wall are two different businesses. Two: manage chokepoint exposure by survivability, not probability — for the names most dependent on a single geopolitical chokepoint, apply a tail discount, cap exposure, don’t bet odds. Three: price sovereign demand as a stabilizer of the upstream pool (downside floor) and hedge it as a geopolitical tail-risk source — it both lengthens duration and ties that duration to politics.

The upstream name you’re looking at — how much of its scarcity premium is politically reinforced, and how much is tied to a geopolitical cliff? Comments open.

Sources (framework argument, June 2026): export controls, chokepoints (geographic concentration of lithography/packaging/leading-edge), sovereign AI as structural industry judgments based on public policy and reporting; no judgment on any nation’s policy. Precedent mechanics in Chapters 5 & 6.

— From Chapter 13 of a book in progress, working title The Deflation Sandwich

#AI #Geopolitics #DeflationSandwich