My submission to the 2026 Bridgewater x Global Citizen global economics challenge. Fu Zhou ("Fred Z"), publisher of the Scoop.
Twelve forecasts in three tiers: five general macro forecasts on the AI x Modern Mercantilism collision, five specific forecasts in areas where I hold live positions or have published research, and two niche forecasts based on firsthand experience that I do not think the consensus is pricing at all. Every forecast is binary, has a named public resolution source or condition, and a defined threshold. The probabilities are where the evidence puts them, not where they would look boldest.
Macro forecasts on the collision of AI capex and modern mercantilism.
At least one of the five major maritime chokepoints (Hormuz, Malacca, Bab el-Mandeb/Suez, Taiwan Strait, Panama) sees a sustained drop of 50% or more in transit calls lasting 30+ consecutive days between 2027 and the end of 2036.
Two disruptions that would qualify occurred in the last three years alone. In a mercantilist world, chokepoints are tools of policy, not accidents.
US federal net interest payments exceed 25% of federal revenues in any fiscal year before FY2033, on a full-year basis.
FY2025 closed at 19%, and CBO's baseline has interest doubling from $1.0T (FY2026) to $2.1T (FY2036). The baseline path gets near 25% late in the window. A term-premium shock gets there early.
The NY Fed ACM 10-year term premium averages above 150 basis points across a full calendar year before the end of 2032.
Duration risk gets repriced. Treasury supply keeps rising on the CBO path while the buyers who historically did not care about price, foreign central banks, keep rotating toward gold.
Combined capex of Microsoft, Alphabet, Amazon, and Meta falls year-over-year in at least one calendar year before the end of 2031.
2026 guidance is roughly $725B, up 77% from $410B in 2025, the third straight year above 60% growth, with capex running at 46-54% of sales for three of the four. No capex boom in modern history kept compounding at that rate without a digestion year, and grid interconnection queues are a physical limit that guidance cannot spend through.
The US average effective tariff rate is still above 10% at the end of 2030.
Tariff revenue is now a real line item the budget leans on, and one of the named drivers of FY2026 revenue growth. That turns tariffs from a negotiating stance into a fiscal dependency, and neither party will unwind a revenue line the budget already depends on.
Struck down and replaced the same week. The instrument changes; the policy survives. That is the forecast.
Q1 FY2026 ran at 22.1% of quarterly revenues.
Each of these extends a live book or published piece of research on the Scoop. See the research desk for the underlying write-ups.
The 7-day moving average of daily transit calls through the Strait of Hormuz does not exceed 60 before January 1, 2027.
I hold this with real money: NO on normalization behind a resolved six-leg ladder and two open legs. Normalization needs swept mines, war-risk insurance premiums coming back down, a finalized Iranian toll framework, and fleets moved back off Cape of Good Hope schedules. A signed MOU with unresolved terms delivers none of that.
China puts new export restrictions or licensing on at least two more critical minerals or processed materials, beyond what is restricted as of mid-2026, before the end of 2029.
Gallium and germanium in 2023, graphite and antimony in 2024, rare earths in 2025. The playbook is a ratchet, not a phase. Restricting exports is cheap for the restrictor and existential for the restricted.
A Chinese-developed model holds the #1 overall spot on LMArena, or whatever replaces it as the consensus public leaderboard, for 30+ consecutive days before the end of 2030.
Export controls limit compute. They do not limit talent, electricity, or state prioritization. Barely above even is where I honestly think the number sits, and it is well above where most US-based forecasters would put it.
By the end of 2030, US law or regulation formally splits event contracts in two: sports-outcome contracts under state gambling rules, while economic, financial, and weather contracts stay under exclusive federal CFTC jurisdiction.
You can already see the split forming in the lawsuits. States sue federally regulated exchanges over sports contracts; nobody sues over Fed-rates or CPI contracts. My forecast is that this hardens into formal structure, with the economically substantive products getting the crypto treatment: slow acceptance first, then formal federal integration.
Before the end of 2031, at least one S&P 500 company discloses in an SEC filing that it uses event contracts listed on a CFTC-regulated prediction market exchange as a commercial hedge.
An ice cream company hedging a cool summer, an airline hedging a shutdown-driven FAA disruption, a retailer hedging a tariff schedule. Adoption runs small-to-large. Volume from early adopters is what eventually makes the corporate hedge big enough to show up in a filing.
Two forecasts grounded in firsthand observation rather than published data.
By the end of 2030, China runs a systematic exit-restriction regime for AI researchers: either a formal exit-approval requirement, or credible documentation of routine passport surrender across leading AI labs and university departments.
During my senior-year study abroad at Tsinghua in Beijing, it was common knowledge among the students and researchers around me that senior AI people at top institutions had their passports held. The practice already happens informally. The forecast is that as the pay gap to US labs stretches into the millions, Beijing formalizes keeping people in.
Laws or binding regulation specifically restricting or conditioning datacenter development (moratoriums, mandatory ratepayer-protection statutes, or special large-load tariff regimes) get enacted in at least five US states before the end of 2029.
America's binding constraint on the AI buildout is not talent leaving, it is public consent. Rising home power bills, water use, and land fights are turning datacenters into the most contested local infrastructure of the decade, and state-level backlash moves faster than federal industrial policy can override it.
N1 and N2 are really one forecast applied to both countries: by 2030, the decisive constraint on both national AI programs is domestic and political. China has to hold its people in. The US has to hold its public's consent. Neither constraint is technological, neither shows up in capex guidance, and neither is priced.
Portfolio note. These 12 forecasts are deliberately correlated through one framework instead of scattered for diversification. If the framework is wrong, they should fail together. That is a feature. A forecaster whose predictions cannot fail together has not made a forecast; they have made a hedge.
A physical-layer framework for the collision of AI and modern mercantilism
Modern Mercantilism and AI look like two different stories, one about trade policy and one about technology. I think they are the same story: a competition for physical capacity that markets keep pricing through narrative.
Mercantilism works through physical things that states can seize, close, tax, or license: straits, fabs, minerals, refineries, ports, and increasingly, people. AI, despite the software branding, is the most physically intensive buildout since electrification: chips, concrete, transformers, water, and above all megawatts. The 2026 capex guidance from the four largest US hyperscalers, roughly $725 billion and up 77% year-over-year, is not a software budget. It is an industrial mobilization.
When these two forces collide, every strategic input to AI becomes a mercantilist target (chips, gallium, rare earths, grid capacity, researchers), and every mercantilist tool becomes AI-relevant (export controls, tariffs, chokepoint leverage, exit bans). The next decade is a race between states to accumulate physical capacity, and markets are doing a bad job pricing it.
Markets systematically misprice physical-layer events because they trade the narrative layer.
1. The narrative-physical lag. In March 2026, with US strikes on Iran already underway, Kalshi priced a 7+ day closure of the Strait of Hormuz at 19-43 cents depending on the window. The physical signal, active military escalation in a chokepoint that carries 20% of global oil, was public. The market anchored on the historical base rate ("closures are rare") instead of live conditions. I bought YES across three windows; all three resolved YES. Then the same lag ran in reverse on normalization. Every diplomatic headline moved the market several points in an afternoon, while the physical layer barely moved: war-risk premiums that had not compressed, vessels still on Cape of Good Hope schedules, an unresolved Iranian toll framework, unswept mines. When prediction markets and insurance markets disagree on geopolitical risk, I side with the people who lose money when they are wrong. Six legs resolved, six correct.
2. The hope premium. If you extract implied per-month normalization probabilities from the Hormuz contract curve, the same way you would strip forward rates out of a yield curve, near-term windows have been consistently pricing 18-22 percentage points of normalization probability per month while the 3-6 month windows price 4-5pp per month. That is a 4-5x gap, stable across three months of snapshots, that repeated NO resolutions failed to close. Whether that is rational front-loading of diplomatic odds or retail headline-chasing is genuinely open, but the tradeable part was not ambiguous: optimism about resolution is a risk premium, and someone gets paid to take the other side. I took it.
3. Rules are the territory. Kalshi's World Cup ad markets priced Gatorade YES above Pepsi YES three separate times in ten days. That inversion is impossible under the contract rules, because PepsiCo owns Gatorade, so every Gatorade ad is by definition a Pepsi ad: P(Pepsi) is always at least P(Gatorade). The market was trading vibes about brands; the contract resolves on a rules PDF. At decade scale, the same category of error is everywhere. "A deal was signed" is not "traffic returned." "Capacity was announced" is not "capacity was built." My forecasts are written to resolve on the water, not on the handshake.
I spent my junior year at Yonsei in Seoul and my senior year at Tsinghua in Beijing, two of the campuses closest to the semiconductor and AI frontlines of this collision. The most underpriced fact I brought home from Tsinghua is this: among the students and researchers around me, it was common knowledge that senior AI researchers at top institutions had their passports held. Not a rumor. Assumed, the way you assume a badge is required to get into a lab.
That observation is where the forecast pair N1/N2 comes from. Mercantilism started with goods, extended to capital, and is now reaching people. China's frontier AI talent is its scarcest strategic input, and the pay gap to US labs is measured in multiples, not percentages. What I saw on the ground, before it shows up in any policy document, is that Beijing's answer is retention by control.
The United States has the mirror-image problem. Its talent flows in, but its buildout runs on public consent: land, water, and above all electricity that shows up on household bills. Seoul treats chip fabs as national champions and grid buildout as close to a patriotic project. The contrast with US datacenter fights is stark. State-level backlash is the American version of the exit ban: a political limit on physical capacity that no capex guidance line accounts for.
Winners are whoever holds physical capacity that narrative cannot substitute for.
States that control electrons. AI demand makes the grid the binding constraint of the decade. China's electricity and manufacturing advantage faces America's capital advantage: $725B a year of private capex is a mobilization no other system can finance. The decade's central contest is electrons versus capital, with each side's domestic political constraint as the swing variable.
Whoever holds chokepoints and processing monopolies. Rare earth separation, gallium, advanced packaging, straits. The ratchet only tightens, and replacing a processing monopoly takes a decade of permitting and chemistry, not an announcement.
Neutral reserve assets. The 2022 reserve freezes taught every non-aligned central bank that reserve assets carry counterparty risk. Roughly 1,000 tonnes a year of official gold buying is the physical response, happening underneath a mostly stable dollar share of reserves.
The venues that price event risk itself. When Bretton Woods ended and currencies floated, it created the FX derivatives market. When rates were deregulated, it created the swaps market. Modern mercantilism is doing the same thing to event risk. Tariff schedules, chokepoint closures, and export-control decisions are now first-order P&L drivers for ordinary companies, and the instruments that price them are prediction markets. The markets I trade for edge today are the hedging infrastructure of the mercantilist decade.
Losers are balance sheets that are long the narrative layer. Long-duration sovereign creditors: the US fiscal path, with interest doubling to $2.1T by FY2036 on CBO's own baseline, runs into a shrinking pool of price-insensitive foreign official buyers. That is not a default forecast; it is a term-premium forecast. The bond market is pricing "fiscal consolidation will arrive" the same way Kalshi priced "traffic will normalize": on hope, ahead of evidence. Chokepoint-dependent importers without alternatives are next. And anyone trading announcements: the most repeatable losing position of the decade will be buying the diplomatic headline, the capacity announcement, the signed MOU, while ignoring the insurance premium, the interconnection queue, the held passport.
A fair objection: will AI itself, including systems like Bridgewater's AIA, arbitrage these narrative-physical gaps away? Eventually, at the margin, in liquid markets. But the decade's defining events happen where the gaps are widest and the correcting capital is structurally absent: novel contracts, geopolitical one-offs, thin books where institutional size cannot enter without becoming the price. My own analysis of the Hormuz curve found $8.2M of resting depth across five mid-curve contracts against $2.6M on the long-dated leg. Those are books where a fund cannot collect the premium but a disciplined small book can. AI makes the priced layer more sophisticated while mercantilism keeps generating unprecedented physical events that no training distribution covers. That gap is not closing this decade. It is the trade.
For the next ten years, read the physical layer (megawatts, minerals, transits, insurance premiums, held passports, resolution rules) and fade every market that is trading the press release.
Everything above was traded before it was written. All positions on Kalshi; entry prices are quantity-weighted averages across fills; all figures published as they happened on the Scoop with downloadable price-history data.
| Contract | Market | Dir. | Avg entry | P&L | ROI |
|---|---|---|---|---|---|
| Iran closes Hormuz 7+ days, before May | Closure | YES | 19.4¢ | +$2,419 | +387.5% |
| Iran closes Hormuz 7+ days, before August | Closure | YES | 42.0¢ | +$386 | +134.0% |
| Iran closes Hormuz 7+ days, before 2027 | Closure | YES | 43.2¢ | +$2,002 | +122.5% |
| No normalization before May 1 | Normalization | NO | 43.0¢ | +$228 | +132.6% |
| No normalization before May 15 | Normalization | NO | 30.0¢ | +$94 | +98.4% |
| No normalization before Jun 1 | Normalization | NO | 71.0¢ | +$406 | +40.8% |
Total realized P&L: +$5,535. The closure market alone returned +$4,808 across three legs.
| Contract | Dir. | Avg entry | Recent | Cost | Mkt value | Return |
|---|---|---|---|---|---|---|
| No normalization before Jul 1 | NO | 50.62¢ | 95.4¢ | $3,538 | $6,669 | +$3,130 (+89%) |
| No normalization before Aug 1 | NO | 52.92¢ | ~53¢ | $3,180 | $3,186 | ~flat |
The standout entry, May closure YES at 19.4¢ after US strikes had already begun, is the clearest example of the narrative-physical lag: the market was pricing under 1-in-5 odds of a 7+ day closure while the escalation was already observable. Beyond Hormuz, three other high-conviction positions this year (government shutdown duration, DHS reopening timeline, and a definitional arbitrage between two structurally identical contracts) have resolved correctly or are trending that way.
Forward-rate extraction from binary contract curves. Because sequential "before date X" contracts are cumulative, the implied probability of the event happening in a specific window is the difference between consecutive YES prices, and dividing by the window length gives an implied per-month rate. Same idea as stripping forward rates out of a fixed income yield curve. Applied to the Hormuz curve across six snapshots (Apr 1 to Jun 30):
| Window | Apr 1 | Apr 15 | May 1 | Jun 1 | Jun 30 |
|---|---|---|---|---|---|
| 0-1 month | +18.1pp | +21.8pp | +19.2pp | +21.7pp | +22-37pp |
| 1-2 months | n/a | +3.3pp | +8.7pp | +17.6pp | +14.1pp |
| 2-3 months | n/a | +7.1pp | +9.4pp | +8.4pp | +7.8pp |
| 3-6 months | +3.7pp | +4.3pp | +4.1pp | +4.5pp | +4-6pp |
| 6-9 months | n/a | +1.7pp | +2.9pp | +4.1pp | +2.6pp |
Term-structure execution. Backtested against actual Kalshi price history, rolling short-dated NO contracts (entering at 1-month-out median prices) returned +134.8% from April 1 versus +118.3% for buying and holding the July 1 leg. That is a gap of roughly 16 percentage points, with five reassessment points along the way instead of one, and no early drawdown. I will flag honestly that this gap may be a liquidity risk premium rather than free alpha. If that is true, it is also why it persists: institutional size cannot collect it without becoming the price. Execution is maker-only limit orders into headline-driven volatility spikes (near-dated contracts have daily standard deviations of 14-20¢ versus 3.5-6¢ at the long end), because 1-3% taker fees would eat a meaningful share of any leg's edge.
Rules-first resolution analysis. Before sizing anything, I read the resolution PDF as the territory. Kalshi's World Cup ad rules make every Gatorade ad count as a Pepsi ad, yet the market inverted P(Gatorade) above P(Pepsi) on three separate days (+9.8¢, +9.1¢, +7.7¢). That is a violated inequality, not an opinion. I published it as a guaranteed-positive structure that was still uneconomical after roughly 5¢ per leg of roundtrip friction. Knowing exactly why an arb is real but untradeable is the same skill as knowing why an MOU is real but insufficient.
The base rate is brutal: two 50%+/30-day disruptions in 2023-2026 alone (Bab el-Mandeb, Hormuz). A decade at even a conservative 15% per year hazard implies around 80% cumulative, and the mercantilist thesis says the hazard is rising, not mean-reverting.
FY2025 net interest was $970B, 19% of revenue; Q1 FY2026 ran at 22.1% of quarterly revenues; CBO's February 2026 baseline doubles interest to $2.1T by FY2036 with debt at 120% of GDP. The baseline only gets near 25% late in my FY2033 window, which is why this is 50% and not higher.
The milder version of G2's shock: steadily rising duration supply meeting a buyer base whose biggest official participants are rotating into gold, roughly 1,000 tonnes a year since 2022.
Big-four capex went from $410B (2025) to about $725B guided (2026), a third straight year above 60% growth; capex-to-sales at 46-54% for three of the four. 65% prices one down-year in five, not a bust. Interconnection queues and transformer lead times force a digestion year even if demand holds up.
Tariff revenue is a named driver of FY2026 federal revenue growth, and the strongest form of policy stickiness is a revenue line the budget already depends on. Both parties are mercantilist now; the debate is about targeting, not the level.
Extends my live book. Normalization needs mine clearance, war-risk premiums compressing, a finalized toll framework, and fleets moved back off Cape schedules. The tail I acknowledge: a burst of pent-up transits after clearance spiking the 7-day average through 60. I hold NO anyway. The market's roughly 52% pricing conflates 'deal signed' with 'traffic restored.'
Four straight years of new restrictions is a trend with asymmetric costs: cheap for Beijing, existential for manufacturers that import the inputs. The AI collision adds targets, because every element in an accelerator supply chain is now leverage.
The compute gap from export controls is real. So are the offsetting advantages: electricity, talent density, open-weights momentum, and state prioritization. Barely above even is the honest number, and it is materially above the US consensus.
The de facto split already exists in the lawsuits. Crypto is the template: years of jurisdictional ambiguity resolving into formal federal integration for the economically substantive core. It is 60% rather than higher because 'formal' is a demanding threshold; the split could sit in unresolved case law past 2030.
Mercantilist policy makes event risk a P&L line item, small compliance-light shops adopt first, their volume deepens the books, and disclosure follows use. Weather and commodity derivatives took roughly a decade to travel that path. Event contracts started that clock around 2024.
Grounded in direct observation at Tsinghua: passport-holding for senior AI researchers already happens informally. Resolution here is reporting-based, which is softer than I would accept in a Kalshi contract, and I flag that honestly.
Residential electricity bills are the price American voters see most directly, and datacenter load growth is the most attributable new driver of them. Five states in three and a half years only requires the current trajectory to continue through one more election cycle of rising bills.
I am stating these before anyone else does, because a forecast whose author cannot name its weaknesses is a narrative, not a forecast.
Horizon mismatch, owned upfront. My verified record is weeks-to-months resolution. This challenge is a decade. I am not claiming short-horizon wins prove 2036 foresight. I am claiming they prove the process: operationalized thresholds, named data sources, physical-layer evidence over narrative, sizing to confidence, and publishing the reasoning before resolution.
Small sample. Six resolved legs and three broader directional calls is not statistical proof of skill. It is six pre-registered, publicly documented, real-money tests of the exact mechanism this framework depends on, with zero failures so far.
Correlated book. All twelve forecasts share one engine. If markets are actually better at pricing the physical layer at decade horizons than at monthly ones, the framework degrades and the forecasts miss together. I accept that concentration knowingly. It is what makes the set falsifiable.
Anecdote is not a dataset. N1 rests partly on firsthand observation from one campus in one year. I have weighted it as strong qualitative evidence about what the state prefers, not proof of national policy. The 70%, instead of 85%+, reflects exactly that discount.
Named tripwires. I would update fast and publicly on: sustained reserve-share and gold stabilization for 8+ quarters (weakens G3), hyperscaler ROI disclosures that justify the capex curve (weakens G4), non-China mineral processing coming online at scale (weakens S2), a federal statute formally embracing sports contracts (kills S4), or documented loosening of Chinese researcher travel (kills N1). One market is not a theorem, and the tripwires make my future updates checkable instead of retroactive.
All underlying price data is from Kalshi's public market feeds (KXHORMUZNORM-26MAR17, KXWCADS-26JUL19), published with downloadable CSVs at scoop.market. Fiscal anchors: US Treasury Monthly Statements and CBO's February 2026 Budget and Economic Outlook. Reserve anchors: IMF COFER (2026Q1). Capex anchors: company guidance from Q4 2025 / Q1 2026 earnings calls.
Positions disclosed where held. All positions on Kalshi. Not financial advice.