Source: What If China Wins The AI Race? (Andrei Jikh) | 2026-11-13 | the corpus’s 99th source
The video argues three things. The AI fight is closed-source pricing power against open-weight abundance. China gains by giving models away. And, through Luke Gromen, either AI outcome ends in money printing. This version replaces the earlier draft. It reflects that Jensen Huang has since resigned.
Verdict on the conclusion
It partly stands. The direction holds, the strong form does not, and the hard-money thesis holds as a barbell with conditions.
| Claim | Verdict |
|---|---|
| Medium-term money creation | Stands, mainly because of fiscal dominance already in the corpus ($40T debt, about $1.1T interest, about $1.8T deficit, Fed reserve-management purchases). AI accelerates it. |
| “Either AI outcome forces printing” | Too strong. Two plausible exits exist (below). |
| Tax base at risk if AI works | Stands, and the video understates it. |
| Gold for the deflationary leg, Bitcoin for the inflationary leg | Stands as logic. Bitcoin behaves as a risk asset in the first leg. |
| Energy and metals | Stands under either outcome. |
Stress test of Gromen
Tax base. The video says about half of tax revenue comes from jobs. In FY2024, individual income taxes were about 49% of $4.9T in receipts and payroll taxes about 35%, so wage-linked revenue is up to about 84%. Corporate taxes were about 11%. Only the displaced share matters, and individual income tax includes non-wage income.
Exit 1, “AI works”: income shifts from wages to capital, and a tax system can follow. Trump floated government stakes in AI firms (Source 91). Nominal growth above the interest cost also shrinks debt relative to GDP. Neither is automatic, and both need legislation, but neither requires printing.
Exit 2, “AI fails”: the bust may be absorbed. Hyperscalers largely fund capex from cash flow, and stress concentrates in neoclouds and private credit (Gartner: over $3.5T of exposure and most neoclouds gone by 2030, Source 93). A bailout depends on whether losses reach banks and insurers.
Near-term evidence cuts against “printing now.” The Fed hiked in September to 3.75–4.00%, and Treasury yields fell on Nov 5 as money fled to bonds. The balance sheet still shows QE-lite, so policy is mixed.
What AI adds. It removes exits. Losses and gains concentrate in a few names, and the timing is uncertain.
Two supporting checks. The “no new US generation” claim is roughly right on output. EIA says US electricity consumption was essentially flat for nearly two decades, and production grew only about 0.5% a year from 2003 to 2023. Capacity was added and retired in that period. The conclusion that the grid is under-built is consistent with Sources 93 and 98. I could not verify the gold-versus-Nasdaq figure, which depends on endpoints, so I’m leaving it out.
Hard outside money: what holds
Bitcoin fell 5.7% to about $75.9K on Nov 5 with the equity selloff. That fits the barbell: in a deflationary shock Bitcoin sells first and gold holds up better. So the thesis rests on sizing, horizon and self-custody, not on timing. Energy, power equipment and metals are the least path-dependent leg. Positions in productive inner-ring equity capture the same transfer as inside money.
Fact-check
| Item | Status |
|---|---|
| Nvidia board probe, then Huang’s resignation | Confirmed. The board accepted his resignation as President, CEO and director on Nov 11, announced Nov 12. CFO Colette Kress is interim CEO while keeping the CFO role. A successor search is open. Nvidia’s Q3 FY27 results are due Nov 17. |
| Reason for the resignation | No reason given in the announcement, and no findings announced. FBI, congressional and shareholder actions are reported. |
| Kimi K3 (about 2.8T parameters, weights released late July) triggered the open-model fight | Confirmed |
| July 24 letter by 25 companies; the biggest closed labs were absent | Confirmed |
| “The administration backed off” | Overstated. Reporting says case-by-case, no decision. |
| “Chinese models are 90–95% as good at 1/100th the price” | Unverified. A preliminary US-UK assessment found K3 still trailed leading US models. |
| The leak was timed to remove an open-model champion | Unproven. An intelligence memo had circulated in the spring, and the resignation gives no reason. |
| Democrats are preparing bills | Confirmed as proposals: independent pre-release evaluation and liability. Trump opposes mandatory rules, so a veto is possible, and riders on must-pass bills are the likely route. |
| Nvidia’s reported $26B open-model plan | Per the video only |
Chapter 6: the Cantillon effect
The open-weight price collapse sorts players into toll collectors and lenders. Toll collectors sell chips, power, land and compute-hours, and they are paid up front whatever the model layer earns. Nvidia is the clearest case. Lenders hold the residual: bondholders, private credit and insurers. That is the preprint’s mechanism, with the inner ring spending at yesterday’s prices and the cost landing downstream. Falling prices for AI-linked services keep official inflation calm while the transfer shows up in assets and inputs.
Huang’s exit adds a lesson. The chokepoint everyone treated as untouchable had a governance gap of its own. The preprint’s attestation logic applies to human principals as well as agents.
Will the post-midterm climate hurt SPCX?
SpaceX is conflicted. As a compute landlord it benefits from abundant, cheap models. As a model vendor (Grok, Cursor, Grokbot) it wants pricing power, which open weights erode. That is the weaker half of the $500B ARR target. Federal bills face a veto through January 2029. A bipartisan liability bill (the AI LEAD Act) and frontier-model audit proposals do exist, and more than half of voters called AI and data centers a major factor in their vote. Restrictions on Chinese models would be mostly neutral for compute-hours.
Impact on orbital pricing (US$30B/GW)
Model prices fell sharply while compute rents rose (Source 96), so the Jevons effect has held. The risk is efficiency gains outrunning demand growth, which is Gartner’s impairment case. My earlier view stands: the $30B/GW figure is ground-scarcity rent through 2028, and Musk’s own orbital illustration is about $10B/GW at scale.
Impact on data linkages
Open weights are portable. A 2.8T-parameter model is roughly 1.4–2.8 TB, so at 10–100 Gbps a one-time load takes minutes to about half an hour (my arithmetic). Ongoing traffic is small prompts and outputs. That suits orbital, edge and sovereign inference. Closed models require shipping proprietary weights to the lessor.
Will this fuel insatiable demand?
In compute-hours, yes. Moonshot temporarily limited signups when demand exceeded its inference capacity. In revenue per token, no. Demand shifts toward owners of the scarce physical layer.
Does SPCX retain its moats?
| Item | Direction |
|---|---|
| Landlord position (wins under open or closed) | Strengthened |
| Counterparty risk | Higher. Anthropic, the largest anchor at about $37B a year, is the lab most exposed to open-model price pressure. |
| Nvidia supply and leadership | Now realized. SpaceX installs about 200,000 GB300s a month and co-designs Starmind with Nvidia. Allocation sits under interim leadership. Near-term continuity is likely, since commercial and engineering teams and CUDA remain. |
| Texas footprint | Unchanged from Source 98 |
| Orbital and launch | Pending Flight 15 |
The agentic ecosystem
Governance has to attach to agent identity, not model access. Open weights run anywhere, so one lab’s refusal policy or a pre-release audit of US labs does not bind the agents that matter. One outlet reports that Hugging Face defended itself with a Chinese open-weight model after American closed models refused to help. That is a single account, but it shows refusal-based safety failing defenders. An open, model-agnostic Values Passport is the layer that survives both a closed-lab regulatory moat and an open-weight world.
EWS
| Signal | Status |
|---|---|
| Nvidia leadership (interim CEO, successor search) | 🟡 Event occurred |
| Nvidia earnings, Nov 17 | 🟡 Watch Rubin supply and 2027 allocation |
| Nvidia open-weight stance under new leadership | 🟡 New |
| Investigation outcome | 🟡 Open, no findings |
| Administration decision on Chinese open models | 🟡 Case-by-case |
| Federal AI bills and riders (Jan 2027) | 🟡 Veto-limited |
| Anthropic payment capacity | 🟡 Watch |
| Flight 15 (about 06:45 SGT, Nov 14, if on schedule) | 🟡 Binary event |
| Gromen falsifiers | |
| Fed turns to cuts or QE while Treasury yields rise | Supports the thesis |
| Federal individual income and payroll receipts fall year on year | Supports the thesis |
| Neocloud or private-credit failures reach banks or insurers | Supports the thesis |
| Capital-tax or equity-stake legislation passes | Weakens the thesis |
| Nominal growth outpaces interest costs | Weakens the thesis |
Synthesis
Gromen’s conclusion stands as a medium-term direction driven mainly by fiscal dominance, which AI accelerates but does not create. The claim that both AI outcomes force printing is stronger than the evidence, and the hard-money thesis is best held as a sized, self-custodied barbell. The instruction is unchanged: the identity and values layer must be built before the infrastructure it governs becomes default.
Sources:
- Nvidia 8-K, CEO transition
- Nvidia 8-K, board action
- Variety on the resignation
- Nvidia board probe (Implicator)
- Open-model letter (Dealroom)
- Kimi K3 weights release (Caixin)
- Kimi K3 benchmarks and policy talks (Implicator)
- Democrats’ AI bills (CDO Magazine)
- Federal revenue FY2024 (USAFacts)
- EIA on flat electricity consumption
- EPRINC US power demand growth