---
title: "Part III: Principalities and Powers"
part_of: "The Commanding Heights"
part_of_url: https://commanding-heights.com/commanding-heights.md
authors: ["Will Manidis", "Joe Roberts"]
date: 2026-10-29
url: https://commanding-heights.com/#part-iii
summary: "America and China will become the only countries capable of sovereign expression because only they can continually develop frontier models. Intelligence will splinter into a commercial and sovereign frontier, with the latter being locked behind SCIFs and developed by lab subsidiaries for statecraft, not cash flow. China can publish cheap open-weight models and accept low valuations for its labs because it is not subject to the American 401(k) class. These sovereigns will exercise control over model-producing firms without formally announcing nationalization."
---

## Part III: Principalities and Powers

> *America and China will become the only countries capable of sovereign expression because only they can continually develop frontier models. Intelligence will splinter into a commercial and sovereign frontier, with the latter being locked behind SCIFs and developed by lab subsidiaries for statecraft, not cash flow. China can publish cheap open-weight models and accept low valuations for its labs because it is not subject to the American 401(k) class. These sovereigns will exercise control over model-producing firms without formally announcing nationalization.*

They are building a separate future in China, in case you couldn’t tell. It will be a form of modernity with little resemblance to our own. The Genesis period of machine intelligence resembles many other episodes of industrial history within China. There is nothing new under the sun for China. Picture a furious competition over talent and resources, causing a cascading price compression to the bare minimum necessary to operate, and then perhaps something less than that. All against all, within the market, under the party-state. The same dynamic is playing out in AI. Chinese machine intelligence is now competitive in capabilities for the vast majority of consumer uses. Chinese labs such as Alibaba and DeepSeek will continue to supply open-weight, low-cost commodity models across the entire world. In short, they will cross the American intelligence moat. If the American buildout is premised on a sufficiently large value captured by labs, the price compression of Chinese labs globalizing will upend this arrangement to finance the perpetual frontier. The immediate consequence of price compression is an inability to finance scaling as had been anticipated. What good is a frontier model if only a portion of the expected compute ends up being built?

We distinguish commodity intelligence, a commercial frontier sold at a premium, and a sovereign frontier held in state reserve. One lab may operate in just one vertical, or across all three. Today, China sets the commodity floor and America leads the commercial frontier, though the capability delta between the two is shrinking and, increasingly, so is the willingness of consumers to pay the difference in token prices. This new paradigm had been visible for more than a year beforehand, though in the eyes of many in English-speaking countries, it struck like lightning in the summer of 2026. On July 16, 2026, Moonshot AI released Kimi K3, a 2.8T-parameter model whose weights could be downloaded from anywhere, costing a trivial amount, and the capabilities of which rivaled the American commercial frontier for most tasks. The next day, in a speech to the World AI Conference in Shanghai, Chinese leader Xi Jinping declared that Chinese machine intelligence would remain open-weight and widely available.[^58]

[^58]: Moonshot AI, “Kimi-K3,” model card, Hugging Face, accessed September 29, 2026. Laurie Chen, Casey Hall, and Eduardo Baptista, “Xi Pitches China as Leader of New Global AI Order, Challenging US Dominance,” *Reuters*, July 17, 2026.

This sparked an immediate American response. Michael Kratsios, director of the OSTP, accused Moonshot of building Kimi K3 through industrial-scale distillation of Fable. He separately alleged that Moonshot had trained the model on NVIDIA’s most high-capability chips, remotely accessed in Thailand. More recently, the Cybersecurity and Infrastructure Security Agency suggested much the same. The Treasury Secretary, Scott Bessent, threatened financial sanctions and placement on the Commerce Department’s sanctions list.[^59] American officials seemed to recognize the scale of the threat that open and capable Chinese intelligence represented. So did the labs and their manifold horsemen. Augurs scrabbling in the forum, discourse on the everything app. Ban the models! It’s all distillation, China can’t really compete at the frontier! Create regulatory uncertainty! Open-weight models are indispensable to American startups! Your data, in a Chinese model? Unthinkable. They might still be conducting discourse on the everything app yet. But real regulatory action never came. All that remains of the Kimi K3 moment are pronouncements on dot gov sites. It is hard to find a regulated workflow that doesn’t use Kimi in production today, and the American public, and even the intelligence class, has totally forgotten about this dustup.

[^59]: Doina Chiacu, Chris Sanders, and Courtney Rozen, “US Accuses China’s Moonshot of Stealing from Anthropic’s Fable for Latest AI Model,” *Reuters*, July 22, 2026. CISA, NSA, and FBI, Joint Cybersecurity Advisory AA26-251A, September 8, 2026. Michael Kratsios (@mkratsios47), post on X, July 22, 2026. Scott Bessent (@SecScottBessent), post on X, July 22, 2026.

We see clearly that in America, the state is not yet ready to behave like a sovereign. We use each word very deliberately. The latter is distinguished as the final backstop of authority, the buyer and seller of last resort. It is the latter to which the full faith and credit belongs. Atlas on the Potomac, or perhaps more a hesitant Samson, holding San Francisco in one hand and New York in the other. And ultimately, as the financing of American machine intelligence is placed under unbearable strain, labs and shareholders will cry out and the American sovereign will answer their invitation.

### The Artifact

It is the artifact that the sovereign will act to protect. **Here, we define the artifact as a formula of intelligence that can be reproduced**. Any widely available model will be distilled into an artifact and commodified. This begins the process of price compression. In the past, the state acted to control the export of hardware in the hopes that hobbling compute at a particular moment would grant America a decisive edge in machine intelligence. But rusted and worn down, the initial export-control apparatus of the early machine intelligent era has fallen apart. The hardware has escaped containment and slipped loose from the low-lying fields of the Netherlands. The Chinese are a modern, civilized people. Even if all you still think they can do is copy, why would EUV machines that strike molten tin with regularized laser pulses through mirrored refraction to print circuits with 13.5-nanometer light be impossible?[^60],[^61] In lapidary form: there is no technology beyond the reach of Chinese hands. The Chinese Intelligence Apparatus is far more complex than thousands of Blackwell chips wrapped in tinfoil in carry-on luggage.[^62]

[^60]: U.S. Department of Justice, “Three Charged with Conspiring to Unlawfully Divert Cutting Edge U.S. Artificial Intelligence Technology to China,” March 19, 2026. ASML, “Light & Lasers” and “Lenses & Mirrors,” accessed September 29, 2026.

[^61]: In October 2026, a customer at Micro Center’s Tustin, CA, store who sought to purchase a workstation including an NVIDIA 5090 reported being required to fill out and sign a Claude-styled document entitled “Advanced Computing Product Purchase Declaration.” It read “NO EXPORT - NO EXCEPTIONS” and required an address and identification.

[^62]: Might there be a corresponding *Away* luggage trade?

In response to the commodification of the artifact and the resulting compression of value, the American state will pursue a new paradigm. If the first export control system was built for objects, the second control system will be aimed at the constituent parts of the artifact. The subsequent paradigm will be more adaptive, though much less formal. We expect the relationship between lab and sovereign to become far more intimate and bidirectional. Export-control decree by whispered insinuation, not via public policy and certainly not via think tank.

The United States is moving toward an increasingly fortified protection of sovereign frontier intelligence, including stronger protection of model parameters. Additionally, the release of models from labs has been slowed down by not-so-voluntary pre-release review and exceptional state controls on model access.[^63] Even so, internal capabilities have continued to grow apace. There is a cleaving, right now, between the abundant, commercially available intelligence and the artifact held in a bunker in San Francisco or a guarded tower in New York. In an act of self-preservation, the sovereign frontier will need to be locked inside, beyond the access of both common Americans and Chinese. The commercial frontier and commodity models will remain outside, feeding the gaping maw, 8B mouths stretched wide to consume them.

[^63]: White House, “Promoting Advanced Artificial Intelligence Innovation and Security,” Executive Order 14409, June 2, 2026, sec. 3. Anthropic, “Statement on the US Government Directive to Suspend Access to Fable 5 and Mythos 5,” June 12, 2026. OpenAI, “Previewing GPT-5.6 Sol: A Next-Generation Model,” June 26, 2026. Anthropic, “Expanding the Cyber Verification Program,” October 6, 2026.

The reason is this: a frontier artifact is an extraordinary capital project condensed into a reproducible file that can be carried in a palm.[^64] The scale of capital is unparalleled, premised on the tremendous gains and secured value that American frontier intelligence promised. But the commodification of the artifacts converts a capital-intensive and scarce good into a widely accessible input. This, well before the investor recoups his cost, well before the solidification of the expected value. If a frontier model is widely accessible, it will be distilled. If it is distilled, it will be commodified and its value compressed to the barest margin. Subsistence profits, familiar in the rice paddy. The continued development of a protected sovereign frontier will be necessary to serve two converging interests: the value proposition of the line and the state’s totalizing interest in surety of the largest financial outlay in American history.

[^64]: In “Great Work of Time” by John Crowley (1989), the protagonist, a brilliant but indolent inventor of a form of time travel, decides the safest way to get rich without perturbing history is to go back to 1856 British Guiana, buy a one-cent British Guiana Magenta, the most valuable stamp in existence, of which only one copy survives at present, mail himself a letter with it, wait for it to arrive cancelled, and hide it somewhere he knows will survive to the present. One wonders if today a time traveler going in the opposite direction, even a matter of months, would simply pocket an SD card of frontier weights.

### From the Other Side of the Pacific

The air is different in China. It’s clear and sweet in Inner Mongolia. In Chongqing, the oppressive heat can hardly escape the steep inclines of the city or the mountains that enclose it. In Beijing, the air is heavy with just cigarette smoke. No longer the smog that characterized it in the early 2000s. The Chinese have modernized past all that. In the hutongs, old men who remember hungry days will gather after a long day of loitering in public parks. They remember the ration coupons during the 70s. Now they clutch 512 GB Chinese-made phones. This is retirement. Every surface is either oily or sticky. They sip Red Star erguotou slowly, stooped over, necks craning over to watch the crickets fight. Payment by WeChat. Sinomodernity in the smallest moments. From Zhongnanhai to the upland villages of Liangshan, they labor beneath the same line that we do. Only, seen from China, the line looks different.

**China has no desire, nor any interest, to participate in the American intelligence paradigm of the past.** The Chinese party-state of today and the future will see little reason to protect enterprise value in a model layer. Not that the Chinese model-layer value is zero. Instead, it is the near-minimum. Firms, provincial governments, and the party-state itself do hold economic interests, sometimes in the labs themselves and their secondary performance. Certainly, China maintains an interest in the overall growth of its economy, which will depend more and more on AI-enabled efficiencies and the supporting financial commitments. But the level of capital needed to keep pace at the commercial frontier and break the American value proposition is meaningfully lower. Moonshot is valued at about $50B on the verge of its IPO and DeepSeek’s latest reported funding round values it at about $90B including new capital. Other major Chinese labs trade on public markets at fractions of American competitors. Compare this to the trillions in value attributed to OpenAI and Anthropic, not to mention the trillions in valuations of related firms and trillions more expected as AI diffuses throughout the entire world.[^65]

[^65]: Tracy Qu and Raffaele Huang, “China’s Moonshot AI Confidentially Filed Hong Kong IPO,” *Wall Street Journal*, September 3, 2026. Kane Wu, Yantoultra Ngui, and Eduardo Baptista, “DeepSeek Wrapping Up $15 Billion Fundraising to Fuel AI Race, Sources Say,” *Reuters*, October 22, 2026. Echo Wang, “Anthropic’s IPO Prospectus Shows Sweeping AI Vision, Surging Costs,” *Reuters*, September 28, 2026, updated September 29, 2026.

What explains the divergence between the Chinese and American technological arrangements? Why are our labs seen as world-eaters, while theirs are valued at fractions? It is tempting to appeal to the cultural differences. Or point toward four thousand years of imperial history, mass famine, mass upheaval. Ascribe it to the political repression, as though it is so radically different from ours, and try to explain the whole from one scrap of ephemera. Such lazy answers: we play chess; they play Go. China is fascinating, not least because it causes the most original minds to lapse into the most unimaginative and incurious formulations.

The answer is hidden in the explanatory power of the dollar and the renminbi. Why is the Chinese sovereign more willing to commoditize the commercial frontier than the American sovereign? It is down to financial markets and the mass populations, here and there, subservient.

Where is wealth stored in the average Chinese household? Since the end of the Cultural Revolution, it is no longer stuffed under mattresses or buried in some field outside the village until the Red Guards pass. Today, the way that Chinese wealth is stored is fully modern: banks, checking accounts, a system of checking, savings, and withdrawal that looks not unlike your own despite being a closed capital account.

But the composition of that household wealth is different, which goes a long way in explaining why the Chinese party-state is less compelled to prop up the stock market. The Chinese see themselves as a democracy, just of a different sort. A state ultimately answers to its people, whether or not it is formally democratic. *Panem et circenses*, both being superabundant. In ensuring prosperity, the Chinese state responds to a set of incentives quite different than ours. Their attitudes about the governance of machine intelligence, the level of state capital expenditure, the adaptive interference of regulators and censors: it all is downstream of whether such policies advance the primary imperative of the modern state: wealth preservation and creation.

![Dot chart of household assets as a share of the total, 2022. Housing: 47.4 percent in China, 27.5 in the US. Cash and deposits: 24.7 and 9.2. Equity, businesses and funds: 15.4 and 34.4.](https://commanding-heights.com/img/household-assets-china-us-chart-print.png)

*Household asset composition in China and the United States, 2022.*

Source: Authors’ calculations from World Bank, *China Economic Update: Advancing Reforms, Enhancing Prospects* (December 2025), 22, table 2. Board of Governors of the Federal Reserve System, “B.101.h: Balance Sheet of Households,” *Financial Accounts of the United States*, September 11, 2025, 2022 column.

In China, far more wealth is held in housing and cash than in America. This is why the 2021 real estate crisis was so consequential. When the central government tightened borrowing requirements in 2020, the business models of various property developers ceased to function. No more filling Chinese cities to the brim with identical concrete towers and inhabitants straight from Sichuan village towns. Recall one of the largest property developers in China, Evergrande, defaulting on more than $300B in liabilities. Construction firms collapsed, leaving rows of towers standing in their wake. But even after this turmoil, real estate and housing remain the most vital part of Chinese household wealth. Even in 2022, 47% of Chinese household wealth was still held in real estate. Cash and deposits, an enduring sign of the definite pessimism of the Chinese, are extraordinarily high at 24%. Equities are only 15% of Chinese household wealth.[^66] This quantitative indifference toward markets and the superfixation with housing is a central condition shaping the Chinese conception of the intelligence economy.

[^66]: Patrick Hendy, “Evolving Financial Stress in China’s Property Development Sector,” *Reserve Bank of Australia Bulletin*, September 15, 2022. World Bank, *China Economic Update: Advancing Reforms, Enhancing Prospects* (December 2025), 22, table 2. “The Human Cost of China’s Property Crisis,” *Financial Times*, November 28, 2023.

The party-state will not act to boost the share prices of MiniMax or Zhipu if it doesn’t affect the annuities of hundreds of millions of senior citizens. The labs themselves will do fine, returning a tidy, reasonable profit for themselves. And the rest of the economy, which is not quite so levered on the success of a few individual labs, will keep pace. There are new qualitative productive forces, those Commanding Heights seen by Chinese eyes, and they are not determined by the rise and fall of the line.[^67]

[^67]: Xinhua, “Xi Stresses Development of New Productive Forces, High-Quality Development,” February 2, 2024, National Committee of the Chinese People’s Political Consultative Conference.

The Shanghai Composite Index has been essentially flat, up only 29% over the last ten years. S&P is up 254% over the last ten years, and there are those beyond counting who would say that is not nearly enough.[^68] In America, the whole spectacular arrangement grinds to a halt if the line does not go up enough. If the Dow Jones was only up 29% over the last ten years, there would be riots. The Villages ablaze, golf carts smoldering in sand traps, target-date funds with targets on their backs.

[^68]: Authors’ calculations from Shanghai Stock Exchange, *SSE Statistical Yearbook 2017*, “SSE Composite Index—Daily Index,” and *Newsletter—August 2026*, “SSE Monthly Data.” Raymond James & Associates, “Domestic Stocks Hit All-Time Highs in August” (2016), 1. S&P Dow Jones Indices, “S&P 500 \[SP500\],” FRED. August 31, 2016–August 31, 2026, price returns in local currencies, excluding dividends: Shanghai Composite, 29.2%. S&P 500, 254.0%.

**This is because American household wealth is determined far more by defined-contribution retirement plans.** In 2022, retirement accounts accounted for about 32% of household wealth. Stocks and assets at financial institutions came in at 10% and 9%, respectively.[^69]

[^69]: Briana Sullivan and Shomik Ghosh, *Wealth of Households: 2022*, Current Population Reports P70BR-202 (U.S. Census Bureau, November 2024), 6–7, fig. 3. These figures exclude the wealthiest 1% of households.

In the past, American workers would work at the same company for decades, and at the end of their career they would receive a wristwatch and a defined-benefit pension plan. As these became fewer and fewer, the benefits from Social Security played a larger role in financing America’s growing population of the elderly. But Social Security cannot pay out the same benefits for much longer. As defined-benefit annuities receded, contribution-based retirement plans, with matching contributions from the dozen employers that an American office worker bounces between over the course of his career, became the indispensable part of American retirement.[^70] Put your money in the market, let it grow and grow, and at the end of forty-five years, you will have enough to retire. That was the nursery rhyme. Settle in Florida to molder in the humidity.

[^70]: John J. Topoleski and Elizabeth A. Myers, *A Visual Depiction of the Shift from Defined Benefit (DB) to Defined Contribution (DC) Pension Plans in the Private Sector*, IF12007 (Congressional Research Service, updated March 25, 2026), 1–2. Social Security Administration, “Social Security Board of Trustees: Projection for Combined Trust Funds Remains Consistent with Prior Year,” June 9, 2026.

But these retirement schemes ride the highs and lows with the stock market. Which, these days, is to say the highs and lows of the intelligence economy. Itself an indicator of the successes and failures of various training runs. Your retirement plan already relies on Claude, even if you can’t see it, or even know what it is.

Recognizing the centrality of the intelligence economy to the American stock market, it is entirely sensible that the needs of such a sovereign would differ substantially from the Chinese ecosystem, where labs are a much smaller part of the economy, easily substitutable, and the stock market itself inches upward at a sluggish pace. When half of the country’s household wealth is tied up in the stock market, as compared to fifteen percent in China, the particular needs of the sovereign become apparent.

American policymakers must defer to the needs of a national constituency whose livelihoods are determined, in large part, by fluctuations in corporate valuations. Chinese policymakers take heed of a much wider array of indicators, such as housing, bank deposits, and minimal annuities.

Valuations in China are increasingly determined by the market, but without the trillion-dollar exuberance of those in America. There is also active consent, it seems, from the Chinese labs to this paradigm. In a leaked four-hour investor call, DeepSeek CEO Liang Wenfeng candidly described his strategy for the firm’s commercialization and his plan for an enduring edge. Liang explicitly stated that DeepSeek’s models would remain open-weight for the foreseeable future. Even more shocking to American capital, DeepSeek would not aim to maximize profits against all other interests. He would instead guide the firm, in which Liang holds the majority of shares, toward what he termed a “reasonable profit.”[^71] What would the American landscape look like if the investors and financiers who underwrote it believed that AI would return just a “reasonable profit,” rather than extracting maximum shareholder value from as large a surface area as possible?

[^71]: Coco Feng, “Low Profile, High AI Ambition: What Leaked Comments Reveal about DeepSeek’s Liang Wenfeng,” *South China Morning Post*, July 26, 2026.

Perhaps one only needs to look towards the safetyist movement to see what American AI with Chinese Characteristics could look like. Although, increasingly, all sides of the debate are open to central planning at the cost of shareholder value: an oddly bipartisan suggestion that seems to reoccur is using the Defense Production Act (DPA) to dictate the allocation of compute between training and inference.

We have primarily focused on the domestic ecosystem that encourages the Chinese treatment of the artifact. But the international advantage should be obvious as well. With limited domestic downside, pursuing a commoditized form of intelligence is the primary way that Chinese labs and party orchestrators can dull the American edge in machine intelligence. A runaway value proposition stopped short by Chinese hands. Short of the intrigue of exfiltration and espionage, weights smuggled out in briefcases and across oceans in planes, this is something more simple yet far more totalizing: the banality of financialization, commodification, balance sheets, and the unbearable imperative of making a return.

The Chinese paradigm determines the tail end of the intelligence curve. The premium of an American commercial frontier model is thereafter legible as a spread over a visible benchmark. For every workload that a Chinese model can perform, a closed-weight provider must justify their asking spread. This premium will be under a permanent audit by the clients and continually compared against the best from the other side of the Pacific. Independent bodies have already sprung up to provide such market intelligence to firms. The pricing differential will become increasingly legible to market participants. Intelligence adolescent is on its way to New York, preparing ritualistic sacrifice by the opening bell. Markets index the value of commodified machine intelligence in China. The American paradigm is only now beginning to catch up.

![Scatter chart of net improvement against median cost per task in Arena’s Agent Mode, September 2026. US models top the frontier and Chinese models hold its cheap end, with a 10.6 times price premium between the cheapest US and Chinese models at or above 4 percent net improvement.](https://commanding-heights.com/img/arena-agent-mode-frontier-chart-print.png)

*The Pareto Frontier of Arena’s Agent Mode. Source: Arena.*

Source: Arena, “Agent Arena: Overall - Pareto Frontier,” chart snapshot as of September 30, 2026.

### The Shrinking Aperture

The Chinese offensive paradigm, today and into the future, will be built on forcing a retreat of America’s most capable models from the commercial frontier to the sovereign frontier, with the public shut out. The intelligence aperture snaps shut. Absent any action on the part of the state, price compression will continue, starving the frontier of funding.

Do not expect a lack of access to America’s sovereign frontier to blunt all Chinese momentum. It is at present unclear to what degree distillation is responsible for Chinese labs being able to keep up with the American commercial frontier, though it plays a substantial role. If we understand distillation as a necessary response to a shortage of chips, a reaction to insufficient scale, then we should expect that this deficit will only last a few more years. The Chinese do not even need to be capable of reliably producing the most advanced 5nm and 7nm semiconductors. They will be able to, sooner than you’d expect, produce less exquisite chips en masse and scale them across the softly rolling hills of Inner Mongolia and the vast rugged flats of Guizhou.[^72] Grids like farmland seen from an airplane, data centers in orderly rows powered by hydroelectric dams along the Yellow River. We have said before: there is no technology beyond the grasping reach of the Chinese state.

[^72]: Huawei, “Huawei Pioneers a New Computing Architecture for the AI Era: Making One Million Processors Work as One Computer,” September 17, 2026. Xinwen Lianbo, “A Rapidly Expanding Computing Network Set to Create Millions of Jobs,” Digital China Summit, June 25, 2026.

From the Chinese perspective, there may be a particular advantage to commoditized intelligence in the near future. A widely used open-weight model is a sponge for data, accumulating more the more widely it is used. We see an eventuality where the widespread use of a Chinese model could improve its capabilities, especially as we approach the era of continual learning. In this eventuality, then the following question becomes all the more consequential: How much of the future will run on Chinese intelligence? Fortifying America’s sovereign frontier will do nothing to diminish the appeal of ever-accessible Chinese intelligence.

We already have seen the spread of Chinese intelligence on the basis of price advantage and a narrowing delta of capabilities**. Over the course of the last year, publicized Chinese models went from a marginal portion of global token consumption on OpenRouter to more than a majority by volume by summer 2026.**[^73]

[^73]: OpenRouter, “AI Model Rankings,” usage data through September 29, 2026.

It is worth noting that OpenRouter is a distorted subset of global token consumption and represents those already inclined toward open-weight models with high capabilities. But the fortification of the American frontier is only in its early stages. Continue extrapolating this trend. For how many global users will Chinese models with near-equal capabilities and near-floor prices be the automatic choice? A minority? A majority? Even more? Without any effort to compete in commercial intelligence, which is impossible without a change in the expectations of value capture for the intelligence stack as a whole, Chinese models will continue to win share by volume and later benefit in capabilities as a result.

Leopold was right to insist that America’s frontier had to be secured against Chinese theft.[^74] But the financial threat survives even with perfect security. China can publish intelligence sufficient for most of the world’s work and destroy the value proposition that undergirds the American buildout. An American lab’s commercial customers will not be able to finance the next training run at the sovereign frontier. They couldn’t pay for that model if they tried. And of course they will, pleading their case to be read into the SAP. The line must be sustained, the run still needs to happen, and only the sovereign will be left to pay for it.

[^74]: Leopold Aschenbrenner, “IIIb. Lock Down the Labs: Security for AGI,” in *Situational Awareness: The Decade Ahead*, June 2024.

A Chinese sovereign frontier can still appear. None of our analysis requires that the entirety of the Chinese machine intelligence paradigm will remain open and index-setting in perpetuity. But in the short term, the discussions that the party may restrict overseas access to top Chinese models,[^75] greeted with relief in English-speaking countries, will not come to pass. If and when a Chinese sovereign frontier emerges, it will not necessarily negate the Chinese open ecosystem. Far from it. Those unfamiliar with Sinomodernity will be surprised to find that the modern Chinese state can encompass both a published model to compress foreign valuations and, in the future, a withheld Dongfeng model as strategic reserve.

[^75]: Fanny Potkin, “Beijing Is Looking at Curbing Overseas Access to China’s Top AI Models, Sources Say,” *Reuters*, July 7, 2026, updated July 9, 2026.

Such a paradigm provides the sovereign a considerable amount of flexibility and leverage over its only competitor. At no point will the Chinese party-state need to choose once and for all between an open ecosystem and a sovereign frontier. The party-state retains more options. It can restrict access. It can also publish and set the floor. Once China hits a critical mass of domestic chip fabrication in the near future, it will be able to do both simultaneously.

Few anticipated that the price of inference would fall so quickly, putting such pressure on the expected value of training. Chinese labs were seen as mirror images of our own, equally jealous in their protection of their margins. What America does, surely China will do. Wrong then, devastating now, and a mistake likely to be made again.

Sinomodernity is an arrangement at present strange to us and a vision of a future stranger still. They have latitude for the foreseeable future to continue to act at odds with our expectations. Expect their paradigm to hold.

As the Chinese continue their regime of price compression, American labs will be forced to finance the *n*th training run regardless of changes to their pricing power. The alternative, letting simple competition slow the rate of progress, would devalue the artifact too much. No, training expenditure must continue long after the ordinary returns on invested capital have begun to deteriorate. New financing mechanisms will be employed to keep the party going far after the value of the original expenditure is written down. We are watching the cascade today, from operating cash flow, then to IG bonds, devolving and devolving until finally coming to rest against the backstop of the sovereign.

In the end, the sovereign will truly announce itself by financing the continued diffusion of machine intelligence. Debt as we think of it will not meet the needed expenditure given the rigid structure of principal, maturity and coupon payments in the face of open-weight releases that impair the collateral. The final funding gap will be filled by instruments subscribed to the artifact itself, laundered out of short-termism, desiring to ride the ascent without regard for the calendar, an unshakable faith in the reshaping.

### Of Last Resort

We should not expect too much from looking to historical analogies. The sovereign retains a multitude of tools to exercise its will, few of which require swaying public opinion or making large displays of authority. Leopold understood this. One golden share, rightly structured, could empower its owner with effective control of the entire firm.[^76] Providing support through procurement, the indirect underwriting for the physical infrastructure and necessary hardware, or a stake in the lab itself (offered or taken), will amount to a new layer of governance, administered beyond the public view in intimately whispered imperatives.

[^76]: United States Steel Corporation, “Fifth Amended and Restated Certificate of Incorporation,” June 18, 2025, art. VI, sec. 1(a)–(j), exhibit 3.1 to Form 8-K.

Intensive credit underwriting and off-balance-sheet SPV structuring will reduce to a simple reliance upon a stream of government payments, the funded procurement contracts delivering salvation to the labs and their lenders. Private capital will continue financing the contractor after the state becomes its indispensable customer. State equity will make total Washington’s interest in the lab, but it will be the continuing public expenditure that will sustain the underlying work to be done.

A funding wave will break against the walls of the central bank. If the sovereign chooses to continue funding AI at the market-determined Epoch annual rate of 3.5x frontier training costs,[^77] it will place rapidly increasing demands on Treasury financing responsibilities. The Treasury will find buyers at higher interest rates, but servicing new borrowing and refinancing maturing debt will become more expensive. Washington will suppress those yields once they begin to constrain the sovereign program.

[^77]: Epoch AI, “Trends in AI,” section “Training Cost,” accessed September 30, 2026.

In 1942, the Federal Reserve fixed the yield of Treasury bills at ⅜ of a percent and capped the yield of long-term government bonds at 2½ percent by buying whatever private investors did not want at those rates. The Treasury-Fed Accord of 1951 ended this arrangement and liberated the Fed to pursue its own monetary policy, despite President Truman’s insistence that cheap financing for the government was needed to fight the Korean War.[^78]

[^78]: Robert L. Hetzel and Ralph F. Leach, “The Treasury-Fed Accord: A New Narrative Account,” *Federal Reserve Bank of Richmond Economic Quarterly* 87, no. 1 (Winter 2001): 33–34, 40–43, 51–52.

In the new Cold War, expect Washington to reverse that Accord. When maintaining the frontier becomes a matter of national defense, expect the Fed to once again fix the yields on Treasury notes and to buy whatever is needed to keep them there, even if Congress has to change its mandate to make it happen. Maintaining the frontier will take precedence over price stability and departmental power struggles.

### Two Sovereigns, One System

We tell you that nationalization will not look like nationalization in either America or China. To protect the artifact, do not expect to see any draconian legislation, nor much more of the discord that early characterized the relationships between the state and the country’s leading labs. Both sides will find it mutually agreeable and necessary. Leopold rightly recognized that nationalization could take a multitude of forms. Perhaps the labs would willingly join a national project through defense contracts or joint ventures with the sovereign.[^79] The Chinese and American sovereigns will recognize that frontier capability, like Polaris or Ju Lang, remains a strategic necessity long after its costs cease to be justifiable on commercial terms. The labs will accept sovereign authority as the price of remaining at the frontier.

[^79]: Leopold Aschenbrenner, “IV. The Project,” in *Situational Awareness: The Decade Ahead*, June 2024.

![Black-and-white photograph of a Polaris A-2 missile rising out of a cloud of smoke as it launches from the surfaced submarine USS Henry Clay off Cape Kennedy, 1964.](https://commanding-heights.com/img/polaris-a2-launch-1964.jpg)

*Polaris A-2 launch from USS Henry Clay, off Cape Kennedy, 1964.*

Picture credit: U.S. Navy, “Surface Polaris Launch,” April 20, 1964, photograph, USN 1094722, National Archives and Records Administration.

The program to develop the Polaris SLBM would never have made it out of Lockheed’s Corporate Policy Committee if it were a speculative commercial initiative. It was newfangled technology in a vertical that did not exist, and had only one prospective customer. The research project’s very existence required the sovereign to commission and underwrite the proposed work to get Lockheed to take the ambitious swing at storing ballistic missiles in submarines and then reliably launching them from underwater. Instead of the government just calling up its in-house USN R&D or Sandia and tasking them with making America’s first SLBM, they found the organizations best placed for success, where the subject-matter experts with the best chance of successfully swinging and hitting already were. They brought them under the hood, read them into the SAP alongside the Atomic Energy Commission’s weapons labs, and let them run their magic.[^80]

[^80]: L. E. Root and G. A. Busch, “Company Planning in the Defense Industry,” in *Proceedings of the Defense Market Planning Seminar, March 15, 1960* (Washington, DC: Electronic Industries Association, 1960), 60–61. William F. Whitmore, “The Origin of Polaris,” *Proceedings* 106, no. 3 (March 1980): 56–59.

This sovereign frontier will not be a re-creation of the USG offensive cyber program circa 2020, an in-house operation employing image board members from UIC developing zero-days at Fort Meade for GS salaries. Washington is not going to attempt to reconstitute an OpenAI of civil servants, nor does it have a practical reason to try. Capital has already organized global LLM talent into San Francisco’s labs, and the USG won’t outbid the private sector for employees; they have already awarded contracts to their employers.[^81]

[^81]: U.S. Department of War, “Classified Networks AI Agreements,” May 1, 2026.

Now, ask yourself why the most terrifically exciting and soon-to-be-valuable private companies would become so close with the state that they might be alleged to resemble an appendage. The answer is simple. The sovereign as shareholder can reduce risks that no private counterparty, no buy-in from the array of dollars and capital, ever could. We live in the age of private diplomacy,[^82] but even that term is simply a recognition of the growing influence that individual companies can have. It is a complement to state power, not a substitute. The state has no true competitors, and those who forget this, or ignore it, will spend their entire lives building something that could be crushed with a phone call from Washington. Lampreys on the Leviathan. The lab is no exception. It needs permission now to release its most capable model, predictable rules for foreign access, support for infrastructure, financial certainty, all to continue advancing the frontier. It will also need political legitimacy at a time when the gains, widely distributed among private actors, will still accrue to a small number of holders rather than the public. The new American arrangement, both financial and political, cannot be provided for by an S-1. It needs a sovereign to ensure its continued existence.

[^82]: Ian Bremmer, “The Technopolar Moment: How Digital Powers Will Reshape the Global Order,” *Foreign Affairs* 100, no. 6 (November/December 2021): 112–28.

The appreciation in value alone will be an afterthought to the sovereign. At first, this arrangement will need to ensure a certain level of stability at a time when stability is increasingly undermined. American public ownership, or influence, addresses a fundamentally political question that is unique to the United States. Between economic armageddon, runaway capabilities, and annihilation, the sovereign must strike a balance, a new settlement. Take extraordinary measures to ensure the line goes up, diffuse the technology, and reap the rewards in a way that asymmetrically favors America. Even if not explicitly stated in any single piece of legislation or executive order, assent of the masses will be necessary at some level. The taxpayer, a voter, man in another light, is being asked to support this new settlement. In turn, American workers will enjoy a privileged form of life when compared to the rest of the world. Above the new aberrant arrangement, the sovereign can and will solidify the expectation that the lab will be treated as a national asset instead of an ordinary vendor.

The commercial frontier will continue to command a premium. But America must also compete in commodity intelligence. Who will develop the model used by billions for tasks insignificant alone but of vital importance in the whole? If American commodity models cannot compete with the Chinese index on price, they will not be used. And if they will not be used, they will eventually be outpaced in capabilities upon the arrival of continual learning. To sustain accelerating frontier development and open-weight availability paced by the Chinese, the sovereign will ultimately need to ensure the righteous dispensation of capital. It will not be called nationalization and it will herald the unification of the interests of the state and the lab.

### Peculiar Laboratories

![Painting of the west front of Westminster Abbey, with a procession of the Knights of the Bath in red robes crossing the square before it.](https://commanding-heights.com/img/westminster-abbey-canaletto-1749.jpg)

*Canaletto, Westminster Abbey with a Procession of the Knights of the Bath, c. 1749.*

Picture credit: Westminster Abbey collection. Westminster Abbey, [“Canaletto Painting of the Abbey Goes on Display](https://www.westminster-abbey.org/abbey-news/canaletto-painting-display),” March 2, 2020.

The English have a name for such institutions set apart, answering only to the sovereign: royal peculiars. Where monasteries and churches had previously answered to Rome, the Reformation resulted in the Crown assuming the role of the sovereign. Henry VIII dissolved the constitution of Westminster Abbey in 1540. Successive monarchs exerted their personal preferences in the church, inviting the monks in, expelling the monks. Peculiars were subject to such whims and no others. Standing outside of the ordinary jurisdiction of bishops and answering to the Crown alone.[^83] A separate dispensation at a time of great turmoil.

[^83]: Westminster Abbey, “History of Westminster Abbey,” accessed September 29, 2026.

Now is a time not unlike then (cf. Anglomodernity). Waves of upheaval are already buffeting the outer reaches and they will soon be unavoidable. A member of technical staff is worth tens of millions if not a multitude more on an annual basis, capable of carrying out his work. From the perspective of an adversary, how much more would that researcher be worth dead or, worse, captured by state interests? We’re already seeing steps taken in China to protect essential tech workers and their immediate family members from being seized abroad and used as leverage. This was something Leopold got exactly correct: the exponential growth of machine intelligence as a determinant of comprehensive national power will come at the expense of the freedom of those who create it.

They put the scientists at Los Alamos to keep them safe and obscure. But obscurity will be impossible in the years to come. This puts the safety of those laboring on the project in question as well. What does security look like for the nation’s most vital assets and the hardware they need when every conventional idea about security no longer applies?

America a fortress, and inside its walls is yet another stronghold occupied by the labs’ sovereign research teams, packed away in their SCIFs. The work is silent and air-gapped. They’re contorting the model into unimaginable shapes, putting it through its paces. Such peculiar laboratories will answer only to the sovereign, subject to its whims and no others. Cloistered in the hills or out at sea. Like Latin in the Middle Ages, the modern cenobites will not work in the commonly spoken language. Researchers in habits obeying rules they need not understand. This is the arrangement that will endure.

### A Fearful Symmetry

In China, formal nationalization will be even less needed, despite the expectations of too many to the contrary today. They’re ahead of the curve. The party-state, that modern wonder, has always itself been a sovereign. In 1949 it emerged with an inherent authority that the American sovereign must now consciously claim for itself. Governance of machine intelligence in China is now as America will be: highly adaptive and hardly seen. Today, Chinese model developers developing generative algorithms must register with the Cyberspace Administration.[^84] This initiates an ongoing period of review that never really ends. The regulators, agents of the sovereign, will always be in contact as needed to assign responsibility for content and require public safety assessments, continual audits, and algorithm filing. The Chinese sovereign retains an even wider range of tools, formal and informal, seen and unseen, that the state can use to dispense its will. The principality and its power. In China, popular assent is a much less fickle thing. It has less need to crudely purchase popular assent, or at least tranquility, because the Chinese party-state never ceased to function as a sovereign. It was always there, the power of last resort. Less tolerant than it was, more aware of its precarious position. Gone are the days of belief in historical inevitability. There’s a reason that the Tiananmen Square protests happened the same year as the beginning of the collapse of the Soviet Union. If not the party-state, nothing stands in the way of chaos.

[^84]: Cyberspace Administration of China, “关于发布生成式人工智能服务已备案信息的公告（2026年7月至8月）” \[Announcement of Filed Generative AI Services, July–August 2026\], September 14, 2026.

There are those who would react to China’s ascent up the technological ladder with trepidation. We are watching it with great interest and the expectation that it will herald a parallel neoteric governing arrangement. It is a peculiar form of modernity that plays out every day across the Pacific, and we expect it to become far more peculiar from here.

The inescapable presence of the party-state sovereign, and its perception abroad, will influence how diffusion occurs. Those channels depend on the trust of users and a currency accepted beyond the issuing state. While America diffuses through a web of private rails that are not ultimately superseded by the sovereign in the same way, at least yet, China diffuses internationally through firms that are ultimately answerable to the party-state. But the reach of the sovereign only goes so far. For China, capturing the value of the capabilities it builds beyond its borders will be harder.

The rails of American diffusion rest on a foundation poured over two centuries. We expect a flattening in the language of business as a result of translation technology, but English will retain an irreplaceable value. New firms often choose a Delaware corporate wrapper and sell into markets governed by American commercial standards, financed by New York dollars. Trust, implicit or otherwise, here is a prime asset. Billions of people can keep savings in a dollar that reaches them through a private company on a phone because of the familiarity of the interface.

None of this advantage will be guaranteed into the future. The structure favors America, but it is precarious. The course of events in a democracy is uncertain. The next four years will show whether we are in for another hundred years of durable economic capture conducted in English or Mandarin. Perhaps both.

------------------------------------------------------------------------

The envelopment of the lab by the sovereign is more about function than form. Nationalization was once an act of public transfer, handing over the paper deed unwillingly. In the 19th and 20th centuries, the authorities would kick open the factory door and send the workers home. Foot soldiers would be left behind to guard the machines. A magistrate would formally change the title to the property. A workman would change the sign above the door. Private property was abolished by a very, conspicuously, public decree. Mass rallies, ferment among the firmament in Red Square. Authoritative radio broadcasts by the new order. That was a time of flags and uniforms, nations defined by blood and states delineated by borders on paper maps. Antiquated symbols and referents were of limited utility then. They have no place in the world bearing down upon us. Modernity, we know to be more exacting and more private, and far less forceful. In the past, you had to seize the Commanding Heights with soldiers and rifles. Today you need only own them.

The sovereign is, in practice, concerned with control beyond form. Where control already exists, or is offered, ceremonial nationalization becomes an unnecessary tool. The American sovereign is offered authority over the frontier because the work needs to be sustained. The Chinese sovereign needs no such offer. The modern state, Chinese or American, is less a butcher today than a surgeon, employing subtle tools with mechanical precision. And among our most precise is the Neo-Dollar.