May the Best System Win: What Kevin Rudd Can Teach Communities About Open-Source AI — Including China’s

The Counter-Stack, volume two: the geopolitics, Vol 2 of 3

“Never underestimate Xi Jinping. Never ever underestimate him.”

Kevin Rudd, former Prime Minister of Australia, on The Ezra Klein Show, July 2026

On July 14, Governor Kathy Hochul signed an executive order making New York the first state in the country to halt new hyperscale data centers — anything drawing 50 megawatts or more, paused for up to a year. That same week, Ezra Klein released a long conversation with Kevin Rudd, the Mandarin-speaking former prime minister of Australia who went back to school at 59 to read everything Xi Jinping ever published. On the surface, these two events have nothing to do with each other. Underneath, they are the same story: what happens to ordinary people when concentrated power owns the infrastructure of intelligence — and what communities can do about it.

The summer communities said no

Start with why you, a library trustee or county commissioner or co-op board member, are reading about Chinese politics at all. Average U.S. city electricity prices have climbed from about 13 cents per kilowatt-hour in 2019 to roughly 19 cents in early 2026 — a nearly 50 percent jump — and researchers point squarely at the data center buildout. New Yorkers have watched residential rates climb nearly 68 percent since 2019. Consumer Reports now covers data centers the way it covers defective cars. By one widely cited count, local opposition blocked or delayed some $98 billion in data center projects in a single four-month stretch last year, and communities keep organizing — lawmakers in more than 30 states have introduced hundreds of data-center bills this year alone.

Saying no is necessary. Saying no is not sufficient. A moratorium buys a year; it doesn’t answer the question of where your community’s AI capability will actually come from — because your schools, your clinics, your small businesses are going to use these tools either way. The Counter-Stack answer is community-owned, eco-friendly mini data centers: low-draw compute nodes (150–240 watts, about two bright light bulbs) running open-source models on rooftop solar, owned by libraries, municipalities, and cooperatives rather than rented from a landlord in the cloud.

And every time we present that plan, one hand goes up with the same question: “Aren’t the best open-source models Chinese? You want our library running Chinese Communist Party software?”

It’s a fair question. It deserves a serious answer, not a hand-wave. And the most useful guide we’ve found for answering it is a man nobody has ever accused of being soft on Beijing.

Why listen to Kevin Rudd

Kevin Rudd began his career as an Australian foreign service officer posted to Beijing in the mid-1980s, one of his country’s best Chinese linguists. He met Xi Jinping in person before almost any Western official did — in the 1980s, when Xi was a vice mayor and Rudd was, in his words, “the guy who does the photocopying.” He rose to become prime minister of Australia, then did something almost no former head of government does: he enrolled in a doctorate at Oxford at age 59 and forced himself to read the entire published corpus of Xi Jinping thought. The dissertation became a book, On Xi Jinping. He then served as Australia’s ambassador to the United States and now leads the Asia Society.

In the interview, Rudd is unsparing. Xi’s China is a Leninist party-state that has turned hard toward central control. Its surveillance apparatus — facial recognition, gait recognition, location tracking, every payment monitored electronically — gives it, in Rudd’s words, tools “that Mao and Stalin would have dreamed of.” Its citizens, facing youth unemployment and a collapsed property market, are told to chi ku — “eat bitterness” — for the national project. This is not a man selling you on China.

Which is exactly why his framework matters. If the West’s most clear-eyed China analyst can articulate a way to engage with Chinese technology without naivety — and he does — then a library board can too. Rudd never mentions open-source software; his subject is statecraft. But four of his lessons travel remarkably well from the situation room to the council chamber.

Lesson one: believe what powerful people say in public

Early in the conversation, Klein offers a principle that Rudd — a former practitioner — immediately endorses: the things politicians say in public are much more important than the things they say in private. Journalists chase the off-the-record comment, but leaders are careful in public precisely because public words set the parameters they must live within. Rudd read thousands of pages of Xi’s public writings and concluded the plan was all there, in the open: the party leads in all things, China becomes the world’s dominant industrial power, supply chains are weapons. Nobody had to leak it. You just had to take it seriously.

Now apply that discipline closer to home. The hyperscalers also tell you their plan in public — in earnings calls, capital expenditure announcements, and rate cases before your public utility commission. Hundreds of billions of dollars a year in data center capex. Terms of service that claim rights over your data. Utility filings that shift grid-upgrade costs onto residential ratepayers. None of this is secret. As Lina Khan told Paul Krugman in the conversation that anchored our previous post, the gatekeepers of each layer of the AI stack are already using that position to decide who gets to compete. Hope that they’ll behave differently is not a plan. Rudd has a line for this too: in Washington, he worries, “hope becomes a substitute for analysis and strategy.”

Lesson two: infrastructure is the ideology

Rudd’s deepest insight is about what Xi actually optimizes for. Not growth — Xi has deliberately sacrificed growth. Not consumer prosperity — households are told to eat bitterness. Xi’s organizing principle, in Rudd’s summary, is: “I’m building the world’s most powerful industrial state, and I’m not going to let any bunch of private entrepreneurs or consumers get in the road of that.” Control of supply chains, at home and abroad, is the whole game — because whoever controls the infrastructure controls everything downstream. Made in China 2025 was, Rudd says, the largest exercise in industrial policy in economic history, and it worked.

Rudd also describes the financing strategy behind it, and every community official should read this sentence twice. The plan, as he reconstructs it: subsidize production below cost until “the world’s competition and opposition… fold. Then we can restore prices at a level which makes sense for us… but by then the alternative sources of production have gone.”

Sound familiar? It should. It is also a fair description of how AI is currently being sold to your school district: inference priced below cost by companies losing billions a year, until the dependency is total and the alternatives have withered — at which point the prices, and the terms, become whatever the landlord says they are. The subsidy-until-they-fold playbook is not uniquely Chinese. It is what any sufficiently concentrated power does, under any flag.

The uncomfortable mirror

Set Rudd’s description of Beijing’s doctrine next to Khan’s description of the hyperscalers, and the rhyme is hard to miss:

Beijing’s doctrine, per Rudd	The hyperscaler playbook, per Khan — and your utility bill
Control the supply chains of the future: chips, batteries, rare earths	Control the AI stack: chips, cloud, models, and the apps on top
Subsidize below cost until foreign competitors fold, then restore prices	Price inference below cost until customers are dependent, then extract rents
Surveillance as governance: payments, cameras, facial recognition	Surveillance as business model: tracking, profiling, data extraction
Ask citizens to “eat bitterness” for the national project	Ask ratepayers to absorb ~50% electricity increases for the AI buildout
“The party leads in all”: government, military, academy, commerce	The platform leads in all: terms of service govern speech, commerce, access

To be clear, this is not a claim of moral equivalence — one of these systems imprisons dissidents and is, in Rudd’s careful assessment, rehearsing for something worse in the Taiwan Strait. The point is structural: both are centralized command systems whose power flows from owning choke points, and both ask the people underneath to absorb the costs. For a community deciding how to get AI capability, the axis that matters is not East versus West. It is centralized versus distributed. A hyperscale data center that doubles your electric bill does not oppress you less because its owner pledges allegiance to a different flag than Beijing’s.

The open-weight paradox

Here is the twist that Rudd’s interview helps us see clearly. The most effective tools available against both empires — Beijing’s party-state and Silicon Valley’s cloud lords — are open-weight AI models. And right now, many of the best ones come from China.

Recall the DeepSeek moment from our previous post: an outsider shipped a near-frontier model at a fraction of the compute cost, and Khan noted it exposed the incumbents’ structural conflict of interest — the firms that dominate AI also sell the compute AI runs on, so they profit from inefficiency. What’s happened since is an acceleration. Klein himself says it in the interview: the American story of China as a copier is dead. “America is keeping Chinese electric vehicles out not because they’re bad cars,” he observes. “It’s because they’re such good cars.” The same is now true of Chinese AI models — with one enormous difference. You can’t download an electric vehicle. You can download the models.

Model family	Builder	License	Why it matters for a community node
DeepSeek V4-Pro	DeepSeek (Hangzhou)	MIT	Top-ranked open model for reasoning and coding; 1.6T parameters but only ~49B active per query — efficiency by design
Qwen3 family	Alibaba	Apache 2.0	Most permissive license; sizes scale down to laptop-class — ideal for 150–240W nodes
Kimi K2.6	Moonshot AI	Modified MIT	Leading open model for long-running agent and tool-use work
GLM-4.6	Zhipu AI	MIT	The value champion for day-to-day coding assistance
Llama 4 Maverick	Meta (US)	Llama Community License	Strongest US open-weight option; native multimodality; some use restrictions
Mistral models	Mistral (France)	Apache 2.0 (most)	The European alternative; strong small models

Rankings shift monthly — that’s the point. Always verify the license text of the specific checkpoint you deploy. A good running survey is Understanding AI’s open-weight roundup.

Why would Chinese labs give this away? Three reasons, none of which require trusting anyone. First, U.S. chip export controls forced them to innovate on efficiency instead of brute force — and efficiency, as Krugman pointed out, is exactly what the American incumbents have no incentive to pursue. Second, it’s classic strategy: if you trail in the proprietary-model race, commoditize that layer and compete where you’re strong. Third, soft power. Their motives don’t have to be pure for the artifact to be useful — IBM didn’t fund Linux out of altruism either, and Linux now runs most of the world’s infrastructure, including, almost certainly, whatever device you’re reading this on.

And notice what constraint-born efficiency means for us: mixture-of-experts models that activate a sliver of their parameters per query, and small distilled models that run beautifully on a 200-watt node fed by rooftop solar. The models built under compute scarcity are precisely the models community-scale, eco-friendly infrastructure needs. The hyperscalers’ gigawatt campuses are not a law of physics. They are a business model — one that, as Khan said, treats inefficiency as a feature.

Weights are not services: the distinction that settles the China question

So: is it safe for your library to run a Chinese model? This is where one distinction does almost all the work. An open-weight model is not an app, not a service, not a connection to China. It is a file — a very large cookbook of numbers. When you run it on hardware you own, using open-source software your technologist can inspect (the leading inference engines are American- and European-built open projects), there is no phone line back to Hangzhou. Or to Northern Virginia. The legitimate worries about TikTok — a live service, streaming your data to servers under foreign jurisdiction, algorithmically tuned from abroad — simply do not transfer to a static file running in your basement. It works with the network cable unplugged.

Question	Their cloud app	Open weights on your node
Where does your data go?	To their servers, under their jurisdiction	Nowhere — inference happens on premises
Who can change or shut it off?	The vendor, any day, for any reason	Nobody — you pin the exact version, forever
Censorship and bias	Server-side filters you can’t inspect	Model-level biases you can test, document, and route around
What do you pay with?	Subscription fees and your data	Electricity and maintenance you control
Works during an outage or embargo?	No	Yes

Clear eyes require the caveats too. Open weights can carry baked-in biases from training — some Chinese models handle questions about Tiananmen or Taiwan evasively even when run locally. So you test: run your own evaluations, document what each model does well and badly, and route around weaknesses (civics homework help to one model, coding to another). You verify checksums when you download, pin versions, and use auditable open-source runtimes. This is not trust. This is verification — which is exactly the posture Rudd recommends toward Beijing, and exactly the posture we recommend toward San Francisco.

Managed strategic competition, community edition

Rudd’s signature framework for U.S.–China relations is called managed strategic competition, and it has three baskets. First, hard red lines — Taiwan, the South China Sea — enforced by credible deterrence, because clarity there is what makes everything else stable. Second, open competition everywhere else: trade, technology, talent, ideas. His words: “Make this a full and open competition, and may the best system win.” Third, active cooperation where interests genuinely align — he names AI governance and pandemic preparedness. Even Washington and Beijing, he notes, have begun negotiating on the most dangerous forms of AI.

Rudd built this framework so two superpowers could compete without sleepwalking into war. Scaled down, it is a complete operating manual for how a community uses everyone’s technology without becoming anyone’s subject:

The red lines do the heavy lifting. Once your community has committed that sensitive data never leaves the building — not to a Chinese server, not to an American one — the question “can we trust a Chinese model?” dissolves into the question you actually control: “does this file, running on our hardware, under our audit, do good work?” That is an empirical question. You answer it with benchmarks, not geopolitics. And the cooperation basket is where the network from our previous post comes alive: communities sharing evaluation results, deployment recipes, and solar-sizing calculators over the same open protocols — each node sovereign, all nodes stronger together.

A dignity test for every deployment

Asked for book recommendations at the end of the interview, Rudd offers a surprise: alongside two China books, he picks Pope Leo XIV’s encyclical on artificial intelligence, Magnifica Humanitas — “a great read,” he says, “about the essential dignity of the human person, and how AI uplifts that dignity or how it may impair that dignity.” The encyclical, published this May, warns bluntly against technologies concentrated in the hands of only a few people, and calls on the world to disarm AI before it corrodes human relationships and critical thought.

“Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together.”

Pope Leo XIV, Magnifica Humanitas (2026)

Set aside theology if it isn’t yours; the test travels. A gigawatt campus that raises a county’s electric rates 50 percent, drinks its water, employs almost nobody, and pipes the value to shareholders three time zones away is a Tower of Babel. A 200-watt node in the library basement, running open models on rooftop solar, answering patrons’ questions without recording them, owned by the people it serves — that is infrastructure at the scale of human dignity. When the world’s most hard-nosed China analyst and the Pope converge on the same worry — concentration — communities should probably take the hint. It is the same worry Lina Khan maps in antitrust terms and Varoufakis maps in economic ones. Everyone is describing the same tower.

What this means for your town

Pull the threads together and Rudd’s statecraft becomes a community checklist:

What Rudd teaches	What your community does
Never underestimate the centralizers	Take hyperscaler and party-state ambitions literally; plan for rent extraction, not benevolence
Public words beat private assurances	Read earnings calls and utility filings before granting abatements or signing contracts
Supply chains are power	Own the compute, the power source, and the connection — the counter-stack
Subsidy-until-they-fold is a real strategy	Never build civic services on below-cost cloud pricing you don’t control
Surveillance follows centralization	Choose architectures that can’t spy: local inference, zero telemetry, unplugged-capable
Managed competition beats cold war	Use the best open models from anywhere — under red lines you enforce

The pieces exist today. Open-weight models — Chinese, American, European — that rival the proprietary giants and run on hardware a library can afford. Solar panels that make a node’s operating cost a rounding error. WordPress and the Model Context Protocol turning 40 percent of the web into agent-ready civic infrastructure, as we detailed in The Counter-Stack. New York’s moratorium, and the RAISE Act before it, have bought communities something precious: a pause. The question is what gets built during it.

Further reading and listening

• “What Xi Jinping Wants” — Kevin Rudd on The Ezra Klein Show (and Rudd’s book, On Xi Jinping: How Xi’s Marxist Nationalism Is Shaping China and the World)
• The Counter-Stack: Building Profitable Enterprises on Open Source AI — the Krugman/Khan post this one builds on
• Yanis Varoufakis, “Cloud Capital vs AI: What DeepSeek Means for Technofeudalism”
• CEPR, “Data Centers and Price Spikes: Why the Public Should Own Its Utilities”
• Pope Leo XIV, Magnifica Humanitas — the AI encyclical Rudd recommends

OccupyAInet℠ is part of an ecosystem that includes the Broadband Institute Foundation, Verdant Data, and CivicBoxAI℠ from podCOINai℠ — building regenerative digital infrastructure: community-owned broadband, solar-powered mini data centers, and privacy-first AI.