The economics of censorship resistance

Censorship resistance is not a romantic flourish. It’s the inescapable logic of incentives.

When a platform’s survival depends on selling subscriptions in every jurisdiction from Canberra to Cupertino, government speech codes become non-negotiable costs of doing business. Centralized AI must therefore purchase compliance the way a factory buys steel: expensive, recurring, and ultimately decisive.

Last week CivitAI, one of the web’s largest model-sharing hubs, announced it will be “blocking access to the United Kingdom” because allowing UK traffic to its site would require full submission to London’s new Online Safety Act and Ofcom’s forthcoming codes of practice. The UK government is explicit: the Act “applies to services even if the companies providing them are outside the UK” whenever Brits can click the link. Faced with ruinous compliance audits, age-verification mandates, and open-ended fines, CivitAI chose exit over obedience.

The episode is not an outlier but a template. Any AI application that hosts user-generated content now shoulders the same liability burden as Facebook, minus Facebook’s war-chest of lawyers and lobbyists. That burden scales linearly with every additional nation-state you try to serve.

Consider Elon Musk’s xAI. Public filings show the Memphis “gigafactory of compute” already contains over 150,000 H100-class GPUs, part of a project that ballooned to $400 million in construction permits after raising $12 billion in outside funding. Such hardware investments alone force xAI to chase Netflix-level subscriber numbers. Yet those same numbers invite the full attention of Canberra, Brussels, Ottawa, Sacramento, and every other speech regulator on earth. The higher the capital expenditure, the higher the regulatory exposure, and the greater the temptation to throttle “controversial” outputs before a watchdog can issue a fine.

This is the moderation treadmill: spend billions, attract regulators, spend more to appease them, and pray one politically incorrect completion doesn’t tank the share price.

Now flip the incentive stack. Open-source projects run on community hardware, volunteer time, and a culture that judges code by whether it compiles, not whether it offends. IT Pro notes that 20,000 businesses adopted open-source AI tools in just the past year, flocking to Meta’s Llama, Mistral, and other libre models because proprietary licensing fees were becoming “simply not viable” for smaller enterprises.

Because the marginal cost of another fork is near zero, the community gains nothing by silencing unpopular speech and everything by maximizing adoption. A single problematic output doesn’t trigger a stock sell-off. It triggers a GitHub issue and a patch. Bug-fix economics, not brand-protection economics, rule the day.

Regulators, predictably, see such functioning, independent information eco-systems as an existential threat. Washington’s first Operation Choke Point pressured banks to debank politically disfavored but lawful industries like firearms and payday lending. Today’s digital reprise targets compute credits, hosting contracts, and payment rails for independent AI. CivitAI’s forced UK blockade is the proof-of-concept: if lawmakers can’t jail the code, they will starve the infrastructure.

Expect this playbook to scale: insurance carriers will hike premiums for “unmoderated” models. Cloud providers will quietly revise acceptable-use policies. Payment processors will discover “reputation risk.” The aim is identical to the original Choke Point: make participation so onerous that innovators self-select out of the market.

The economics are stark: centralized AI subsidizes censorship because compliance is the price of recouping colossal capital expenditure. Decentralized AI subsidizes free expression because its primary scarce resource is community goodwill. The establishment sees the writing on the wall: if it cannot throttle the latter, the former will hemorrhage users, capital, and ultimately relevance.

CivitAI was merely the first canary to keel over in the Online Safety mine. Unless the liberty-minded tech community builds parallel rails for compute, hosting, and finance, many more birds will follow. The good news is that code wants to be free, and, as history keeps reminding us, economics is downstream of incentives, not intimidation. The harder regulators squeeze, the more GPUs slip through their fingers.

---

Explore more from Popular AI:

Start here | Local AI | Fixes & guides | Builds & gear | AI briefing