Killer Switch: why decentralized AI is needed

AI has discreetly become critical infrastructure – as vital as electricity or the internet. Every business already depends on AI, often without even realizing the extent of that dependence:

• Customer Support: Chatbots handle millions of inquiries – API shutdown = paralysis of support services.
• Code Generation: Developers use Copilot and ChatGPT daily – productivity drops by 30-50% without them.
• Analytics: Models analyze data, generate reports, forecast demand.
• Content: Marketing, SEO texts, translations, design – AI automates routine tasks.

The API of a single company – OpenAI, Google, or Anthropic – becomes a single point of failure (SPOF) for thousands of businesses. OpenAI has experienced global outages multiple times: each one halted hundreds of thousands of applications built on their API. AWS us-east-1 downtimes caused cascading failures across the internet – and AI-dependent services suffered first.

OpenAI's projected losses – $112 billion by 2030 – question the sustainability of its business model. What happens if a company, on which millions of businesses depend, goes bankrupt or radically changes its terms? Dependence on a single AI provider is not just a technical problem; it's an existential business risk.

Centralization of AI creates four categories of risks, each of which is already manifesting in practice:

Censorship. Every AI company sets its own rules about what a model can and cannot answer. Models refuse to discuss certain topics, generate certain content, or answer “inconvenient” questions. For businesses, this means: you do not control the capabilities of the tool on which you build your product. Rules can change at any moment — without warning and without the ability to influence the decision.

Kill switch. Russia, Iran, China, North Korea, and dozens of other countries already face restrictions on access to AI services. A single decision by the US government — and OpenAI blocks its API for an entire country. This is not a hypothesis: it is already happening. For countries building their economy on AI, this is equivalent to being disconnected from the power grid.

Price blackmail. When a business is completely dependent on one company's API, the monopolist can raise prices without limits. OpenAI regularly changes tariffs, and companies that have invested millions in integration are forced to pay — because switching to an alternative will be even more expensive. Small businesses suffer first: $2.50–15 per million tokens is an unbearable amount for a startup in a developing country.

Data leaks. Every request to centralized AI passes through the servers of a single company. All your prompts, customer data, business logic, confidential documents — all of this is stored and potentially analyzed. Corporate secrets shared via ChatGPT are effectively under OpenAI's control — with all the inherent risks of leakage.

Decentralization eliminates each of these risks at the architectural level — not through management promises, but through the protocol's mathematics.

No single point of failure. The Gonka network runs thousands of independent MLNode instances around the world (~4,648 GPUs from ~113 participants). If one node goes offline, the Transfer Agent instantly redirects the request to another. A failure of 10%, 50%, or even 90% of the nodes does not stop the network — it continues to operate as long as there is at least one active node. This is the same principle that makes the internet resilient: a network, not a server.

No censorship. No company controls how the AI responds. Models in Gonka run without corporate “guardrails” — the network is open. This doesn't mean a lack of responsibility — it means the user, not a board of directors in San Francisco, decides on the acceptability of content.

Market-driven pricing. Competition between thousands of independent hosts ensures market pricing. The cost of an AI request through Gonka is $0.003 per million tokens, ~840 times cheaper than OpenAI. A monopolist cannot raise prices because there is no monopolist.

Privacy. Requests are not stored on one corporate server. They are processed in a distributed manner: the Transfer Agent directs the request to an MLNode, the node processes it, and returns the result. Your data is not aggregated in one place, which drastically reduces the risk of leaks. This is analogous to the difference between an intranet (one owner) and the internet (a distributed network): the latter is resilient by definition.

Decentralized AI is not only a technical solution but also a geopolitical tool. AI is called the “new oil” – and controlling AI computations becomes a matter of national security. A country without access to AI finds itself in the same position as a country without oil in the 20th century – dependent and vulnerable.

Bhutan became the third state-miner in the world (after El Salvador and the CAR), using cheap hydropower. This small Himalayan country with a population of less than a million people made a strategic bet on crypto – and is showing interest in DePIN projects like Gonka.

Uzbekistan is considering integrating state data centers into the Gonka network. For a country that cannot buy millions of GPUs from NVIDIA (quotas are limited, priority given to the USA and allies), connecting existing capacities to an open network is the only realistic path to AI sovereignty.

This pattern scales: dozens of “second-tier” countries – from Kazakhstan to Nigeria, from Vietnam to Argentina – have GPU capacities but lack access to advanced AI models. Gonka provides them with infrastructure: connect your GPUs, receive GNK for work, use AI without intermediaries. Digital sovereignty is not a slogan, but a concrete architectural possibility: countries, companies, and developers gain access to AI without dependence on Big Tech and without the risk of a kill switch.