AI has become embedded in crypto, and any disruption could hit the industry hard. Irina Karagyaur, co-founder and CEO of ecosystem growth agency BQ9, expects the biggest centralized AI systems to crash by 2030, undone by their own “success”. She believes decentralized physical infrastructure networks (DePin) could save the day.
We sat down with Karagyaur, who is also the president of Creators4Impact, to discuss why she thinks the business model behind the leading large language models (LLMs), like OpenAI’s GPT-4, is unsustainable, as noted in a recent essay. We also discussed how the “coming AI crash” will impact crypto assets.
Cryptonews: In your essay, you warn that centralized AI will “crash” not because of ethics or regulations, but because of infrastructure bottlenecks. Why do you think this risk is still being downplayed by major AI firms and investors?
Irina Karagyaur: It’s being downplayed because no company wants to miss out. When GPT-3 took the world by storm and exploded in popularity in late 2022, it triggered a crisis across the major tech companies. None of them, not even Google, which had been a pioneer in AI for years, had anything comparable ready for public release.
Sensing the shift, Microsoft quickly deepened its partnership with OpenAI, investing billions and integrating GPT models into Bing and the Office suite. Since then, other major players have scrambled to catch up.
Meta accelerated the development and release of its open-weight Llama models to assert leadership in open-source AI. Amazon ramped up investment in its Bedrock platform and custom foundation models. Even Apple, long silent in the generative AI race, began to roll out its own offerings in 2024–2025.
The philosophy has been, “The only risk is not to take the risk.” In this landscape, tech company leadership cannot afford to sit on the sidelines.
CN: Could you elaborate on why centralized LLMs, even if hugely successful, are inherently unsustainable at scale?
IK: There are many reasons for that. In summary, in our current paradigm, when technology is new, we rely on incremental brute force to grow it, and “more is better.” We don’t place the weight of our efforts on efficiency until we hit a wall.
For instance, Deepseek claims to have trained its V3 model for just US$5.6–6 million, versus around $100 million+ for GPT-4. The context was their difficulties in getting the tech and the budget, but on the other hand, excellent professionals and governmental support. When they disrupted the AI industry, all that was heard was “necessity is the mother of invention”.
That is exactly the point of my article, highlighting the need to develop different approaches to make AI, as a whole, sustainable and inclusive.
DePIN Solves Centralized AI Pitfalls
CN: You talk about Filecoin, Render, and DIMO as some of the key players in solving AI’s infrastructure crisis. What makes such projects suited to serve AI compared to traditional cloud providers like AWS or Azure?
IK: I’d like to highlight that there are other DePIN solutions doing the same in addition to the ones mentioned before. The point is that these decentralized technologies will solve the root cause of the problem.
Decentralization factors in people and nodes within the business model, as infrastructure cost (capex + opex). So we can split the costs among millions or even billions of nodes around the world. It is a cost-efficient approach that supports sustainable growth and human rights.
CN: Critics say centralized systems benefit from economies of scale. How can DePIN ensure decentralized infrastructure matches the performance and reliability of hyperscale data centers like those in the “Stargate Project”?
IK: Obviously, centralized systems can only survive from economies of scale. Especially if you want to create a monopoly. You will need to hinder competition and also build centralized systems as big as possible to take the market inside your cloud.
The Stargate Project was not [built] for technical reasons, performance, or reliability. The dominant narrative for its support from the U.S. government [is] to reach the supremacy of the American AI.
On the contrary, DePIN running AI means that running AI is a people’s job and a person is paid for that by the community. The whole point is to understand humanity’s own priorities. We can opt to make sure AI exists and coexists with us creating jobs and sustainable models. Or we can opt to run blindly, putting AI above us while destroying society, the environment, and even AI itself.
CN: How does DePIN handle sensitive AI training data compared to centralized systems? Are zero-knowledge proofs or federated learning viable for maintaining privacy in decentralized models?
IK: In theory, DePIN is in a position to handle AI training data in a more secure way by distributing computation and data storage which reduces single point of failure and the risk of data breaches. In practice, they need to be tested.
DePIN systems are required to integrate privacy-preserving techniques (like FL or ZKPs) on top of their infrastructure to protect sensitive data. These features aren’t built-in by default.
AI-Related Crypto Tokens Must Decentralize
CN: If major LLMs crash by 2030 as you predict, what would be the ripple effect on the crypto ecosystem, particularly AI-related tokens?
IK: The decentralized models will rapidly take the market share of those major LLMs left behind. Those who bet on decentralization and crypto rewards will have a big chance to grow exponentially.
CN: Many crypto tokens are now tied to AI narratives — some even depend on partnerships or APIs from centralized models like OpenAI or Anthropic. Are these projects at existential risk if those models fail?
IK: Absolutely. They have to create resilient business models by becoming truly decentralized.
CN: ChatGPT went down recently and left many crypto traders in a lurch. I would think a crash in major AI systems should have a more devastating impact on AI-driven crypto tokens.
IK: Not necessarily at this point in time. Sentiment around AI is largely positive. Crypto community is very optimistic about AI agents and related tokens. Therefore, since the ChatGPT outage in June lasted just a couple of hours (in context with electricity outages in some EU countries) it had limited impact on the sentiment. In the event that such outages become more frequent and people become more aware of the bottom line, we may well see a bigger impact on the markets as well.
CN: Could a collapse in AI infrastructure draw more attention to base-layer crypto networks like Bitcoin and Ethereum as resilient, censorship-resistant foundations for digital value?
IK: I’m convinced it will. Just like now, we see crises with centralized crypto exchanges (CEXes). It has become evident how those large centralized companies are extracting value not only from their massive user bases but also from the entities whose tokens they list. Decentralized networks and solutions will win over time. I’d say sooner rather than later.
Avoiding the Coming AI Crash
CN: Your essay proposes “profit sharing” to avoid the AI crash. Can AI and crypto-powered infrastructure realistically share profits with millions of individual contributors (e.g., node operators)?
IK: Actually the contrary: not sharing profits is impossible to imagine. Currently, AI uses the commons for training (and needs to be updated constantly). That is a source of income pending to be shared. This situation will change soon since many cases of intellectual property rights infringement have been taken to court.
On the other hand, AI mainly needs energy, compute, and storage, and in some cases, the provision of those resources is now funded with taxpayer money. In that sense, for years, Bitcoin and Ethereum and now many others have been successfully running nodes and paying the bills. Why can’t something as lucrative as AI do the same?
CN: You also speak of the “Internet of Value” emerging from the fusion of AI, blockchain, and the metaverse. What does a fully functional AI-Web3 application look like to you in 2030?
IK: One of my aspirations is that by 2030, such applications will become everyday tools that truly work for people, preserving our privacy and being built around our needs.
Let’s take a government website where citizens go to consult public services as an example. Imagine a citizen opens a single, easy-to-use portal. Their digital, not centralized ID gives them access to services like healthcare, voting, or applying for support, and their AI assistant helps them navigate the system, fill out forms, and even suggest ways they can earn tokens by contributing to community projects.
Crucially, the AI recognizes the value of their data and pays them for it, turning everyday digital interactions into sources of income. All of this happens with full privacy and transparency, protected by decentralized blockchain, and decisions — like where public funds go — are made through DAO-based online voting in a clear and accountable way.
This is the Internet of Value: a fairer digital world in which we own our data, shape our systems and work with AI to build a future that includes everyone.
CN: Could decentralization lead to new forms of inequality, where only those with hardware, bandwidth, or tokens participate in the upside?
IK: Of course! That is why we need the blockchain industry leadership to support this vision and the governments to intervene and regulate. We have to avoid creating disguised monopolies in the name of decentralization. For that, we have diverse Foundations, NGOs, or cooperative models that can be swiftly set up for that purpose.
The post Centralized AI Crash by 2030 Could Send Crypto Reeling | Interview appeared first on Cryptonews.
