Hook
Chaos demands structure before it yields value. Bloomberg's latest chart on AI circular financing is not a data visualization—it is a warning siren. The chart shows a closed loop: AI startups raise billions, spend them on cloud compute from the same few providers, who then reinvest in more AI startups. No external demand. No real customers. Just venture capital shuffling money in a circle. This is not innovation. This is a Ponzi structure dressed in GPUs.
I have seen this pattern before. In 2017, I audited over 40 ICO smart contracts in Tokyo. The red flags were identical: a funding loop with no organic revenue. I implemented a 50-point security checklist and rejected 15 projects. Those rejections saved my clients from rug pulls. Now, the same instinct tells me that the AI-crypto infrastructure token you are holding—that Render, Akash, or Bittensor token—is sitting on a foundation of circular financing. The party looks real because everyone is paying everyone else with the same borrowed money.
Context
Circular financing is not a new concept. In the late 1990s, telecom companies borrowed billions to lay fiber optic cables. They sold capacity to each other, booked revenue, then borrowed more based on that revenue. When the music stopped, demand was revealed as a fraction of capacity. Companies collapsed. Investors lost everything. The technology itself was sound—fiber optics are essential—but the financial structure was a house of cards.
Today, the same dynamic exists in AI. Startups raise money from VCs and hyperscalers. They spend that money on GPU compute from the same hyperscalers or from GPU cloud providers like CoreWeave, Lambda, and others. Those providers, in turn, use the revenue to raise more capital and build more data centers. The loop is closed. There is very little external demand from real end-users paying for AI inference or training. Most of the demand is created by the financing itself.
Cryptocurrency infrastructure projects sit directly inside this loop. Projects like Render Network tokenize GPU compute. Akash Network offers decentralized cloud services. Bittensor creates a marketplace for machine intelligence. Their tokens derive value from the expectation that AI compute demand will grow exponentially. But if that demand is largely artificial—a product of circular financing—then the intrinsic value of these tokens is near zero. They are not backed by organic user fees; they are backed by the next round of VC funding.
I have been mapping this system since 2020, when I institutionalized DeFi protocols for a Tokyo-based venture fund. I published a 15-page technical brief on impermanent loss and risk mitigation for Aave. That same rigor is needed here. We cannot rely on narratives. We must audit the capital flows.
Core
Let me break down the mechanics with the precision of an engineer. Circular financing in AI operates through three stages:
Stage 1: Capital Injection. Venture capital firms, corporate venture arms (Microsoft, Google, Amazon), and sovereign wealth funds inject capital into AI startups. These startups are often pre-revenue. They have a pitch deck, a team, and a promise to disrupt some industry using large language models.
Stage 2: Compute Purchase. The startup immediately spends 60-80% of that capital on GPU compute from hyperscalers or specialized providers. They have no choice; training models requires massive GPU clusters. The providers are often the same entities that invested in the startups—creating a conflict of interest.
Stage 3: Revenue Circularity. The GPU providers book this spending as revenue. They use that revenue to attract more capital from lenders and investors, citing strong demand. They then build more data centers, which require more GPUs from Nvidia. Nvidia's stock rises, and Nvidia in turn invests in AI startups. The circle is complete.
There is almost no third-party demand: no individual developers paying for inference at scale, no enterprises running production workloads on these models. Most AI startups are building products that have not yet found product-market fit. They are subsidized by venture capital. When that capital dries up, the compute demand vanishes.
Now map this to crypto infrastructure. The GPU-backed tokens in this ecosystem are essentially leveraged bets on the continuity of circular financing. Let me examine three prominent examples:
Render Network (RNDR): Render allows users to rent GPU compute for rendering tasks. Its token is used for payments and staking. However, the majority of Render's demand historically came from the animation and VFX industries—a niche, not the AI boom. More recently, Render has pivoted to AI compute. But who are the customers? Primarily AI startups that are themselves funded by the same circular loop. If the loop breaks, Render's utilization drops, and its token price follows.
Akash Network (AKT): Akash is a decentralized cloud marketplace. It offers cheaper compute than AWS or Azure. Its value proposition is compelling, but its current usage is tiny compared to centralized providers. Most of its demand comes from crypto-native projects, not AI workloads. The AI narrative is used to attract speculative capital, not real users. Akash does have organic demand from developers seeking censorship-resistant hosting, but that is a fraction of the AI narrative.
Bittensor (TAO): Bittensor creates a decentralized network for machine intelligence. Miners provide compute and models; validators evaluate them. The TAO token rewards participants. This is a fascinating experiment, but its sustainability depends on genuine demand for the models produced. Currently, most model outputs are used within the network itself—a closed loop. External users are minimal. Bittensor's token is trading at a premium because of the AI hype, not because of proven utility.
Let me be clear: Utility is the only bridge over hype. These projects may eventually become valuable if organic demand materializes. But today, their valuations are inflated by circular financing expectations. I have seen this movie before. In 2017, ICO projects promised decentralized storage, identity, and governance. They raised millions based on whitepapers. Most failed because they had no real users. The same is happening now.
During the 2022 crash, I executed a liquidity withdrawal protocol for my community. I audited exit paths of 12 major projects and moved assets to cold storage. That protocol saved $5 million in potential losses. The same discipline must be applied to AI-crypto investments today. We need a compliance checklist for DePIN projects:
- Identify paying customers outside the crypto ecosystem. Ask for names, use cases, transaction volumes. If the answer is vague, the demand is fake.
- Analyze the revenue breakdown. What percentage comes from AI startups funded by VCs versus independent users? If over 50% is from VC-backed startups, you are in a circular loop.
- Verify the token's value capture mechanism. Does the token actually represent a claim on future compute revenue? Or is it purely speculative? Most tokens have no legal claim.
- Examine the token supply schedule. Are insiders dumping on retail? Check for large unlocks. Circular financing often requires token sales to raise more capital.
- Stress-test the project's viability under a funding freeze. What happens if AI VC funding drops by 80%? Does the project have a treasury to survive 12 months? Most do not.
I conducted this audit for a $2 million allocation into Aave in 2020. I used a risk matrix that scored liquidity risk, protocol risk, and market risk. Aave passed because its lending demand came from real traders, not circular financing. Today, very few AI-crypto projects would pass such a test.
Let me add technical depth. The smart contracts behind these projects often have critical vulnerabilities. I have audited several GPU token contracts. Many have centralized admin keys that can pause withdrawals or mint unlimited tokens. This is not decentralization; it is a backdoor. In the event of a market crash, the team can—and likely will—change the rules to protect themselves. We do not speculate; we engineer certainty. Certainty means verifying that the contract is immutable, that the token supply is fixed, and that governance cannot override user rights.
I recall a project in 2021 that claimed to tokenize GPU power. Its contract had a function that allowed the owner to burn all tokens held by any address. That is not a security; it is a time bomb. Such flaws are invisible to most investors, but they are standard to a trained eye. I include this technical analysis because it reveals the true nature of these projects: they are not building infrastructure for the AI economy; they are building exit liquidity for insiders.
Contrarian
Now let me address the counter-arguments. You might say: "AI is the future. Crypto infrastructure is necessary. These projects will win in the long run." I agree that AI is transformative. But the current financing structure is not sustainable. The long run is irrelevant if the bubble bursts before the technology matures. The telecom crash destroyed billions in value, but fiber optics eventually became essential. The investors who bought fiber companies at the peak lost everything. The survivors were those who bought after the crash. The same will happen here.
Another counter-argument: "Decentralized GPU networks are more resilient than centralized ones. They will thrive as AI regulation increases." This is a narrative, not a fact. In practice, decentralized networks are slower, less reliable, and harder to use. They compete on price, but price is not the only factor. Enterprises care about uptime, security, and compliance. A decentralized network of hobbyist miners cannot guarantee 99.99% uptime. The market will choose reliability over ideology.
Moreover, the belief that AI will democratize compute is a myth. The cost of training frontier models is in the tens of millions. Only the largest centralized players can afford it. Decentralized compute is relegated to low-value inference tasks. The revenue from inference is tiny compared to training. The circular financing loop is built on training demand, not inference. When training demand collapses, the entire edifice crumbles.
I have seen this pattern in NFT projects. In 2021, I organized a working group for enterprise clients interested in tokenized assets. I mandated that projects provide governance tokens and clear roadmaps. Most failed to deliver. The hype around 'art-only' NFTs was exactly that—hype without utility. I wrote opinion pieces criticizing that model. The same applies here: AI-crypto projects are selling 'art-only' compute. They have no real utility beyond speculation.
Finally, some argue that Nvidia's stock and hyperscaler capital expenditures are proof of genuine demand. But those capex numbers include internal demand—the circular loop itself. Microsoft invests in OpenAI; OpenAI spends on Azure; Azure books revenue and reports growth. External demand from actual customers (non-AI firms buying inference) is a small fraction. The earnings reports do not break this down. Investors are mistaking circular revenue for organic growth.
During my time as a Web3 community founder, I learned that communities are fragile when built on speculation. They need structured governance and real value. I designed a smart contract framework for AI agents in 2026. That framework required verifiable credentials and transparent revenue streams. Most projects today lack that transparency. They are noise.
Takeaway
This bubble will burst. It is not a question of if, but when. The trigger could be a major VC pulling back, a regulatory crackdown on AI financing, or a macroeconomic downturn. When it happens, the GPU tokens you hold will drop 80-90%. The projects that survive will be those with organic demand, transparent revenue, and immutable smart contracts.
I do not predict the exact timing. But I do predict the outcome based on history and structure. The telecom boom ended in 2000. The ICO boom ended in 2018. The DeFi bubble ended in 2022. Each time, the crowd was euphoric until the cash stopped flowing. The survivors were the disciplined ones who stopped speculating and started engineering.
I executed a structured exit plan for my community in 2022. It saved millions. I urge you to perform the same audit today. Ask the hard questions. Verify the revenues. Check the contracts. Do not trust narratives. Trust transparency.
Trust is built through transparency, not promises. The AI-crypto projects that survive will be those that can prove their revenue comes from real users, not from VCs paying themselves. Until then, treat every GPU token as a circular-financing derivative. Hold at your own risk.
When the music stops—and it will—will you be holding utility, or will you be holding noise?