The numbers surged, but the room felt empty. Last week, Axios reported that Anthropic’s revenue run rate has hit $65 billion ahead of its highly anticipated IPO. For most, this is a validation of the AI boom—investor confidence, market dominance, a new benchmark for tech valuations. But as someone who has spent the last decade building decentralized protocols, I see something else: a quiet signal that the infrastructure of the AI gold rush is being built on sand. The same pattern that led to the DeFi liquidity mining bubble—subsidized growth, inflated metrics, and a disconnect between token value and real utility—is now playing out in the AI sector. And if we don’t learn from the past, the collapse will be felt not just in Silicon Valley, but across the entire blockchain ecosystem that is increasingly intertwined with machine learning models.
This is not a critique of Anthropic’s technology or its team. They are building remarkable things—Claude, the safety-focused assistant, is genuinely impressive. But the $65 billion run rate figure deserves a deeper dissection. Where is this revenue coming from? Is it organic demand for AI inference, or is it subsidized by venture capital, enterprise trials, and hype-driven API usage? I’ve seen this playbook before. In 2020, during DeFi Summer, protocols like Compound and Uniswap saw TVL spike to hundreds of billions, only to crash when liquidity mining incentives were halved. The same dynamic is emerging in AI: companies are burning cash to acquire users, and the run rate is a reflection of spending, not sustainable earnings.
As a decentralized protocol PM who has audited over 50 smart contracts for Gitcoin’s quadratic funding mechanism, I’ve learned to separate signal from noise. The signal in Anthropic’s run rate is not that AI is overvalued—it’s that the current centralized infrastructure for AI is inherently fragile. The high cost of inference, the opacity of model training, and the lack of verifiable compute are all problems that cry out for decentralized solutions. And that is where the blockchain industry can step in, not as a speculative side bet, but as the foundational layer for the next generation of AI.
The Context: AI Meets Blockchain
Anthropic, founded in 2021 by former OpenAI researchers, has positioned itself as the ethical alternative in AI. Its flagship model, Claude, emphasizes safety and alignment. The company has raised over $7 billion from investors like Google, Salesforce, and Spark Capital, with a valuation rumored to be north of $30 billion. The $65 billion run rate, if accurate, would imply a revenue multiple of roughly 2x—a figure that would make even the most optimistic venture capitalist pause. For context, OpenAI’s run rate was estimated at $3.4 billion in 2024, and Anthropic is claiming 20x that? Something is off.
But let’s set aside the skepticism for a moment. Even if the run rate is inflated by one-time contracts or forward bookings, the trend is clear: the demand for AI inference is skyrocketing. Every startup, from fintech to healthcare, is integrating LLMs. This is where blockchain enters the picture. Decentralized compute networks like Akash Network, Render Network, and io.net are already offering GPU time at a fraction of the cost of AWS or Google Cloud. But their adoption is still nascent, partly because the AI industry is locked into centralized cloud contracts, and partly because the infrastructure for decentralized inference is not yet mature enough to handle the scale of a Claude or GPT-4.
I remember a conversation in early 2022, during the Terra/Luna collapse, when I questioned the sustainability of algorithmic stablecoins. Many dismissed me as a pessimist. But the collapse taught me that any system that relies on continuous subsidization—whether through minting rewards or VC cash—is unstable. Anthropic’s run rate, if it is indeed driven by subsidized API usage, is no different. The parallel is uncomfortable but necessary: just as Terra’s Anchor Protocol offered 20% APY to attract deposits, AI companies are offering below-cost inference to attract developers. When the subsidies end, the users will leave.
The Core Analysis: Decentralized Compute as the Missing Layer
When the graph spikes, the soul remains quiet. The spike in Anthropic’s revenue run rate is a distraction from the real story: the infrastructure for AI is still horribly inefficient. The cost of running a single inference for a large language model can be measured in cents, but when multiplied by billions of queries, it becomes a massive expense. Most of that cost goes to cloud providers like AWS, Azure, and Google Cloud. These are centralized points of failure, subject to censorship, price hikes, and geopolitical risks. The blockchain community has been building alternatives for years, but the AI industry has largely ignored them.
Let me illustrate with a specific technical insight. ZK Rollups, which I have analyzed extensively in my work on Layer 2 scaling, are often touted as a solution for verifying computation. The same technology can be used to verify AI inference. Imagine a world where you can query an LLM and receive a zero-knowledge proof that the output was generated by the claimed model, with the correct parameters, and without any tampering. This is not science fiction—projects like Modulus Labs and Giza are already working on this. But the proving costs are still absurdly high. In my audit experience, I’ve seen ZK proofs that cost more to generate than the inference itself. Until these costs drop by an order of magnitude, decentralized AI will remain a niche.
Anthropic’s run rate, however, provides a powerful incentive to solve this problem. If the demand for AI inference is truly worth $65 billion—or even $10 billion in real terms—then the market for decentralized compute is enormous. The infrastructure builders who can reduce the cost of trustless inference will capture a significant portion of that value. I’ve been involved in similar transitions before. When I worked on the quadratic voting mechanism for Gitcoin, we reduced the cost of public goods funding by orders of magnitude compared to traditional grant-making. The same kind of innovation is needed in AI.
Take the example of tokenized AI compute. Several projects are attempting to create a market where GPU owners can rent out their hardware to AI developers, with payments settled on a blockchain. The idea is sound, but the execution is flawed. Most of these projects are simply rebranding cloud computing with a token, without addressing the core issues of trust, latency, and model portability. I’ve seen this pattern in the Bitcoin Layer 2 space: 90% of so-called “Bitcoin L2s” are Ethereum projects rebranded for hype, and the real Bitcoin community doesn’t acknowledge them. Similarly, many “decentralized AI” projects are just centralized APIs with a token wrapper.
The Contrarian Angle: Is the Run Rate a Mirage?
Here is the counter-intuitive thought: the $65 billion run rate might actually be a negative signal for the long-term health of the AI ecosystem. It reminds me of the Uniswap v2 liquidity mining crisis. In 2020, I was a Senior PM for a DeFi liquidity protocol, and I witnessed the chaos when protocols launched liquidity mining programs that attracted mercenary capital. The TVL spiked, but the users were not loyal. They left as soon as the incentives ended. The same is happening in AI. Companies like Anthropic are offering massive discounts on API usage to capture market share. Developers are building on top of these APIs, but they are not locked in. When the subsidies stop, they will switch to the next cheapest provider.
This is not sustainable. The real value in AI is not in the inference API; it is in the proprietary data and the model itself. But even that is fragile. Models are becoming commoditized—open-source alternatives like Llama and Mistral are catching up quickly. The moat Anthropic claims is safety, but safety is a feature that can be replicated. What cannot be replicated is the trustless, verifiable infrastructure that blockchain can provide. If Anthropic wants to build a lasting business, it should consider integrating decentralized verification into its offerings. But that would require a fundamental shift in mindset, from centralized control to distributed trust.
During the Terra/Luna collapse, I felt a profound sense of grief. I had believed in the dream of algorithmic stability, and seeing it shatter made me question everything. The same emotional vulnerability is present in the AI industry today. Many people are betting their careers on the assumption that the current growth rates will continue indefinitely. But history tells us that bubbles burst, and when they do, the survivors are those who built on solid infrastructure. For the blockchain world, this is an opportunity to demonstrate that decentralized systems are not just for speculation—they are for building the resilient foundations of the future economy.
The Takeaway: Building the Ethical Infrastructure for AI
When the graph spikes, the soul remains quiet. The soul of the AI industry is its potential to empower creators, not just investors. As a Creator Rights Defender, I believe that the next wave of AI should be built on open, verifiable, and decentralized infrastructure. Anthropic’s run rate is a wake-up call: the demand is real, but the supply side is broken. We need to bridge the gap between AI and blockchain, not through hype, but through practical engineering.
In my work with protocol engineers lobbying for clear regulatory frameworks ahead of the Bitcoin ETF approvals, I learned that translating technical concepts into policy is necessary but not sufficient. We also need to translate the ethical imperative of decentralization into business models that work. The $65 billion run rate is a signal that the market is ready for a new kind of infrastructure. The question is whether we, as builders, can deliver it.
So, I leave you with a rhetorical question: If Anthropic’s revenue is truly $65 billion, where is the proof? Where is the on-chain verification? Where is the decentralized governance that ensures the model is not a black box? Until those questions are answered, the quiet soul of the market will keep whispering—and those who listen will be the ones who build the future.