We are watching a strange kind of dawn. On August 15, Anthropic PBC quietly revealed to potential investors that its second-quarter revenue had grown 13x year-over-year, crossing $11.5 billion in preliminary figures. Adjusted operating profit turned positive for the first time in Q2 2026. The numbers are staggering—$787 million in the same quarter last year, $4.73 billion in Q1 this year. The engineers at Anthropic are building something that the market is craving. But as I read the leaked documents, a different kind of signal flickered in my mind, one that has nothing to do with share prices or valuation multiples.
We are watching the creation of a new sort of walled garden, one that covers the entire sky. And the question for those of us who believe in decentralization is not whether Anthropic’s technology works—it clearly does—but whether we are willing to let the most powerful cognitive tool ever built be controlled by a single boardroom. From code audits to community heartbeats, I have learned that trust is not a protocol, it is a practice. And practices require many hands, not just one.
Context: The Architecture of Centralized Intelligence
Anthropic’s business model is built on a foundation that is invisible to most users. Their flagship model, Claude, is trained on vast datasets, fine-tuned through reinforcement learning from human feedback, and served through APIs that require a central server. The revenue growth reflects a simple truth: enterprises are hungry for reliable, safe, and powerful AI. They are willing to pay for it. The $11.5 billion quarterly figure comes from a mix of API subscriptions, enterprise licensing, and custom model fine-tuning. The company’s adjusted operating profit turning positive suggests that the cost of training and inference is now being covered by recurring revenue—a sign of a mature business.
But here is the architectural tension. Every query sent to Claude is processed by a server owned by Anthropic. Every piece of data that flows through the API is visible to the company. The model’s weights are proprietary, protected by trade secrets and patents. The company’s board has the power to change the model’s behavior, to censor outputs, to prioritize certain customers over others, or to comply with government requests. In a centralized system, trust is a property of the institution, not the protocol. And institutions can be compromised, bought, or corrupted.
This is not a criticism of Anthropic’s team, which includes some of the most thoughtful researchers in the field. It is a critique of the architecture itself. In 2026, I helped draft the Decentralized AI Bill of Rights, a consensus document signed by 500 Web3 organizations. One of its core principles was that the decision-making power over AI models should be distributed across stakeholders, not concentrated in a single entity. Anthropic’s revenue numbers are a testament to the efficiency of centralization, but they also highlight the risk: the more valuable the system, the more tempting it becomes to misuse it.
Core: The Data-Dependency Paradox
Let me dig into the numbers with a technical lens. Anthropic’s revenue surge is largely driven by enterprises that are integrating Claude into their workflows. For example, a major bank might use Claude to automate compliance checks, a healthcare provider might use it to analyze medical records, and a logistics company might use it to optimize supply chains. These use cases require the model to process sensitive data. The data is sent to Anthropic’s servers, where it is temporarily stored for inference and potentially used for fine-tuning. The bank, the healthcare provider, and the logistics company are effectively outsourcing trust to Anthropic’s security team, its compliance officers, and its internal policies.
But here is the paradox. The very data that makes Claude useful is also the data that makes it dangerous. If Anthropic’s servers are breached, or if a rogue employee accesses the data, the consequences are catastrophic. And even if the company is perfectly run, the data is still subject to subpoenas, government requests, and legal obligations. In a decentralized AI system, the data never leaves the user’s device. The model is split into small pieces, each processed locally, and the results are aggregated via a consensus mechanism. The user retains control over their data. The cost is lower efficiency, higher latency, and more complex engineering. But the benefit is sovereignty.
Based on my experience auditing the Telegram Open Network whitepaper in 2017, I recognize a pattern. The TON whitepaper promised a decentralized internet, but its incentive structure ignored small-holder participation. The project collapsed because the community felt unheard. Anthropic’s revenue growth is impressive, but it is built on a similar blind spot: the assumption that users will be satisfied with a black box, as long as the outputs are accurate. I have seen this before. Users start with enthusiasm, then demand transparency, then demand control, and finally leave for alternatives that offer both.
In 2020, when I founded the Mumbai Chain Guardians, we translated 50 technical upgrade proposals for Aave and Compound into simple, empathetic guides. The goal was to help users understand what was happening under the hood. Transparency reduced panic and built trust. Anthropic’s model is opaque. The company publishes alignment research, but the actual weights, training data, and inference logs are hidden. The $11.5 billion revenue is a measure of convenience, not trust. And convenience without trust is a fragile foundation.
Contrarian: The Bull Case for Centralized AI—and Why It’s a Trap
Let me play the devil’s advocate. There is a strong argument that centralized AI is simply better. It is faster, cheaper, more reliable, and easier to integrate. The open-source and decentralized AI models, like LLaMA or Bittensor’s subnetworks, require significant technical expertise to run. They are not as polished. They lack the safety guarantees that Anthropic’s extensive red-teaming provides. For a large enterprise, the choice is obvious: pay for Claude, get instant value, and move on.
Moreover, the revenue growth suggests that the market is voting with its wallet. The $11.5 billion figure is not a speculative bubble; it is based on actual usage. The adjusted operating profit turning positive means that Anthropic is not just a hype machine—it is a sustainable business. Investors are right to be excited. And the AI industry is still in its early stages. The total addressable market for AI services is trillions of dollars. Anthropic is well-positioned to capture a significant share.
But here is the contrarian angle that I believe is missing from the mainstream narrative. The very efficiency of centralized AI creates a monoculture risk. If all enterprises rely on the same model, the same infrastructure, and the same decision-making processes, then a single vulnerability—a bug in the model, a policy change, a regulatory action—can cascade through the entire economy. Decentralized systems are less efficient, but they are more resilient. They are like diverse ecosystems versus a mono-crop farm. The farm produces more food per acre, but a single disease can wipe it out. The ecosystem produces less, but it survives.
Furthermore, the revenue growth may be masking a deeper problem: the commoditization of trust. Anthropic is monetizing the trust that users place in its brand. But trust is not a protocol, it is a practice. And practices cannot be scaled indefinitely. As the company grows, the relationship between the user and the model becomes more transactional. The human touch disappears. The ability to audit the model, to verify its decisions, to contest its outputs—all of these become harder. The company becomes a gatekeeper, not a partner.
In my 2021 work with the Tata Trusts on the Heritage on Chain NFT project, I saw how blockchain could be used to preserve cultural dignity. The project raised $150,000 in ETH, with 70% going directly to artisan communities. The key was that the community owned the tokens. They controlled the narrative. Anthropic’s model, by contrast, is owned by the company. The community has no say in how it evolves. The $11.5 billion revenue is a testament to the company’s ability to capture value, but it says nothing about the distribution of value. In a decentralized model, value flows back to the participants. In a centralized model, value flows to the shareholders.
Takeaway: The Infrastructure of the Heart
When I look at the Anthropic numbers, I do not see a threat. I see a challenge. The challenge is to build a decentralized AI infrastructure that is not just a toy, but a real alternative. We need systems that can handle enterprise workloads, that can match the latency of Claude, that can provide the same safety guarantees—but without centralizing control. The technology exists. Bittensor, Gensyn, and other projects are making progress. But they need funding, talent, and adoption.
Anthropic’s revenue growth is a signal that the market is ready for AI. It is also a signal that the market is being conditioned to accept centralized control. The next few years will determine whether we build bridges or walls. Building bridges where DeFi once built walls is my life’s work. And I believe that the same principles that guided the Ethereum community, the same resilience that kept the Mumbai Chain Guardians together during the 2022 crash, can be applied to AI.
Digital artifacts that remember who we are—that is the promise of blockchain. Anthropic’s Claude remembers your prompts, but it forgets your values. The audit was just the beginning of the bond. The real bond comes when the community has a stake in the model, when the model is transparent, and when the users are not just consumers but co-creators. Liquidity flows, but culture remains. And the culture of decentralization is not about rejecting efficiency—it is about demanding that efficiency serve the many, not the few.
So let us celebrate Anthropic’s success. It proves that the market is hungry. But let us also remember that hunger can be satisfied in many ways. The path forward is not to fight centralization, but to out-build it. We have the tools, the experience, and the values. The question is whether we have the courage to use them. Trust is not a protocol, it is a practice. And it is time to practice it louder.
