A leaked tender document, whispered across the Pacific, has landed on my desk. It speaks of a plan so audacious it could reshape the global AI landscape—and yet, as someone who has spent years teaching the ethics of blockchain, I see a different story. Anthropic, the AI safety company founded by former OpenAI rebels, is quietly negotiating to secure 1.4 gigawatts of data center capacity in Australia. By year-end, they want one gigawatt live. The price tag: $15 billion. This is not just an infrastructure play; it is a moral statement on the nature of power in the digital age. The document, shared by a source who remains anonymous for fear of reprisal, reveals a six-week window for a final investment decision. And while the crypto world obsesses over memecoins and layer-2 throughput, I am haunted by a question: in our rush to decentralize everything, have we forgotten that the most critical resource—computational power—is being concentrated into fewer hands than ever before?
Let me step back. Anthropic, valued at over $60 billion after its latest funding round, has traditionally leaned on Amazon Web Services for its compute needs. But the scale of their ambition—training models that rival or surpass GPT-4o and Claude 4—demands independence. This 1.4GW figure is staggering: it is roughly equivalent to the power consumption of a mid-sized city. For context, OpenAI’s largest single cluster on Azure is around 0.3GW. Anthropic is not merely scaling up; they are building a fortress. The document suggests splitting the capacity into four or five smaller contracts, a tactic I recognize from my days auditing smart contracts for the ZEIP-20 working group. Back in 2017, I spent six months scrutinizing token transfer logic in Nairobi, discovering 42 edge cases where code favored centralized validators. The lesson was clear: technical neutrality masks systemic bias. Now, Anthropic’s strategy of multiple vendors is a form of risk diversification, but it also reveals a deeper truth—they are betting that controlling the hardware layer is the ultimate moat. In blockchain, we call this ‘sovereignty.’ In AI, they call it ‘competitive advantage.’
The technical specifics demand our attention. 1.4GW of power, activated within months, requires modular construction, liquid cooling, and low-latency networking on a scale rarely seen. I reached out to a former colleague who now advises hyperscale data centers for a living. ‘You’re looking at 100,000+ GPUs,’ he said. ‘H100s or B200s, probably InfiniBand. The heat alone would melt a traditional facility.’ This is why the choice of Australia matters: cheap renewable energy, political stability, proximity to Asian markets, and land that isn’t locked in litigation. Yet the hidden risk is chip supply. Any disruption in NVIDIA’s production—or a sudden export control tightening—could turn this ambitious timeline into a ghost campus. During my time auditing DeFi protocols, I learned that a single oracle failure can cascade into a billion-dollar liquidation. Here, the oracle is the semiconductor supply chain. And unlike a smart contract, you cannot simply fork the hardware.
But let’s talk about the elephant in the room: the contradiction at the heart of this project. I built The Open Ledger in 2020 to teach locals in Kenya how to navigate liquidity pools and yield farming. We published whitepapers in Swahili, emphasizing that decentralization means shared ownership, not just distributed ledgers. Yet here is a company—arguably the leading ‘responsible AI’ voice—building a centralized compute empire that could become the single largest user of electricity on the continent. In my NFT collective, Savanna Voices, we structured a DAO to ensure 70% of royalties went back to artists. We fought against the extractive nature of hype cycles. Now, I see Anthropic doing the same thing: extracting value from a region’s resources (land, power, labor) while the benefits flow to shareholders in San Francisco. The irony is bitter. Tracing the moral code behind every token means asking who truly owns the keys to the kingdom.
Here is my contrarian angle: what if this massive investment is actually a sign of weakness, not strength? The pressure to activate 1GW by year-end suggests desperation. Anthropic may be locked in a secret race to finish training a model that requires compute they cannot access through existing cloud contracts. I remember the bear market of 2022, when my platform lost 60% of its donations and I had to rewrite 40% of the curriculum. That vulnerability taught me that desperation leads to shortcuts. If Anthropic cuts corners on cooling or network redundancy to meet the deadline, they risk downtime that could erase months of progress. Moreover, the $15 billion debt may force them to prioritize revenue over safety—exactly the opposite of their founding mission. Building libraries where others build empires is not just a phrase; it is a call to resist the temptation of centralization. The crypto community understands this when it comes to validators, but we remain silent when AI companies amass power over the hardware that runs our models.
What does this mean for us? The blockchain industry has long championed the idea of ‘compute as a commodity.’ Projects like Render Network, Akash, and Golem promise that anyone can rent out idle GPUs from home. But the numbers tell a different story: the total capacity of these decentralized networks combined is less than 0.05GW. Anthropic’s single project is 28,000 times larger. If decentralized compute cannot scale, then the vision of a truly open AI—free from corporate gatekeeping—remains a fantasy. In 2026, as I co-authored the African AI-Blockchain Ethics Charter, we argued that access to compute should be a human right. But rights are meaningless without physical capacity. Walking away from the hype to find the soul means admitting that our ideals are outpaced by economic reality.
The takeaway is not doom, but a challenge. Anthropic’s plan will likely proceed, and it may even succeed. But as someone who has spent a decade teaching that code is law, I must remind you: the law is only as just as the infrastructure that supports it. I have seen firsthand how the OpenSea royalty surrender destroyed the creator economy in NFTs. I have watched DAOs fracture because multisig admins held the real power. Today, I see the same pattern repeating at a macro scale. The question is not whether Anthropic can build this data center—they will—but whether the rest of us can build alternatives that are truly distributed. Ethics is not a feature; it is the foundation. If we ignore this lesson, we risk building a world where a handful of companies control the very substrate of thought.
— Tracing the moral code behind every token. — Building libraries where others build empires. — Listening to the silence between the blocks.