SpaceX plans to add over 10GW of computing power by the end of 2027. That is enough to power a small country. Or, more precisely, enough to run the inference engines of the AI oligopoly.
I have spent the last seven years auditing the infrastructure of crypto networks. I have measured the hash rate of mining farms, the latency of DeFi oracles, and the token economics of layer-1s. But nothing prepared me for the sheer scale of this announcement. When SemiAnalysis released its report on Musk's compute targets, I felt the ground shift beneath my feet.
Hype is noise; structure is signal. The numbers are not abstract. They are the bones of a new feudal order. Let me dissect them.
Context: The Infrastructure Race Beyond Crypto
The AI industry has reached a critical juncture. Training models is no longer the bottleneck. Inference is. And inference requires hardware that is geographically distributed, energy-intensive, and exponentially expensive. OpenAI and Anthropic are now deploying on GB300 clusters — Nvidia's next-generation Blackwell architecture — which promise a 30x improvement in inference throughput per watt. But these clusters demand gigawatt-scale power and cooling.
Musk's SpaceX, traditionally a rocket company, has pivoted to building massive data centers in remote locations, leveraging its expertise in energy and logistics. The SemiAnalysis report claims that SpaceX's conservative target is to deliver 6-8GW of incremental computing power in 2027, with upside exceeding 10GW. At a capital expenditure of approximately $50 billion per GW, that translates to $300-500 billion in capex for 2027 alone.
To put that in perspective: the entire global crypto mining industry has a market cap of roughly $30 billion. SpaceX is planning to spend ten times that in a single year on compute infrastructure.
Beneath the yield lies the rot. The yield here is not just financial — it is the promise of artificial general intelligence. But the rot is centralization. Control over compute means control over the future of intelligence.
Core: A Systematic Teardown of the Numbers
Let me reconstruct the economic model from the SemiAnalysis report. I have seen similar projections in crypto — the difference is that this one is grounded in real contracts.
Revenue per GW: SemiAnalysis estimates that when OpenAI and Anthropic provide API inference services on GB300 clusters, each GW can generate over $100 billion in revenue per year. That is a staggering number. At a rental price of $3 per GPU per hour, the annual cost per GW is about $12 billion. The gross margin is 88%. No crypto mining operation has ever achieved that. The best ASIC farms in Bitcoin run at 30-40% margins after electricity and maintenance.
The Microsoft deal: In October 2025, Microsoft signed a $250 billion infrastructure agreement with OpenAI. SemiAnalysis estimates this corresponds to about 7GW of computing power. That is a price of ~$35.7 billion per GW. Compare that to SpaceX's $50 billion per GW capex — Microsoft is paying a premium for turnkey infrastructure. But the real story is the second contract: SemiAnalysis reports it is possible for Microsoft to sign a computing power contract with SpaceX for about 3GW, with a total value of approximately $150 billion. That is $50 billion per GW — exactly the capex cost.
SpaceX's recurring revenue: SemiAnalysis predicts that SpaceX's annual recurring revenue could reach $300 billion by the end of 2027. That would make it one of the largest infrastructure companies in the world, competing with Amazon Web Services.
Now, what does this mean for crypto?
First, the opportunity cost of capital. Every dollar flowing into SpaceX's compute farms is a dollar not flowing into decentralized compute networks like Filecoin, Akash, or Render. The AI industry is choosing centralized, vertically integrated infrastructure over peer-to-peer alternatives. Why? Latency, reliability, and legal compliance. The code does not lie, but the contract can. In crypto, smart contracts are immutable. In the AI world, contracts are signed on paper, enforced by courts, and backed by billions in collateral.
Second, the energy consumption. 10GW of compute requires about 87.6 terawatt-hours per year. That is roughly the annual electricity consumption of Belgium. Crypto mining currently consumes about 150 TWh per year. The AI industry is on track to consume more than crypto by 2028. But unlike crypto, AI's energy use is not decentralized or permissionless. It is controlled by a handful of companies.
Third, the hardware supply chain. GB300 clusters require advanced cooling and networking. The fabrication capacity for these chips is limited. If SpaceX takes 10GW, that leaves less for everyone else — including crypto miners who might want to repurpose AI chips for proof-of-work. Already, I have seen mining farms struggling to acquire new GPUs because the hyperscalers are buying them in bulk.
Aesthetic perfection often hides ethical voids. The beauty of the GB300 cluster — its efficiency, its speed, its sleek design — masks the ethical void of a compute monopoly. The geometry is the bone. The bone is oligopoly.
Contrarian: What the Bulls Got Right
I am not here to simply criticize. The bulls have a point. The scale of this investment is necessary for the development of AGI. Without centrally coordinated infrastructure, the training and inference of frontier models would be impossible. Decentralized alternatives are too slow, too unreliable, and too expensive per unit of performance.
Moreover, the revenue projections are not unrealistic. If OpenAI's ChatGPT can generate $50 billion in revenue today, and if inference demand grows 10x in the next two years, then $100 billion per GW is plausible. The unit economics are not a fantasy.
The bulls also argue that this infrastructure will eventually be commoditized. Just as AWS brought down the cost of cloud computing, SpaceX's massive scale will drive down the price of AI inference. That could benefit everyone — including crypto projects that rely on off-chain computation.
But there is a catch. The commoditization only happens if the market is competitive. Right now, it is not. SpaceX, Microsoft, and OpenAI are forming a vertically integrated triangle. They control the compute, the models, and the distribution. That is not a market. That is a fiefdom.
Silence is the loudest indicator of risk. The silence from the crypto community on this issue is deafening. While we debate the next L2 scaling solution, the real scaling of compute is happening behind closed doors, in boardrooms and data centers that no DAO can audit.
Takeaway: The Accountability Call
I have seen this pattern before. In 2017, ICOs promised decentralized everything. In 2021, NFTs promised creator sovereignty. In 2025, AI promises abundance. But the pattern is the same: a small group of insiders accumulates capital, builds infrastructure, and then extracts rent from the rest of us.
SpaceX's 10GW ambition is not a technological breakthrough. It is a political statement. It says: "We will build the future, and you will pay us to use it." The question is not whether this infrastructure is feasible. It is whether we, as a society, want to allow a single company to control the backbone of intelligence.
The code does not lie, but the contract can. The contracts being signed today are not smart contracts. They are legal documents that lock in lock-in. They are not auditable on-chain. They are not transparent. They are the new iron curtain, drawn not with steel but with silicon.
I measure the depth of the wave. The wave is 10GW deep. The wave is $300 billion recurring. The wave is 88% margins. And I see no decentralized alternative that can match it. Not today. Not by 2027.
If you are holding tokens in a decentralized compute network, ask yourself: what is your GW count? What is your revenue per GW? The answer will be silence. And silence is the loudest indicator of risk.
Beneath the yield — the rot of centralization. Under the mask of abundance — the geometry of control. The bone is there. We just have to look.
