Hook Speed was the only asset that didn't depreciate last cycle. This time, it's capital. Nvidia just posted a capex-to-depreciation ratio that’s climbing faster than a Bitcoin hashrate spike. The ratio is now above 2.0, meaning the company is spending two dollars on fixed assets for every dollar of depreciation. That’s not a normal semiconductor cycle. That’s a signal. Nvidia is betting the entire balance sheet on AI demand that might not exist in the same form six quarters from now. For crypto miners, this is a double-edged sword—one that could cut GPU availability and pricing in ways we haven’t seen since the 2018 crypto winter.
Context Let’s rewind. Nvidia’s H100 and upcoming B200 GPUs are the same silicon that powers generative AI models and, increasingly, proof-of-work mining. But the narrative has shifted. Since 2022, AI has cannibalized mining demand for high-end GPUs. Miners who once bought entire racks of A100s for Ethereum mining now compete with Google, Microsoft, and every VC-funded AI startup for the same chips. Nvidia’s response? Accelerate investment. The company issued $12 billion in corporate bonds in early 2025 and announced a $50 billion capex plan for 2025–2026—mostly for CoWoS advanced packaging and data center infrastructure. The goal: lock in supply chains and scale production faster than any competitor can replicate. But this aggressive capital deployment introduces structural risks that directly affect the crypto mining ecosystem.
Core The first risk is synthetic demand. Nvidia isn’t just selling chips; it’s financing the buyers. Through its venture arm, Nvidia has invested over $1.5 billion into CoreWeave and other GPU-as-a-service startups. These companies then use that capital to order more Nvidia hardware, creating a circular revenue loop that inflates reported sales. Based on my audit experience with DeFi protocols, this looks like a reentrancy exploit—except in the real economy. The TAM (total addressable market) for AI inference is still unproven outside of chatbots. If the next wave of AI startups fails to generate sustainable revenue, the used GPU market will flood with returned hardware. Crypto miners, who operate on thin margins, will be the first to benefit from cheap GPUs—but only after a price crash that destroys equipment value.
Volume tells the truth when price tries to lie. The real metric isn’t Nvidia’s revenue guidance; it’s CoWoS monthly output from TSMC. Right now, TSMC is struggling to ramp CoWoS-S and CoWoS-L beyond 40,000 wafers per month. Nvidia’s B200 demands double the CoWoS capacity of H100. Industry estimates suggest that if CoWoS output grows only 30% in 2025, Nvidia will miss shipment targets by 15%. That shortage will keep new GPU prices high, pushing miners toward older models and ASICs. But it also means that when CoWoS finally scales—maybe by mid-2026—Nvidia will have a glut of chips that need a home. By then, AI demand might have normalized. Miners are the natural buyer of last resort.
Meanwhile, the software monopoly is cracking. CUDA has been the moat that kept miners tied to Nvidia for algorithm customization. But AMD’s ROCm 6.0 now supports most popular mining frameworks, and OpenCL has gotten a second wind from Xilinx FPGA developers. The shift is slow but real. If miners can run the same hashing algorithms on AMD MI300X chips at 80% the efficiency but 50% the cost, the calculus changes. Nvidia’s pricing power erodes. That’s the same pattern we saw in 2018 when ASICs overtook GPUs for Bitcoin mining: the hardware becomes a commodity, and margins compress across the board.
Contrarian Here’s the part the mainstream crypto press misses: Nvidia’s capex acceleration isn’t a sign of impending doom. It’s a strategic play to own the entire compute stack before any competitor can build an alternative. The company is moving from being a chip supplier to an AI factory operator—DGX Cloud, Celestial networking, BlueField DPUs. This vertical integration means that even if the GPU surplus arrives, Nvidia can absorb excess capacity by renting it as cloud compute, not dumping it on the spot market. For miners, this is the real risk: Nvidia could become a direct competitor in the cloud mining sector, offering cheaper and more reliable hashpower than any decentralized pool. We didn’t see that coming.
Arbitrage isn’t just about price—it’s the market correcting its own soul. The current GPU market is split between two narratives: AI hype and mining pragmatism. Nvidia is trying to fuse them with capital intensity. But the fundamental law of semiconductor economics still applies: capacity always overcorrects. The last time Nvidia doubled down on capex, we got the RTX 30 series and the mining boom that followed. Miners who bought cards at $1,500 watched them drop to $400. The same pattern is likely to repeat, but this time the cycle will be driven by AI startups burning cash, not by Ethereum’s proof-of-work.
Takeaway The next six months will reveal whether Nvidia’s bet pays off. Watch CoWoS yields, watch AI startup funding rounds, and watch Nvidia’s data center segment gross margins. If margins slip below 70%, it means pricing power is fading. For crypto miners, the strategic move is to diversify hardware supply chains—L40S, AMD MI300, even Intel Gaudi 3—while keeping dry powder for the inevitable GPU fire sale. Survival is a strategy, but leverage is a mindset. The market is about to test both.
Speed was the only asset that didn’t depreciate last cycle. This time, it’s capital. But capital can turn into a liability faster than a bad smart contract.
