When NVIDIA dropped its Q4 2025 earnings on February 20, 2026, the numbers did something rare: they forced even the most hardened crypto maximalist to pause. Data center revenue hit $35.7 billion, up 112% year-over-year, while gaming—the segment that once housed miners buying RTX 3090s by the pallet—fell 12%. Bitcoin's network hashrate, meanwhile, grew at an annualized rate of just 4.3%, the slowest since the post-FTX recovery. The scent of a resource war is in the air, and math does not care about your conviction—it only cares about marginal returns.
Over the past seven days, I watched three separate mining operations liquidate their GPU inventory on secondary markets. One of them emailed me a spreadsheet that read like a eulogy: 4,200 RTX 4090s, sold at 60% of book value, cash deal. The buyer wasn't a miner—it was a small AI startup in Singapore looking to fine-tune a finance model. The narrative is clear: AI is not just competing for attention or mindshare; it is competing for the physical silicon that was once the backbone of crypto mining. And it is winning.
Context: The History of Shared Silicon To understand the shift, you have to go back to the great GPU bull run of 2020–2021. Miners hoarded cards, driving prices 200% above MSRP. Then came the 2022 crash, when Ethereum transitioned to Proof-of-Stake, flooding the market with cheap GPUs. For a brief window, mining was dead. But AI labs like OpenAI and Anthropic had just discovered that the same tensor cores used to mine Ravencoin could train language models. By 2023, NVIDIA's H100 was serving both worlds: miners used it for high-throughput hashing; AI trainers used it for inference. For a while, coexistence seemed plausible.
That coexistence ended when the math broke. I remember sitting in a coffee shop in Auckland in August 2024, building a simple model comparing the revenue per watt for an H100 in mining vs. renting it as an AI node on a cloud provider. At the time, AI rental rates were 3.2x more profitable per hour than mining. By January 2026, that gap had widened to 5.7x. The invariant is brutal: any algorithm that can be efficiently run on a GPU will eventually be outbid by an AI application that generates higher marginal value. Solitude is the price of clear vision—and in the solitude of my spreadsheet, I knew mining would be priced out of the general-purpose GPU market.
Core: The Mechanistic Reallocation of Capital Let me tell you what the market isn't saying. The common narrative is that AI and mining are separate sectors that happen to use similar hardware. That's a comforting lie. The truth is that both depend on a shared bottleneck: the world's limited capacity for advanced semiconductor manufacturing. TSMC's 5nm and 4nm nodes are running at 98% utilization. Every wafer allocated to an NVIDIA B200 AI chip is a wafer not available for a Bitmain S21 ASIC. And AI chips carry a 70% gross margin for NVIDIA; mining ASICs carry maybe 30% for Bitmain. Math does not care about your conviction.
I've been modeling this since my days auditing Golem in 2017. Back then, I discovered that the reward distribution mechanism assumed fees would remain stable—a naive error. Today, the error is that we treat mining power as a fixed stock. It's not. The hashrate is a function of hardware availability, electricity cost, and expected revenue. AI bids up the price of efficient hardware, which increases the cost of hashrate. If the cost of hashrate rises while the block subsidy stays flat (or falls), the equilibrium hashrate drops. That is what we are seeing now: Bitcoin's difficulty adjustment has been negative for three consecutive periods—the first time since 2018.
But the deeper insight is behavioral. I've spent the last year interviewing institutional investors who allocate to both AI and crypto. Their logic is elegant in its simplicity: AI has a clear demand signal (enterprises paying for inference), while mining has a circular demand signal (miners paying for hardware to earn tokens that other speculators hold). The crowd sees a moon; I see a model—and the model says that capital will flow to the narrative with the most direct real-world utility. That narrative is AI.
Contrarian: The Miners Are Not Dead—But They Are Mutating This is where I annoy the AI cheerleaders. The narrative that AI “kills” mining is too simplistic. Miners are not passive victims; they are adaptive organisms. Consider Bit Digital, the once-pure Bitcoin mining company, which now generates 18% of its revenue from AI cloud services using repurposed GPUs. Hut 8 just announced a $150 million retrofit of its Alberta facility to host liquid-cooled racks for AI training. These are the smart ones.
The contrarian angle is that the competition is actually a forcing function for mining to evolve into something more resilient. Miners who can access stranded energy—solar farms in the desert, hydroelectric dams in Sichuan, flared gas in Texas—will still generate profits, because AI data centers need stable, low-latency power, not variable surplus. Mining can absorb cheap excess energy that AI cannot. This is the invariant in chaos: the lowest-cost producer always survives.
Yet there is a blind spot. The mining hardware supply chain is still dependent on the same foundries that serve AI. As long as AI orders are 10x larger, TSMC will prioritize them. I've seen the future in the scrap market: old GPUs pile up, and altcoin mining becomes a hobby for electricity-subsidized speculators. The golden age of GPU mining for profit is over. ASIC-based coins like Bitcoin will persist, but even their economics will tighten as the cost of new ASICs rises due to competition for silicon.
Takeaway: The Next Narrative is About Energy Sovereignty Where does this leave us? Narratives are liquid; truth is solid. The truth is that mining will no longer be a growth industry for general-purpose hardware. The next narrative is not about hashrate—it's about energy sovereignty. The projects that will survive are those that position themselves as grid-balancing assets or carbon capture mechanisms. The ones that compete head-to-head with AI for GPUs are walking into a trap.
I am positioning my fund accordingly: long on companies with proprietary ASIC designs and access to stranded energy, short on any mining operation that relies on buying off-the-shelf GPUs from the same bins used by AI labs. The shift is already priced into the numbers. The question is whether you are listening to the story the data is telling, or chasing the story you want to hear.
Coding the future, one block at a time.