Hook
Bitcoin hashrate dropped 20.6% in August 2025. Price went up 34.9% over the same period. That’s not a dip. That’s a structural decoupling. The last time this happened was 2012 — before the first halving. Back then, it was a technical glitch. Today, it’s a liquidity signal. Miners are voting with their power lines. They’re not selling coins. They’re selling joules. To AI.
I’ve been tracking this since my 2017 ICO arbitrage days — when we launched three utility tokens in Southeast Asia and watched 80% fail because they had no liquidity model. Skepticism isn’t about doubting technology. It’s about following the flow of capital. And right now, capital is flowing out of Bitcoin’s security budget into AI inference racks.
Context
The Bitcoin network is running fine. Block time is 9 minutes 56 seconds — within the 10-minute target. The difficulty adjustment algorithm (DAA) cut difficulty by 4.19% in August. Hashprice rebounded to $39.36/PH/s/day — above the 30-day moving average. On paper, miners should be happy.
But they’re not.
Puell Multiple sits at 0.73 — the 16th percentile. That’s the lowest since the 2022 bear. Miners are earning less in USD terms than they did a year ago, even with the price rally. The halving cut the block reward to 3.125 BTC. And now, the opportunity cost of mining Bitcoin is higher than ever.
Enter AI.
Institutional capital is pouring into high-performance computing (HPC) clusters. Hyperscalers like Amazon, Google, and Microsoft are starving for GPU capacity. And miners — with their existing power infrastructure, grid connections, and data center sites — are the perfect hosts. IREN, TeraWulf, and Riot have already announced multi-year contracts with AI firms. Riot signed a 20-year deal with Anthropic. That’s not a pivot. That’s a lock.
Core
Let’s map the numbers.
Bitcoin’s 7-day average hashrate fell from 1,150 EH/s in June to 914 EH/s in late August. That’s a loss of 236 EH/s — equivalent to roughly 3.5 million S19 XP miners going offline. But the network’s absolute hashrate is still above 900 EH/s. The 51% attack cost remains astronomical. Security isn’t threatened today.
What’s threatened is the elasticity of hashrate.
In previous cycles — 2018, 2022 — when price dropped, miners turned off old machines, difficulty dropped, margins improved, and hashrate bounced back. That mechanism is now broken. Why? Because the power that was once allocated to Bitcoin is now under long-term AI contracts. You can’t flip a switch back. Power purchase agreements (PPAs) with AI firms are 3–5 years, sometimes longer. The electricity is committed. The GPUs are installed. The ASICs are mothballed or sold.
Look at IREN. They slashed their Bitcoin miner deployment by 17% this quarter. Instead, they’re installing NVIDIA H100 clusters for their AI cloud service. Their CEO explicitly said: “We see higher risk-adjusted returns in HPC.” That’s a direct quote.
TeraWulf did the same. They converted 40% of their facility to AI co-location. Their Q2 2025 revenue from AI services already exceeded Bitcoin mining revenue.
And then there’s Mara, Bitdeer, Riot — they’re actually expanding Bitcoin mining. But they’re the exception. The dispersion is real. The industry is splitting.
This creates a unique dynamic: the DAA is working, but the recovery signal is weaker. Hashprice improves, but hashrate doesn’t respond. Because the miners who left are not coming back — they’ve already found a better yield curve.
Liquidity doesn’t lie. It just moves.
Contrarian Angle
The mainstream narrative is: “Miners are pivoting to AI, which is bullish for their stock valuations and neutral for Bitcoin.” I disagree. The spillover effects are structural.
First, the “miner capitulation” self-healing mechanism is now partially disabled. Historically, the bottom of a Bitcoin bear market was marked by a sharp drop in hashrate followed by a rapid recovery as weak miners died and strong ones bought cheap hardware. That model assumed miners had no alternative. Now they do. AI is a permanent exit ramp.
Second, the security budget of Bitcoin is not just a technical metric — it’s a priced variable. Investors pay a premium for Bitcoin because it is the most secure, most censorship-resistant asset. That premium is backed by the cost to attack the network. If hashrate falls structurally, the attack cost falls. Over time, that premium erodes. The market has not yet priced this risk. The price-to-hashrate divergence is a canary.
Third, the regulatory dimension. The SEC’s regulation-by-enforcement approach has deliberately withheld clear rules for crypto. Meanwhile, AI is getting a red carpet. The CHIPS Act, export controls, and national security funding all favor AI infrastructure. Miners are rational actors. They’re following the regulatory path of least resistance. This isn’t a technology failure. It’s a policy-driven resource reallocation.
Takeaway
Bitcoin’s difficulty adjustment is not failing in a technical sense. It’s failing in an economic sense. The algorithm assumes that miners will always return when margins improve. That assumption is now false. The next 12 months will test whether Bitcoin can sustain a security model that competes with AI for the same physical resources. If the trend continues, we may see a new equilibrium hashrate — 30-40% lower than the historical trend. That changes the risk profile of Bitcoin permanently.
The question is not whether miners will sell. The question is whether they will ever come back.
Skepticism isn’t about doubting Bitcoin. It’s about doubting that the old rules still apply.
