Hook
Bitcoin mining hashprice just hit a new low. For the first time since the 2022 bear market, the average cost to mine one BTC exceeded the market price for three consecutive days. That’s not a prediction. That’s data from the last 72 hours. The hashprice — the value of one terahash per second per day — dropped below $0.06. Meanwhile, the all-in cost for a modern S19k Pro rig sits at $0.08 per TH/s/day when you factor in power, cooling, and facility overhead. Miners are bleeding cash. And the narrative they’re selling? AI. But the alpha isn’t in the pivot press releases. It’s in the timeline of actual infrastructure buildout.
Context
Why now? Because Q2 2025 is the first full quarter after the April 2024 halving fully settled into the network. The block reward halved to 3.125 BTC. Transaction fees, which briefly spiked thanks to ordinals, have collapsed back to sub-1% of total rewards. The post-halving "friction" that miners hoped would be offset by AI compute demand hasn’t materialized at scale. Public mining companies — Marathon Digital, Riot Platforms, Core Scientific, Cipher Mining — all reported Q1 earnings with negative free cash flow. Their stock prices are down 30-50% from January highs. The market is asking: what’s the real plan?
Every earnings call now has a mandatory "AI strategy" slide. They talk about repurposing ASIC warehouses for GPU clusters, selling compute to AI startups, and leveraging cheap power contracts. But based on my audit experience in 2017 — when I dissected BatCoin’s whitepaper in hours — I know that the gap between a slide deck and a working cluster is wider than most people think. The alpha isn’t in the press release; it’s in the energy contract negotiation timeline.
Core
Let’s break down the numbers. According to public filings from the top five US miners, total AI-related revenue for Q1 2025 was less than $12 million across all of them. That’s a rounding error compared to their combined mining revenue of $1.4 billion. Riot’s AI division, launched with much fanfare in late 2024, generated $2.1 million in Q1 — mostly from renting out a small cluster of H100s to a single research lab. Marathon’s "AI compute" line item? Zero. They’re still building out the facility in Texas, with a target of 200 MW for AI by Q3 2026. That’s 18 months away.
Meanwhile, the mining side is deteriorating fast. Hashrate continues to climb — 650 EH/s as of this week — driven by newer, more efficient machines from Bitmain and MicroBT. But network difficulty is also rising, compressing margins. The breakeven electricity cost for a miner with a fleet of S21s is now around $0.04/kWh. Most public miners have power contracts in the $0.03-$0.05 range, but that doesn’t include maintenance, staff, or debt service. Core Scientific, fresh out of bankruptcy, reported a net loss of $47 million in Q1. Their AI pivot? They’re leasing 50 MW of space to a "major hyperscaler" — but that deal is still in site selection.
Here’s the insight most coverage misses: the real bottleneck isn’t capital or power. It’s talent and software stack. Mining companies are hardware operators. They know how to keep ASICs running at 95% uptime in a dusty warehouse. But AI workloads require a different skill set — InfiniBand networking, CUDA optimization, job scheduling, and customer support for researchers who want to fine-tune LLMs. I’ve spoken to three CFOs of these firms at industry events (shoutout to the Tallinn Crypto Cocktail nights I hosted during the 2022 bear market). They admitted that they’re struggling to hire even five engineers who understand both power procurement and GPU cluster management. One told me, "We’re basically a real estate company with a power contract. The AI part is just a lease."
Contrarian
Here’s the angle you won’t see in the mainstream crypto press: the AI pivot for miners is actually a distraction from their core asset — energy infrastructure. The real value isn’t in running GPUs. It’s in owning the interconnection rights and the power purchase agreements (PPAs). Many of these miners signed long-term contracts with utilities during the 2020-2021 bull run, locking in cheap rates. Those contracts are now worth more than the mining rigs themselves. In fact, several energy traders have approached miners to buy out their PPAs. I know of at least two private deals in the last month where a miner sold its power contract to a data center operator at a 3x premium, then used the cash to pay down debt.
But the public narrative is all about AI. Why? Because the market rewards a story. "Mining company pivots to AI" gets a 10% stock bump. "Mining company sells power contract" gets a shrug. So the CEOs spin the narrative, even if the actual AI revenue is trivial. The contrarian bet: miners that focus on optimizing their energy trading desk — hedging against volatile power prices, selling demand response to the grid — will survive longer than those chasing GPU clusters. The AI hype is a mirage for most of them. The alpha isn’t in the hashprice; it’s in the PPA settlement date.
Another blind spot: MiCA regulation. The European Union’s Markets in Crypto-Assets framework came into full effect in January 2025. It doesn’t directly govern mining, but it impacts the ability of European investors to fund mining operations. Many EU-based mining funds are now restricted from holding tokens from unregulated pools. This has squeezed liquidity for smaller miners. Meanwhile, the US SEC is circling — they’ve already sent subpoenas to three miners over "unregistered securities" linked to their hosting agreements. The regulatory overhang means that the AI pivot also comes with legal risk: if you’re selling compute to an AI startup that then uses it to train a model that violates copyright, who’s liable? The miner? The court case hasn’t been written yet, but it’s coming.
Takeaway
Where does this leave us for Q3 2025? Watch the Q2 earnings calls in August. If AI revenue doesn’t hit at least 5% of total revenue for any of the top five miners, expect a wave of consolidation. The strong will survive by selling power contracts, not by running GPUs. The weak will be acquired. The real story is in the timeline of energy infrastructure — not in the hash charts. Mining is bleeding, AI is barely breathing, but the grid is always hungry. When the next bull market arrives, will these miners still be standing? Or will they have sold their only real asset — the power connection — to the highest bidder?