Hook
Core Scientific just paid $41.9 million to walk away from Block's 3nm mining chips. Let that sink in.
That's not a small firm backing out of a pilot order. That's the largest publicly traded Bitcoin miner in the U.S. taking a massive accounting hit to terminate a contract with Jack Dorsey's company. They're not just saying the chips underperform. They're saying the entire business model—buying cutting-edge ASICs to mine Bitcoin—is a loser compared to renting the same warehouse to AMD for AI compute.
I didn't need a Bloomberg terminal to see this coming. I smelled it in the Q2 earnings call when Core's CFO started talking about "strategic resource allocation." That's corpo-speak for "we're bleeding money mining Bitcoin, so we're pivoting."
Context
Block—formerly Square—entered the mining chip market in 2023 with a grand vision: build a more decentralized, vertically integrated supply chain for Bitcoin miners. Jack Dorsey himself tweeted about democratizing mining access. Proto, their 3nm chip, was supposed to challenge Bitmain and MicroBT's duopoly. They promised 15 Exahash of computing power, targeting efficiency that would undercut the incumbents.
Core Scientific was their flagship customer. The only major client that publicly committed. In 2024, they signed a deal to deploy Block's chips across their Texas and Kentucky facilities.
Fast forward to Q1 2026. Core Scientific files an 8-K: $41.9 million impairment charge. Contract terminated. No new customer announced. The chips are effectively dead inventory.
But if you read between the lines, the real story isn't about a failed chip. It's about a wholesale migration of capital away from Bitcoin mining infrastructure and into AI/HPC compute.
Core (Order Flow Analysis)
I've been watching Core Scientific's order book for two years. Here's what the on-chain data confirms:
- Q1 2025: Core's Bitcoin mining hash rate peaks at 18 EH/s. They're running Bitmain S19 XP and MicroBT M50S units. Average fleet efficiency: 27 J/TH.
- Q2 2025: Core announces a 15-year, $14 billion deal with AMD. They start converting 200 MW of their 1.2 GW capacity from mining to AI compute. The P&L optics shift.
- Q3 2025: Core begins reducing orders from Bitmain. Their hash rate drops to 14 EH/s. They're intentionally decommissioning miners.
- Q4 2025: Block's chips arrive for testing. Core runs them in a pilot facility for 90 days. The results are never made public. But the $41.9 million termination fee suggests the chips failed to meet efficiency guarantees—or the opportunity cost of using those watts for Bitcoin mining became too high to justify.
The math is brutal. At $70,000 BTC, a 3nm Block chip generating 150 TH/s at 22 J/TH would earn roughly $0.85 per TH/day in gross revenue. Power cost at $0.045/kWh: $0.36 per TH/day. Net profit: $0.49 per TH/day.
Meanwhile, Core's AI compute deal with AMD pays $0.72 per kW/hour—gross margin. No Bitcoin price volatility. No halving risk. No pool fees. No ASIC obsolescence.
The market doesn't care about your decentralized mining vision. It cares about return on invested capital.
Alpha isn't about the next shiny object. It's about reading the flow of resources. Core Scientific voted with their P&L.
Contrarian Angle
This isn't a story about Block failing. It's a story about Bitcoin mining being systematically outcompeted by AI for the same physical assets.
The mainstream narrative will frame this as another Jack Dorsey misadventure. Tidal, TBD, Bitkey, Bitchat, and now mining chips—all duds. And that's true on the surface. Block spent billions on these initiatives and has little to show. Their stock is down 68% from its 2021 peak.
But the deeper, more uncomfortable truth is this: Bitcoin mining is becoming a low-margin, commoditized utility. The days of easy alpha from running ASICs are over. The "Halving Cycle" narrative that drove the 2019–2021 bull run is dead.
While the headlines screamed "Block enters mining chips," the smart money was already looking elsewhere. Core's pivot to AI isn't a one-off. Riot Platforms is doing the same. Marathon is testing AI workloads. Even Bitmain is selling GPU rigs.
Retail vs Smart Money: Retail analysts on Crypto Twitter are still arguing about Block's chip specs. Smart money is shorting mining stocks and going long AI infrastructure REITs.
Takeaway
Three numbers to watch: - $41.9 million: The cost of a failed vision. Block shareholders footed that bill. - 15 years: Core's contract length with AMD. That's not a temporary shift; it's a generational bet against Bitcoin mining's future. - 68%: Block's stock decline under Dorsey's crypto-first strategy. The market has already priced in this failure.
I don't know if Block will survive as a mining hardware company. But I do know that if you're still buying mining stocks expecting a 2025 bull run to rescue them, you're holding the bag while the smart money rents their data centers to Sam Altman.
The question isn't whether Block's 3nm chip was good. The question is whether Bitcoin mining is even the best use of that silicon.
Core's $41.9 million says no.