Block’s $41.9M Chip Failure: When Code and Capital Flee to the AI Grid

On February 18, Core Scientific paid $41.9 million to walk away from Block’s 3nm mining chip order. Not a renegotiation. Not a delay. A full termination. That’s the sound of a hardware project collapsing under its own weight. Code doesn’t lie. The impairment charge on Block’s latest 10-K says it all. Jack Dorsey’s Proto chip was supposed to challenge Bitmain. The narrative was clean: a vertically integrated Bitcoin miner from the founder of Square. But the reality is a 15 EH/s order that one year later was worth less than the penalty to cancel it. Core Scientific, once the largest public Bitcoin miner, chose to burn $41.9M in cash rather than take delivery of those chips. That is not a pivot. That is a veto from the only customer who mattered. To understand why, rewind the clock. Block’s mining hardware play began in 2021 when crypto was a blank check for any hardware story. Dorsey recruited a team, partnered with Core Scientific (then in Chapter 11), and announced a 3nm ASIC design. The market applauded. But the engineering reality of producing a competitive miner at scale is brutal. Bitmain holds 70-80% share. MicroBT holds another 15-25%. Both have decade-old supply chains, field-proven firmware, and energy efficiency ratios that Block could not match. I’ve seen this movie before. It ends with a fire sale. Core Scientific’s decision to cut ties is a forensic signal. The company had access to Block’s test chips in 2024. They ran them against their own fleet of S19s and M50s. The result: Block’s 3nm miner likely failed on efficiency (J/TH) or reliability. The market didn’t see the data, but Core did. And they paid millions to walk away. That’s the same type of signal I saw during the ICO audit sprint of 2017, when smart contract code revealed vesting cliffs that whitepapers hid. But the story does not end with a failed chip. Core Scientific itself is pivoting. Six months after terminating the Block order, they signed a 15-year, $14B+ contract with AMD to host AI compute. This isn’t a pivot. It’s a headcount reduction disguised as strategy. They’re turning their 400MW of mining capacity into an AI data center. The same land, same power lines, same cooling towers — but instead of SHA-256, they’ll run H100s. This is the most underreported structural shift in Bitcoin mining today: capital is voting with its feet. AI yields 20-30% net margins today. Bitcoin mining margins, post-halving, hover near single digits. Block’s crypto division, meanwhile, is a graveyard. Tidal was shut down. TBD’s Web5 went nowhere. Bitkey wallet adoption is negligible. Bitchat launched and died. Cash App paid $2.5B in fines for compliance failures. The stock is down 68% in five years. This isn’t a series of bad bets. It’s a systemic failure of governance. Jack Dorsey’s CEO authority was unchecked. The board allowed a capital allocation strategy that burned billions on hardware R&D, music streaming, and identity protocols — none of which had a clear path to profit. Based on my experience tracing on-chain insiders during DeFi liquidity trap exposures in 2020, I can tell you when a CEO’s pet project becomes a black hole. The signs are always there if you read the footnotes. The contrarian angle is this: Block’s failure is actually good for Bitcoin mining. It reinforces the moat of the incumbents and discourages capital from chasing new mining hardware narratives. The 3nm market will remain a duopoly. But the larger take is that mining is losing its grip on the best infrastructure. Core Scientific’s real estate will now serve AMD, not the Bitcoin network. That trend is accelerating. Every megawatt that shifts to AI is a megawatt that no longer backs Bitcoin’s hashrate. If this scales, the network’s security model relies on a shrinking pool of price-insensitive miners. What to watch next. Block’s Q2 call will likely announce a full write-off of the Proto business. That’s priced in. The bigger signal is whether other public miners — Riot, Marathon, CleanSpark — announce similar AI partnerships. If they do, the narrative flips from “Bitcoin miner” to “AI REIT with Bitcoin exposure.” That’s a re-rating catalyst. But it’s also a structural risk for Bitcoin’s long-run economics. The question I keep asking: when the last 3nm chip leaves Block’s warehouse, who will buy it? The answer may be no one. Code doesn’t lie. And neither does a $41.9M goodbye.