Hook
April 2025. Core Scientific drops a bombshell: it pays $41.9 million in termination fees to walk away from Block’s 3nm Proto mining chips. The contract was signed less than a year earlier. A penalty that large is not a negotiation tactic—it’s a strategic divorce. Why would a miner choose to burn nearly 42 million dollars instead of taking delivery of state-of-the-art hardware? Because the cost of operating those chips, in both opportunity and capital, would have been far higher. This is not a headline about a single deal gone sour. It is the autopsy of a narrative shift. The age of Bitcoin mining as a standalone profit center is ending. The era of the flexible data center—where AI compute leases outbid hash power—has begun.
Context
Block Inc., formerly Square, is Jack Dorsey’s vision for a decentralized financial future. In 2021, he announced a foray into Bitcoin mining silicon, aiming to break the duopoly of Bitmain and MicroBT. The Proto project was supposed to deliver a 3nm ASIC that could challenge the Antminer S19 and Whatsminer M50 series. Core Scientific, one of the largest publicly traded miners, became Proto’s anchor client. Meanwhile, Block diversified into music (Tidal), decentralized identity (TBD/Web5), self-custody wallets (Bitkey), and a social protocol called Bitchat. All of them have since been shuttered, written down, or quietly abandoned. The only crypto bet that survived—Cash App’s Bitcoin trading—now faces regulatory heat from the CFPB and state regulators, costing Block over $200 million in penalties.
Core Scientific, by contrast, is a survivor of the 2022 credit crisis. It emerged from bankruptcy with a renewed focus on efficiency. In late 2024, it began repositioning itself as a data center operator for high-performance computing. The Proto contract was a remnant of the old strategy. By paying the $41.9 million goodbye, Core Scientific signaled that its future lies not in Bitcoin mining, but in leasing its power-optimized facilities to AI giants like AMD.
Core
The technical heart of this decision is the chip itself. Block never published independent benchmarks for its 3nm Proto ASIC. Competitors like Bitmain already ship 3nm units with proven energy efficiency—around 25-30 J/TH. Without data, the market assumed Block’s chip would be competitive. But Core Scientific’s behavior speaks louder than any spec sheet. They weighed the expected profit from running Proto chips against the profit from hosting AMD’s GPU clusters. The AMD contracts are long-term (15 years), low-risk, and backed by explosive AI demand. Bitcoin mining margins, post-halving, are razor-thin. The $41.9 million penalty becomes a cheap price for flexibility.
I have audited mining operations since the 2020 DeFi Summer. Back then, the question was always: “How many EH/s can you deploy?” Now, the question is: “What is the highest and best use of your power capacity?” That shift is structural. Core Scientific’s pivot echoes a pattern I observed in 2022 when miners rushed to convert natural gas flares into hash power. The arbitrage has moved from energy to narrative.
Let me be blunt: Alchemy fails when the intent is hollow. Block’s intent was noble—decentralize chip production. But the execution lacked the brutal engineering culture needed to beat Bitmain’s supply chain or MicroBT’s manufacturing scale. The $41.9 million termination fee is the market’s verdict. It says: “Your silicon doesn’t justify the risk of being locked into a mining-only asset.”
The ethnographic evidence is clear. Core Scientific’s balance sheet now boasts a $14 billion revenue pipeline from its AMD partnership. That dwarfs any possible return from running Proto miners. The miner has effectively become a REIT for AI compute, not a bitcoin mining company. This is the narrative that investors reward: 140 billion dollars in potential AI revenue over a decade versus uncertain Bitcoin mining cash flows. The stock market has noticed—Core Scientific’s bonds trade at a premium to other miners.
Contrarian Angle
But let me play devil’s advocate for a moment. The conventional take is that Block’s chip was a flop and Core Scientific is a genius for jumping to AI. Yet, what if the chip was actually competitive—just not competitive enough to overcome the narrative gravity of AI hype? The $41.9 million penalty could be seen as a simple IRR calculation: pay a lump sum to avoid inventory risk. Core Scientific might have already found that the cost of integrating Block’s chips into their infrastructure (cooling, power density, maintenance) exceeded the benefits, regardless of chip quality. Moreover, the AI data center boom is itself speculative. If demand for AI compute slows—say, due to a recession or a shift to on-device inference—Core Scientific’s 15-year lease with AMD could become a burden. The same flexibility they cherish now could trap them in long-term contracts with price guarantees.
And what of Block’s chip? It may not be dead. The technology—a 3nm design—has inherent value. Another miner without a strong AI option might buy the inventory at a discount. Or Block could sell the IP to a Chinese competitor. The failure of a singular client does not invalidate the engineering. But in a market driven by narrative, perception is reality. The $41.9 million goodbye sign ensures that Proto will be remembered as a cautionary tale, not a technical marvel.
Takeaway
Bitcoin mining is entering a bifurcation. The survivors will be those who treat their facilities as generic compute hubs, ready to switch from SHA-256 to GPU clusters overnight. The purists who double down on mining-only ASICs will face existential pressure. Block’s Proto chip was a casualty of this transition, but it won’t be the last. The next frontier isn’t hashrate—it’s narrative velocity. When the cost of loyalty exceeds the value of the chip, what binds you to the chain?