Core Scientific paid $41.9 million to walk away from Block's 3nm mining chips. That's not a negotiation tactic—it's a diagnosis. Over the past year, the largest publicly traded Bitcoin miner didn't just cancel a contract; it executed a strategic pivot that signals a deeper structural shift in how capital allocates to proof-of-work infrastructure. The write-off—covering prepayments and termination fees—was announced in a 2026 filing. But the real data point isn't the dollar amount. It's the fact that Core chose to burn $41.9M rather than deploy those chips.
Block's Proto mining chip was supposed to be Jack Dorsey's challenge to Bitmain's dominance. A 3nm ASIC targeting 15 exahash per second. The company claimed "healthy pipeline demand" just months before the cancellation. But that pipeline evaporated when its only public customer—Core Scientific—decided the chips weren't worth running. Core didn't just cancel; it simultaneously signed a 15-year contract to lease its data centers to AMD for AI workloads. The message is clear: mining Bitcoin has become a lower-return business than renting GPU racks.
The core technical question is simple: why did Block's chip fail? Having audited mining hardware for institutional clients during the 2024 ETF wave, I can tell you that the difference between a winning and losing ASIC often comes down to a few percentage points in energy efficiency. Block never published its J/TH (joules per terahash) ratio. Without that metric, any claim of performance is marketing, not engineering. Bitmain's Antminer S21 series operates at around 17.5 J/TH. MicroBT's M50 series competes in the same range. If Block's chip couldn't beat those numbers—or even match them—the economics collapse. Math doesn't negotiate. A chip that consumes 5% more power per hash than the competition doesn't just lose on margins; it loses on capital efficiency. In a post-halving world where mining rewards are halved, every watt matters.
But the failure isn't just technical. It's structural. Core Scientific's decision to pivot to AI isn't an anomaly—it's a canary. The company will likely generate $14 billion in revenue from the AMD contract over 15 years. That's a guaranteed stream, not a volatile mining payout. By contrast, mining Bitcoin in 2026 offers thin margins after the 2024 halving, rising difficulty, and stagnant price. Privacy is a feature, not a bug. But in this case, the privacy of Block's chip specifications became a bug: without verifiable third-party benchmarks, Core couldn't risk deploying hardware that might underperform.
The contrarian angle here is that the 'AI saves miners' narrative has its own fragility. Core's AMD deal sounds transformational, but the AI data center market is equally competitive. Hyperscalers like Microsoft and Google are building their own infrastructure. A downturn in AI investment—or a shift to more efficient inference chips—could leave Core with empty racks. The real story isn't about AI versus mining. It's about the single-customer dependency that killed Block's chip. Core put all its eggs in one basket with Block, then paid dearly to get out. Now it's putting all its eggs in AMD's basket. That's not diversification; it's swapping one concentrated risk for another.
Moreover, Block's failure is a case study in the 'visionary founder' trap. Jack Dorsey's reputation couldn't engineer a better chip. His company's crypto portfolio—Tidal (music), TBD (decentralized identity), Bitchat (messaging), and Bitkey (self-custody wallet)—has been a graveyard of failed bets. Block shares are down 68% over five years. Code is law, but bugs are reality. In hardware, vision doesn't matter if the fabrication yield is low or the power efficiency is mediocre. The semiconductor industry is unforgiving: it punishes late entrants and rewards incumbents with decades of process optimization.
The takeaway is forward-looking. The next question isn't whether Block will survive in mining—it likely won't. The question is: how many more Core Scientifics are out there, ready to pay millions to escape their mining commitments? If the trend continues, Bitcoin's hashrate growth may plateau. The network's security—which relies on miner economic incentives—could become a concern. I've seen this pattern before during the 2021 LUNA crash, where a single protocol's failure cascaded through the ecosystem. Here, the cascade is slower but real: capital leaving mining for AI dries up the demand for new ASICs, making it harder for even Bitmain to justify next-generation chips. Watch for more termination clauses and pivot announcements. The era of easy mining profits is over. The next era belongs to those who can compute, not just mine.