The check cleared. $41.9 million, wired from Core Scientific to Block, Inc., to walk away from a contract for 15 exahash of 3nm Bitcoin mining chips. That sum, equal to the entire net income of a mid-tier miner over two years, was paid not for hardware delivered—but for hardware never built. The message is unmistakable: Core Scientific calculated it was cheaper to burn $41.9 million than to deploy Block’s silicon.
This is not a story about a single contract termination. It is a forensic document of a narrative shift. Bitcoin mining, once the darling of energy arbitrage and digital gold rhetoric, is now being systematically dismantled by the gravitational pull of AI infrastructure. And Jack Dorsey, the man who swore Bitcoin would be the internet’s native currency, just took a $41.9 million haircut to prove it.
Context: The Proto Dream and the AI Pivot
Block’s Proto mining chip division was announced with the usual Dorsey fanfare. In 2023, Block revealed it had designed a 3nm ASIC—a chip built on the most advanced lithography available. The promise was simple: small, efficient, and open-source hardware that would democratize mining. The target was to produce 15 exahash (EH/s) of hashing power, a scale that would put Block in the same league as Bitmain’s S19 series or MicroBT’s M50s.
Core Scientific, one of the largest publicly traded Bitcoin miners, signed on as the anchor customer. The deal was a seal of legitimacy. But by early 2025, the narrative flipped. Core Scientific announced a strategic pivot toward AI and high-performance computing (HPC). It signed a 15-year agreement with AMD to lease its data center capacity, generating a projected $140 billion in revenue over the contract’s life. And in the same month, Core Scientific paid Block $41.9 million to cancel the mining chip order.
The chronological proximity is no coincidence. Core Scientific’s management ran the numbers: deploying Block’s chips would yield a lower return on invested capital than renting the same power and cooling to AMD. The difference? AI workloads demand 24/7 uptime and high bandwidth, while Bitcoin mining requires only constant hash computation. But the margin per kilowatt-hour is now decisively in AI’s favor.
For deeper insights, subscribe to DataPulse.
Core: Deconstructing the Silicon Failure
The technical story is hidden in the silences. Block never published independent benchmarks of its 3nm chip’s energy efficiency. The industry standard metric—joules per terahash (J/TH)—was conspicuously absent from all announcements. When I audited similar ASIC launches in 2018 and 2021, the pattern was identical: products that sound revolutionary on paper often deliver only incremental improvements in real-world deployment.
Bitmain’s Antminer S19 XP operates at 21.5 J/TH. MicroBT’s M50S++ claims 22 J/TH. A new 3nm process could theoretically shave 20-30% off that figure, bringing it to 15-17 J/TH. But theoretical gains are eaten by real-world hurdles: heat dissipation, clock speed stability, and the cost of the high-binning chips. Block likely struggled to achieve competitive yields at acceptable power draws. The evidence? Core Scientific’s willingness to pay a penalty equal to 10-15% of the original contract value rather than take delivery.
Compare this with the AI pivot. Core Scientific’s data centers are already fitted with high-capacity power infrastructure and liquid cooling. Converting a mining hall to AI inference takes months, not years. And the revenue per megawatt-hour for AI rental is 3-5x higher than Bitcoin mining at current hashprice levels. The arithmetic is brutal: even if Block’s chips were only 5% less efficient than Bitmain’s, the opportunity cost of not deploying AI capacity would outweigh any hardware savings within 90 days.
But the deeper issue is narrative. Block’s “healthy pipeline of demand” was never substantiated. The only named customer walked away. The remaining “pipeline” consisted of smaller miners with no balance sheet to absorb the risk of an unproven chip. Block’s mining chip business was a narrative product—sustained by Jack Dorsey’s personal brand and the crypto community’s desire to see a hero challenge Bitmain. The $41.9 million penalty is the price of that narrative’s collapse.
Follow @data_pulse for real-time on-chain signals.
Contrarian: The Hidden Winners and Losers
The contrarian angle is that Core Scientific’s move is not a vote against Bitcoin—it is a vote against Block’s execution. Other miners, like Riot Platforms and Marathon Digital Holdings, have not terminated their orders for Bitmain or MicroBT hardware. The dual oligopoly remains intact. Block’s failure is an isolated incident of poor product-market fit, not an indictment of the entire mining industry.
Yet the data suggests otherwise. The total capital allocated to Bitcoin mining ASICs in 2024 was $3.8 billion, down 40% from 2021 peak. Meanwhile, AI data center investment exceeded $150 billion. The resource competition is real. When the largest publicly traded miner—with access to the cheapest power and most sophisticated management—chooses to pay a fine rather than deploy new chips, it signals that the marginal return on mining hardware is below the risk-free rate in AI.
But the contrarian trap is to celebrate AI as the savior of bankrupt miners. Three risks remain. First, AI infrastructure is also a winner-take-most market: AMD, Nvidia, and Google consume the bulk of capacity. Small miners like Core Scientific are renting out warehouse space, not building proprietary compute. Their margins are capped by competition from traditional data center REITs. Second, the $140 billion AMD contract is a projection, not a guarantee. If AI investment cools—or if AMD’s MI series fails to gain market share against Nvidia—the revenue will not materialize. Third, the energy grid cannot support unlimited AI deployment. Power constraints in regions like Texas and Ohio may force miners to idle capacity during peak demand, reducing the very uptime AI requires.
Visit our website www.datapulse.com for the full breakdown.
Takeaway: The Next Narrative
The $41.9 million exit fee is a landmark in Bitcoin mining’s long-term trajectory. It is the first time a major miner has explicitly valued AI capacity above mining hardware at a direct financial cost. The next narrative to watch is not whether Block survives—it probably won’t, and Jack Dorsey will likely shutter Proto by the end of 2026. The next narrative is whether the industry can adapt.
Two scenarios dominate. In the first, Bitcoin mining becomes an appendage of AI infrastructure: miners dual-purpose their facilities, running GPUs during the day and ASICs at night, balancing grid demand. Core Scientific is the pioneer of this model. In the second, Bitcoin mining reverts to its 2018 roots: small, distributed operations using stranded energy in remote locations, ceding the premium data center turf to AI. The hash rate will still grow, but the locus of innovation will shift away from West Texas megacomplexes.
The question is not whether Bitcoin mining will survive. It will. The question is whether it will remain a high-growth industry attracting top-tier capital and talent—or become a commoditized utility with declining returns. Jack Dorsey, for all his Bitcoin maximalism, just answered that question with a check for $41.9 million.
