Bernstein cautioned investors on Wednesday that Core Scientific's widely cited 75% return on assets from its AI hosting contract is a structural outlier produced by post-bankruptcy accounting, not a realistic target for bitcoin miners now entering the artificial intelligence infrastructure market. The firm puts stabilized ROA at 5% for TeraWulf (WULF) and 4% for Cipher Mining (CIFR), figures it considers far more representative of what new entrants should expect.

The distinction matters because the broader mining industry is repricing itself around AI. Approximately 17 infrastructure deals worth more than $110 billion have been struck since late 2024, contracting roughly 6 gigawatts of power capacity to hyperscalers including Google Cloud, Amazon Web Services, Microsoft, Nvidia, and CoreWeave. Bernstein projects that aggregate AI revenue across its bitcoin miner coverage universe will grow from $1.2 billion in 2026 to $10.7 billion by 2030. Bernstein's core argument is that this growth story is real, but the ROA numbers attached to Core Scientific's entry point are not transferable to miners entering the market today.

Why Core Scientific Is Not a Template

Core Scientific emerged from Chapter 11 bankruptcy in January 2023, which reset its asset base to fair market value at the cycle trough. That reset produced an artificially low capital denominator in its ROA calculation. When the company subsequently contracted 590 megawatts of capacity to CoreWeave, it was pricing AI hosting against a deflated cost basis. Bernstein had previously estimated 75 to 80 percent gross margins on that deal for exactly this reason. Competitors building AI infrastructure today are doing so at current construction costs of $8 million to $11 million per megawatt, financed through hyperscaler-backed project debt at roughly 7.125 percent. Sound financing, but fundamentally different ROA math.

CoreWeave announced an acquisition of Core Scientific at approximately $9 billion in July 2025. Core Scientific shareholders voted it down, and the deal was terminated on October 30, 2025. Core Scientific (CORZ) continues to trade independently on Nasdaq and has committed to converting its remaining bitcoin mining facilities into high-density colocation computing operations. Its 75% ROA figure therefore reflects an independent company still sitting on bankruptcy-reset assets, not an acquired entity or a peer-comparable balance sheet.

In a prior initiation, Bernstein assigned Outperform ratings and price targets of $36 and $32 to TeraWulf and Cipher Mining respectively, framing both as "power landlords of AI," meaning grid-connected operators offering stable, long-duration revenue to hyperscalers. The July 15 note builds on that prior initiation, applying the ROA framework to show why Core Scientific's returns are not a useful benchmark for competitors entering the market now. TeraWulf's contracted HPC pipeline exceeds $12 billion in total value. In Q1 2026, HPC revenue accounted for $21 million of TeraWulf's $34 million in quarterly revenue, crossing above bitcoin mining income for the first time. Bernstein projects mature EBITDA margins of roughly 84% for TeraWulf and approximately 93% for Cipher Mining, driven by stable, long-duration revenue once construction is complete.

Capital markets have already priced in the AI pivot. Miners with secured HPC contracts now trade at 12.3 times next-twelve-month sales versus 5.9 times for pure-play bitcoin miners. TeraWulf was up 122% year to date through June 2026; Cipher Mining gained 69% over the same period. Morgan Stanley and Jefferies have also initiated coverage, signaling broad institutional conviction in the thesis even at compressed ROA expectations.

The Regional Warning

The Bernstein note carries particular weight outside the United States. Pakistan has committed 2,000 megawatts of surplus electricity to bitcoin mining and AI data center operations, the largest government-directed crypto energy allocation in South Asia to date. Finance Minister Muhammad Aurangzeb described it as "a pivotal moment in Pakistan's digital transformation," framing the initiative as a means of turning excess energy into innovation and revenue. Bilal bin Saqib, CEO of the Pakistan Crypto Council, added that the program "could enable the government to generate foreign exchange in USD through bitcoin mining."

India presents a contrasting regional picture. The country is positioning itself as a hyperscale AI data center destination, with grid investment cycles more agile than those of the United States, and Indian utility and infrastructure firms are widely seen as primary beneficiaries of a structural AI power demand shift.

Those ambitions across the region are plausible. The economics require calibration. If Pakistan or regional investors are modeling returns against Core Scientific's 75% figure, the Bernstein analysis is a direct correction. The 4 to 5% stabilized ROA Bernstein assigns to the best-positioned US-listed operators, specifically companies with premium grid access, hyperscaler contracts already signed, and institutional financing in place, should be treated as a ceiling for emerging market deals, not a floor. Political risk, currency exposure, and longer construction timelines compress returns further in South Asia and Africa.

Ethiopia currently contributes roughly 2.5% of global Bitcoin hashrate, part of an African total that accounts for approximately 3% of global mining activity. The country draws around 600 megawatts at 3.2 cents per kilowatt-hour from Grand Ethiopian Renaissance Dam hydropower. South Africa's state utility Eskom is exploring sales of midday solar surplus to miners and data centers. The regulatory environment across the continent remains deeply fragmented: Kenya and Ghana are advancing digital asset frameworks, Nigeria's regulatory resistance has driven much of its mining activity underground, and Ethiopia has yet to produce a clear long-term digital asset policy. For operators in all these markets, the "power landlord" model only works with stable, long-term offtake counterparties. No hyperscaler has established operational-scale infrastructure in sub-Saharan Africa yet.

Network Security Dimension

The shift from mining to AI hosting is not only a financial story; it carries direct implications for Bitcoin's proof-of-work security model. Global Bitcoin hashrate declined from 1,066 exahashes per second in Q1 2026 to 1,004 EH/s in Q2 2026 as miners redirect facilities. Core Scientific alone sold $175 million in BTC in March 2026 to fund its transition, contributing to selling pressure during the period. For developers and users relying on proof-of-work security, the incentive structure of the mining base is shifting in ways that warrant monitoring, even as the AI revenue thesis continues to attract capital into the sector.