Bernstein: Texas Grid's 8 GW Baseload Allocation Fully Assigned, Miners' Competitive Position in AI Infrastructure Transition Strengthened
nashnova research
ERCOT's first screening carved out 8 GW of baseload and 7 GW of study load for AI hopefuls — approved power is now a scarce asset, and crypto miners holding baseload slots are turning into AI infrastructure landlords.
What did the "Batch Zero" screening sort out?
Texas grid operator ERCOT completed its first Batch Zero screening on September 3, sorting applications into three tiers: baseload (mature projects, ~8 GW), study load (pending further analysis, ~7 GW), and excluded projects (deferred to later batches).
This means → baseload is "power you can count on"; study load is "we haven't decided how much you get"; excluded projects won't even re-apply until summer 2027.
The formal verification report is due December 2026; study-load allocations won't be finalized before April 2027 at the earliest.
How much power did each miner get — and who lost out?
IREN was the biggest winner: its Sweetwater campus qualified as baseload with the full 2 GW — the largest single approval in this round.
Hut 8's Beacon Point (1 GW), Galaxy Digital's Helios Phase I (800 MW) and Phase II (830 MW) all made baseload; Core Scientific secured 731 MW, CleanSpark 585 MW, and MARA roughly 480 MW — but MARA's 2 GW Matagorda site was downgraded to study load.
Cipher Digital was effectively demoted: it submitted 3.6 GW for Batch Zero but received only 1.1 GW of baseload (1 GW Colchis + 100 MW Stingray), while 2.1 GW (including 900 MW Apollo) landed in study load.
In plain terms = same application round, vastly different outcomes — the gap comes down to project maturity and prior development history.
Why is approved power becoming more valuable?
Bernstein estimates that of the 8 GW baseload, more than 5 GW is available for new AI deployment contracts — including IREN's 2 GW campus and Cipher's 1 GW Colchis.
New data-center development faces mounting political scrutiny, local moratoriums, and state-level directives — all creating a capacity bottleneck that keeps raising the value of already-approved power.
This means → the grid rules function as a "de-clogging" mechanism — they push speculative queue-sitters out and make sites with real development history more valuable. The scarcer approved capacity gets, the stronger the lease-pricing power.
Why can miners play "landlord" for AI infrastructure?
Crypto miners hold roughly 32 GW in planned power portfolios and can deliver "warm shells" — powered, deployment-ready data-center enclosures — faster than most competitors, directly solving AI's "time-to-compute" pain point.
Over the past two years, miners have signed approximately 9 GW across 20-plus deals worth over $180 billion with hyperscalers, frontier AI labs, emerging cloud providers, and AI chip makers.
This reflects Bernstein's core thesis: no matter how the AI model layer shakes out, power is the hardest constraint — miners with baseload approvals are shifting from compute providers to infrastructure landlords, and the scarcity of that "entry ticket" determines their lease-pricing power.
What are Bernstein's ratings and price targets?
Outperform ratings maintained on: Cipher Digital (target $32), TeraWulf ($36), IREN ($100), Core Scientific ($32), CleanSpark ($35), Riot ($24).
MARA rated Market Perform, target $17.
In plain terms = Bernstein's bull case isn't "who mines the most crypto" — it's who holds the most power and the most mature sites, because those determine who collects rent during the AI compute shortage.
市场有风险，内容仅供研究参考，不构成投资建议。