Bernstein: CoreWeave Rated Underperform, IREN Rated Outperform, With Price Targets of $74 and $100 Respectively
nashnova research
Bernstein used UK AI-cloud firm Nscale's S1 filing as a catalyst to publish a comp report, reiterating CoreWeave at Underperform and IREN at Outperform — the core split is who gets hit first when GPU supply loosens.
Why is CoreWeave rated Underperform?
Bernstein keeps an Underperform rating on CoreWeave with a $74 price target, valued at 25.5× EV/EBIT (enterprise value to earnings before interest and taxes).
The logic: tight data-center supply benefits CoreWeave today, but as industry GPU capacity loosens, it will be among the first and hardest hit.
This means → CoreWeave's growth story rests on a "demand exceeds supply" premise; once that premise softens, the valuation math resets.
Bernstein flags that CoreWeave's ability to retain client stickiness through a looser supply cycle is the key test of this bearish call.
Why does Bernstein prefer self-built capacity over leasing?
Bernstein explicitly favors the self-built capacity model. IREN and Nscale build their own sites; CoreWeave relies heavily on leased facilities.
Self-build advantage: better control over project economics and residual asset value — what the equipment is still worth after its useful life.
Self-build trade-off: higher capex upfront and greater execution risk.
In plain terms = leasing is flexible but you own nothing at the end; building costs more but the long-run math favors it — Bernstein is betting on the latter.
What does Nscale's IPO filing tell us about the competitive landscape?
Nscale's current live capacity is just 55 MW, far below CoreWeave's 360 MW at its own IPO — a significant scale gap.
Yet 76% of Nscale's pipeline under construction is self-built, aligning with IREN's approach; leased sites account for 31% of existing capacity.
Geographically, Nscale is international-first, positioning it for sovereign-client contracts and access to cheaper overseas power; CoreWeave and IREN are U.S.-centric.
Bernstein states plainly: Nscale's S1 filing did not change its valuation framework for any covered name.
Why are crypto miners collectively rated Outperform for AI infrastructure?
Miners hold a planned power reserve of roughly 32 GW and can rapidly deliver energized, ready-to-go shell facilities.
This means → one of AI's biggest bottlenecks is "time to capacity" — whoever can hand over a powered facility fastest wins, and miners are built for exactly that.
Over the past two years, miners have signed contracts to deliver about 9 GW of power capacity across more than 20 deals worth over $180 billion in total.
What are the ratings and price targets across the sector?
IREN: Outperform, price target $100.
TeraWulf: Outperform, $36; Cipher Digital: Outperform, $32.
Core Scientific: Outperform, $32; Riot Platforms: Outperform, $35; CleanSpark: Outperform, $24.
MARA Holdings: Market Perform, $17 — the only name in the group not rated Outperform.
市场有风险，内容仅供研究参考，不构成投资建议。