The Zhitong Finance App learned that Bernstein's latest research report re-estimated the capital expenditure required for CoreWeave to land an order of about 129 billion US dollars, revised the previous incremental capital estimate of about 75 billion US dollars to 44-68 billion US dollars, and narrowed the external financing gap to 180-42 billion US dollars. However, analysts believe that the company's real risk lies in the dilemma of rising financing costs and shortening the term of customer contracts: expansion relies on continuing to “borrow new money,” while contracts supporting financing are getting shorter. Even if AI computing power is in short supply in the short term, once the industry's production capacity is released centrally, CoreWeave may be the first to bear the burden. Maintaining the “outperforming market” rating, the target price was slightly raised to $54.
Model correction: The funding gap has been “slimmed down”, but the pressure to finance 100 billion dollars is still there
Since October 14, CoreWeave's stock price has continued to decline, and the market's core questions about this newcomer AI computing power focus on the source of capital to expand production. Bernstein previously estimated that the company needed an incremental capital of about $75 billion to land and sign an order to be delivered. This report admits that the old model overestimated capital requirements.
The old model is modeled according to the upward trend of a single customer contract and is suitable for demand-side analysis — the scale of production capacity can be closely linked to large customers such as Microsoft and Meta, but directly applied to capital estimates will reduce the implied ratio of revenue to capital expenditure, and include more expenses in the subsequent construction cycle, thereby exaggerating the financing gap.
The installed scale model for this power conversion was re-calculated: Based on the historical unit's installed capital expenditure of about 42 million US dollars/MW, and the benchmark assumption is 40 million US dollars/MW, an additional capital expenditure of 44-68 billion US dollars is required for the implementation of existing contracted reserve orders. After deducting ongoing construction, existing cash, and existing debts and credit, external financing of 180-42 billion US dollars is still required.
If it is further included in the contracted production capacity that will be launched before the end of 2028 and the total scale is close to 4.2 GW, the total additional capital expenditure will rise to US$95-122 billion, and external financing is yet to be raised to US$70-96 billion. The analysis indicates that this additional production capacity will theoretically expand simultaneously with the new contract.
It is worth noting that the estimate is still limited by a serious lack of disclosure of the company's information: the size of megawatts corresponding to the signed order, the annual revenue from the order reserve (RPO only discloses the total caliber for 24 months), and GPU procurement commitments were not disclosed, causing great uncertainty to the calculation.
The real concern: the dilemma between rising financing costs and shortening contract terms
Bernstein's core argument is not whether existing orders can be funded, but rather that both sides of the financing equation are under pressure at the same time. Business expansion requires continuous large-scale external financing, but the term of enterprise-level customer contracts is constantly being shortened, and the contractual basis that supports financing is weakening. As a result, the financing cost corresponding to each additional US dollar of capital continues to rise, creating a dilemma at the strategic level.
In the short term, the current tight supply of AI data centers does support CoreWeave. The company will continue to lose money for the next two years — adjusted EPS forecasts for 2026 and 2027 are -5.14 US dollars and -6.60 US dollars, respectively, and operating profit will not improve significantly until 2027. Despite rapid growth in EBITDA (US$68.26 million in 2026 and US$147.35 million in 2027), high costs such as interest are eroding profits.
Analysts predict that with the release of overall production capacity in the industry and the easing of tight supply situations, CoreWeave will become one of the first and hardest hit companies. In terms of valuation, Bernstein maintained a 23.5 times forward EV/EBIT ratio, corresponding to a target price of 54 US dollars (a slight increase from the previous price of 51 US dollars), but it is still lower than the current stock price, implying a downside of about 2%.
Rating and risk: The bearish position has not changed, and we need to be wary of upward factors
Despite the narrowing of the funding gap estimates, Bernstein still reiterated the “outperforming market” rating. The rating standard is that it outperforms the S&P 500 by more than 15 percentage points over the next 12 months. The report also suggests three upside risks: the delay of Yunda Factory's self-built data center, which favors third-party computing power services; the CoreWeave self-developed software platform maintains the stickiness of customer demand; and the continued shortage of electricity in the data center supports its pricing power.
As of October 26, 2026, CRWV closed at $55.19, with a market capitalization of US$30.439 billion. The cumulative decline during the year was 22.9%. It plummeted 58.4% in the past 12 months, significantly outperforming the S&P 500. Whether this “AI shovel seller” financing narrative can be realized under the intertwining of AI computing power capital expenditure frenzy and supply shortages will remain the focus of the ongoing market game.