①Goldman Sachs upgraded CoreWeave's rating from "Neutral" to "Buy," though the price target was lowered to $106, still leaving about 73% upside from last Friday's closing price; ②The firm believes the market may be underestimating the improvement in contract returns and overestimating the impact of rising financing costs; ③Previously, institutions including Jefferies, Bank of America, Deutsche Bank, and JPMorgan had successively reaffirmed their bullish stance.
Cailian Press, October 19 (Editor: Shi Zhengcheng): CoreWeave, the "new cloud" service provider that has recently become a barometer of AI trading sentiment, received a bullish endorsement from Goldman Sachs on Monday.
Analysts, including Gabriela Borges, upgraded CoreWeave's rating from "Neutral" to "Buy." However, Goldman Sachs simultaneously lowered the price target from $139 to $106. Compared with last Friday's closing price of $61.39, the new target still implies about a 73% upside.
Notably, following CoreWeave's "double whammy" of stock and bond sell-offs last week, institutions including Jefferies, Bank of America, Deutsche Bank, and JPMorgan have successively reaffirmed their bullish stance. Among them, JPMorgan believes that the related concerns are overblown, while Deutsche Bank views the recent decline as a buying opportunity.
What does Goldman Sachs say?
In this research report, Goldman Sachs' core view is that investors may have underestimated the extent of the improvement in CoreWeave's project returns while overestimating the impact of rising financing costs.
Goldman Sachs believes that the main concerns facing CoreWeave center on the funding cycle of its AI infrastructure expansion: the company must continuously raise capital to purchase GPUs and deploy computing power, then rely on fees paid by customers to recoup its investments. If financing costs rise or if customers fail to generate sufficient revenue from their AI services, this business model could come under pressure.
Therefore, the key to CoreWeave's enterprise value is directly related to the magnitude and sustainability of the gap between its unlevered internal rate of return (uIRR) and its weighted average cost of capital (WACC).
Simply put, the key lies in whether the capital invested in computing‑power projects can consistently generate returns that exceed the cost of capital. Goldman Sachs believes that while the market has recently focused heavily on financing pressures, it may be underestimating the positive impact of computing‑power pricing, software revenues, and equipment renewal leases on project returns.
The sharp rise in bond yields does not mean that all forms of financing are equally costly.
First, Goldman Sachs does not deny that CoreWeave's financing costs are rising, but it argues that the trading price of unsecured bonds cannot directly reflect the company's overall new financing costs.
According to a Goldman Sachs research report, the worst-case yield on CoreWeave's unsecured bonds has reached 13.5%–15%. However, the company's expansion of its AI infrastructure primarily relies on deferred-drawdown term loans (DDTL), which are backed by specific customer contracts, GPU equipment, and related cash flows.
In such transactions, lenders not only assess CoreWeave's own creditworthiness but also evaluate the credit quality of end customers, contractual revenues, and the collateral value of the equipment. Consequently, even if the unsecured bond market demands a relatively high risk premium from the company, some projects may still secure financing at comparatively lower costs.
Goldman Sachs estimates that CoreWeave's current weighted average cost of debt is around 8%–9%, and it may rise by 1–2 percentage points in the future, though the increase will likely not exceed this range. The firm has adopted a 10% cost of debt assumption in its model and adjusted its interest expense forecasts accordingly.
This means that while higher financing costs do erode returns, the impact may not be as severe as suggested by the yields on unsecured bonds.
With computing power prices rising, can they outpace the cost of capital?
As financing costs rise, Goldman Sachs believes that CoreWeave's ability to generate revenue per unit of computing power is also improving.
The report estimates that the company's current annualized recurring revenue per megawatt is approximately $10 million. According to company disclosures cited in the research note, some short-term contracts with maturities of 3 to 6 months generate annualized revenues as high as $40 million per megawatt, and the company raised prices by 25% in July and by an additional 10% in October.
However, CoreWeave's average contract term is about 4–5 years, so price increases will not immediately be reflected in all contracts. Goldman Sachs also did not simply apply the higher prices of short-term contracts to its long-term forecasts; instead, based on discussions with customers and competitors, it set $13 million per megawatt as the initial assumption for the annualized revenue of new long-term contracts.
In the example model, which disregards the residual value of the GPUs at the end of the five-year contract, annualized revenue per megawatt reaches $13 million, corresponding to a post-tax unlevered internal rate of return of 15.3%; if revenue rises to $15 million, the IRR increases to 20.9%.
By contrast, the weighted average cost of capital under the 10% debt‑cost assumption is 11.8%.
This is precisely the core of Goldman Sachs' bullish view: if new contracts can generate higher revenue per unit of computing power, project returns may still significantly exceed the cost of capital, and rising financing costs do not necessarily render expansion economically unviable.
This assessment is based on the premise that supply and demand for computing power will remain tight. Goldman Sachs expects that, although there is still considerable uncertainty in forecasting events as far ahead as 2030, the supply-demand imbalance is more likely to worsen than to ease over the next 24 months, with companies' deployment of AI applications in actual production serving as a key driver.
At the same time, long-term contracts have another dimension: while price increases are passed through to overall revenue more slowly, the fixed contract prices help cushion the impact of future declines in spot‑market computing power prices.
It's not just GPU rentals—software is also a source of revenue.
Goldman Sachs also emphasized that there are significant differences among emerging GPU cloud service providers.
According to the bank's industry interviews, CoreWeave's cost-per-unit performance is approximately 20% higher than that of its peers, with advantages stemming from its infrastructure software, cluster management tools, and operational expertise accumulated through large-scale AI workloads.
The company is also expanding into a more comprehensive AI platform through products such as Forge and Omni. Specifically, Forge integrates training, inference, evaluation, and agent development; while Omni enables customers to run their software stacks on infrastructure that is not owned by CoreWeave.
These businesses carry dual implications: on the one hand, software and platform services can drive revenue growth without necessarily requiring a proportional increase in hardware capital expenditures, thereby enhancing the return on deployed computing capacity; on the other hand, deeper integration into customers' development workflows helps strengthen customer stickiness.
Goldman Sachs acknowledged that, as supply increases, the business of simply renting out GPUs may face homogeneous competition. Whether platformization can make progress will affect CoreWeave's ability to maintain a competitive edge once supply and demand return to equilibrium.
Renewal of the legacy GPU lease could generate additional cash flow.
Regarding concerns about asset devaluation triggered by the rapid iteration of GPUs, Goldman Sachs has also offered another perspective.
The report notes that the A100, launched in 2020, has been renewed under a three-year contract extending to 2029, while the H100, introduced in 2023, has also been renewed for another three years at a price higher than the original contract.
This suggests that, for some customers, upgrading to the next‑generation GPU is not always the most cost‑effective choice. Migrating existing infrastructure and workflows entails costs and may disrupt business operations, and the performance gains of new hardware may not be sufficient to offset these expenses.
Goldman Sachs did not attribute all of the aforementioned phenomena to the enduring competitiveness of older GPUs, and it also acknowledged that the current shortage of computing power may have extended the economic lifespan of legacy equipment. However, if such equipment can still be leased after the initial contract expires, it could generate returns beyond those projected in the baseline calculations.
Under scenarios assuming annualized revenue of USD 13–15 million per megawatt and incorporating a sixth-year residual value, the after-tax unlevered internal rate of return can rise to approximately 20%–25%.
Meanwhile, some project‑related loans will be gradually repaid as the initial contracts are executed. Once the equipment has been deployed and the associated debt has been substantially reduced, lease‑renewal income can generate cash flows with greater profit margins, providing internal funding for subsequent expansion.
The upgrade in rating does not mean that long-term concerns have been dispelled.
Goldman Sachs has upgraded its rating from "Neutral" to "Buy," while lowering its price target from $139 to $106 and reducing its forward EBITDA multiple from 22x to 18x.
This adjustment reflects the bank's recognition of buying opportunities following the sharp decline in its share price, while still applying a valuation discount to account for risks such as high levels of debt.
Goldman Sachs points out that the greater long-term risk lies in the possibility of a sharp decline in computing power prices once supply and demand return to equilibrium, which could render project returns at renewal during years 4–6 insufficient to cover capital costs, while the company would still be bound by data center leases spanning years 7–16. Furthermore, some large customers may prioritize using their own in-house computing capacity, thereby affecting contract renewals.
Nevertheless, Goldman Sachs still believes that data on contract economics over the next 12 months may exceed market expectations, and thus views the recent correction as a buying opportunity.