-+ 0.00%
-+ 0.00%
-+ 0.00%

Who did the AI infrastructure panic kill by mistake? Bernstein is optimistic about Riot (RIOT.US) and TeraWulf (WULF.US) IREN (IREN.US) power asset values or are undervalued

Zhitongcaijing·11/16/2026 15:25:04
Listen to the news

The Zhitong Finance App learned that with the recent sharp sell-off of stocks related to artificial intelligence (AI) infrastructure, Bernstein believes that the market may have excessively amplified the financing and operating risks faced by emerging AI cloud service providers, while ignoring the scarce power assets and long-term contract values owned by some infrastructure developers. The bank is particularly optimistic about investment opportunities in Riot Platforms (RIOT.US) and TeraWulf (WULF.US), and believes that IREN (IREN.US), which has recently experienced the most severe stock price decline, also has asset value that cannot be ignored.

Bernstein analyst Gautam Chhugani and his team pointed out in a research report released on November 16 that in the past 24 months, the emerging AI infrastructure companies covered by the bank have signed AI-related contracts worth a total of 132 billion US dollars, corresponding to about 5.5 gigawatts (GW) of electricity capacity. However, in the past two months, the median share price decline of these companies reached about 35%.

Bernstein pointed out that this round of sell-off initially stemmed from market concerns that data center construction might face more regulatory restrictions after the US midterm elections. Subsequently, NVDA.US (NVDA.US) management changes further heightened investors' concerns about the AI infrastructure industry.

Investors worry that after Hwang In-hoon leaves office, Nvidia's new management may no longer actively provide credit guarantees and strategic support to emerging AI cloud service providers as before, thereby weakening the financing capacity of these companies and affecting their data center expansion plans.

However, Bernstein believes that the market does not fully distinguish between emerging AI cloud service providers and data center infrastructure developers with large power assets. The latter can not only be commercialized by operating AI cloud services, but can also monetize value by leasing data centers and selling assets that already have access to electricity.

In a context where electricity supply is increasingly becoming a bottleneck in AI infrastructure expansion, the multi-gigawatt power assets owned by these companies themselves have important strategic value.

Bernstein's research shows that of the $132 billion AI contracts mentioned above, about 85% were data center hosting and leasing contracts, and the rest were AI cloud computing service contracts. In terms of customer composition, AI Labs is the largest customer group, accounting for about 42% of the total contract value; emerging AI cloud service providers account for 23%, large cloud service providers account for 18%, chip suppliers account for 11%, and the remaining 6% come from enterprises and other customers.

However, the bank emphasized that it is not accurate to determine contract risk based solely on the identity of the end customer. Although some AI laboratories or emerging cloud service providers have limited credit qualifications, their leasing contracts have obtained credit guarantees from large technology companies.

Therefore, when evaluating the financial risk of AI infrastructure companies, it is more important to determine who ultimately bears the payment obligation and whether the relevant contracts are supported by investment-grade credit-qualified companies.

For example, Terawulf currently has approximately $27 billion in AI infrastructure contracts, of which approximately 96% of the contract value is backed by Google (GOOGL.US) credit. The company's Energize Data data center project in Kentucky has signed a $19 billion contract with an AI lab, corresponding to 401 megawatts (MW) of power capacity, and has received a credit guarantee from Google. Additionally, Terawulf's data center in Lake Mariner, New York signed a $7.2 billion contract with FluidStack, an emerging AI cloud service provider, which also received credit support from Google.

Cipher Digital (CIFR.US)'s contract structure is also relatively robust. The company currently has an AI infrastructure contract of about $16.5 billion, of which about 55% comes from Fluidstack, which has received Google's credit guarantee, and the remaining 45% comes from AWS, the cloud computing division of Amazon (AMZN.US).

Cipher's Barber Lake project in Texas signed a $9.1 billion contract with Fluidstack, and Google provided about $1.7 billion in credit guarantees. The total value of the two contracts signed by the company and AWS is about US$7.5 billion, and Amazon guarantees related basic rent and operating expenses.

Core Scientific (CORZ.US) currently has about $24 billion in data center hosting contracts, of which about 42% comes from CoreWeave (CRWV.US), about 42% from AMD (AMD.US), and the remaining 17% from emerging AI cloud service providers that have received AMD credit support.

CleanSpark (CLSK.US) has signed a data center contract worth approximately US$6.6 billion with Meta Platforms (META.US), involving 175 megawatts of data center capacity in Georgia, and Meta guarantees related rent and operating expenses.

Bernstein believes that the above contract structure shows that some emerging AI infrastructure companies do not fully rely on Nvidia's credit support. Major technology companies such as Google, Amazon, Microsoft (MSFT.US), Meta, and AMD are also providing demand and credit support for the expansion of the industry.

From a valuation perspective, Bernstein pointed out that after recent drastic adjustments, the ratio of enterprise value to planned power capacity of emerging AI infrastructure developers has dropped to an average of about US$2.8 million per megawatt, down about 35% from the second quarter of this year.

At the same time, if a capitalization rate of 8% to 9% is used to value future net operating income corresponding to the signed contract and take into account the additional debt required for the construction of related projects, the median discount of the enterprise value of these companies compared to the implied value of the contract is currently about 26%.

Bernstein believes that in the current market environment where electricity resources are scarce, some companies can monetize value by selling data center assets with power access conditions even if they do not further expand their AI cloud computing business. Therefore, investors should not value these companies based solely on the operating risks of emerging AI cloud service providers.

In terms of specific investment targets, Bernstein paid particular attention to Riot Platforms and TeraWolf. The two companies have AI data center contracts of approximately US$38 billion and US$27 billion respectively, ranking among the top two companies covered by the bank, with a contract period of 10 to 20 years.

Among them, Riot Platforms' valuation discount was the most significant. The company has now signed a number of data center contracts with AI Labs and AMD, including a project in Corsicana, Texas involving 910 megawatts of power capacity worth $28 billion, and a project in Rockdale involving 191 megawatts of power capacity worth $9.1 billion.

Bernstein estimates that the annual profit before interest, tax, depreciation and amortization (EBITDA) corresponding to the contract that Riot Platforms has signed is approximately US$2.2 billion. Based on a capitalization rate of 8.5% and taking into account the estimated additional debt of about US$6.55 billion required for future project construction, the company's current adjusted enterprise value is discounted by about 50% compared to the implied value of the contract, and the ratio of the enterprise value to the planned power capacity is approximately US$3.3 million per megawatt.

However, Bernstein also pointed out that currently about 98% of Riot Platforms' contract value comes from AI laboratories. Investment-grade credit guarantees for some contracts are still being implemented, and related project financing details need to be further disclosed.

Therefore, whether it is possible to obtain credit support from large technology companies and whether data center construction financing can be completed at an attractive cost will be an important factor affecting Riot Platforms' valuation repair.

In contrast, TeraWolf has a more defined credit support structure. Bernstein estimates that the annual EBITDA corresponding to the contract the company has signed is approximately US$1.05 billion. The valuation is based on the same 8.5% capitalization rate, and the additional debt required for future project construction is taken into account. The current adjusted corporate value is discounted by about 22% compared to the implied value of the contract.

TeraWolf's current enterprise value ratio to planned power capacity is approximately $2.6 million per megawatt. Additionally, the company is actively seeking customers for the Muskie data center campus with 1 GW of power capacity in Kentucky, providing potential opportunities to further expand the contract size in the future.

Bernstein gave Riot Platforms and TeraWulf “outperform the market”. The target prices were $30 and $20, respectively. Based on the closing price on November 13, the potential upside for the two stocks was about 71% and 58%, respectively.

Among the companies covered by the bank, IREN is the company with the biggest drop in stock prices in the past two months, with a cumulative decline of about 58%. Unlike infrastructure developers that mainly generate revenue through data center leasing, IREN directly operates the AI cloud computing business, so it not only faces data center construction and financing risks, but also bears risks such as GPU procurement, equipment deployment, and fluctuations in computing power rental prices.

Bernstein pointed out that investors are currently mainly concerned about whether IREN can successfully implement complex AI cloud infrastructure construction plans, especially when delivery of part of its computing power capacity to Microsoft has already been delayed. Furthermore, most of the computing power capacity that the company expects to be put into operation in 2027 has not yet been locked in customer contracts, putting it directly at risk of fluctuating GPU computing power rental prices.

However, IREN currently has approximately $20 billion in AI cloud service contracts, of which approximately 65% of the contract value comes from customers with investment-grade credit qualifications.

Microsoft is IREN's largest customer, with contracts worth around US$9.7 billion, accounting for 48% of its total contract value; Nvidia's related contracts worth around US$3.4 billion, accounting for 17%; and the remaining 35% from AI Labs and other enterprise customers. IREN also issued $2 billion worth of equity subscription warrants to Nvidia, linked to the milestone of installing 600,000 GPUs.

Bernstein believes that although IREN has established a close partnership with Nvidia, the company's core advantage from other emerging AI cloud service providers is that it has a power asset portfolio of up to 5.8 gigawatts, of which more than 90% of the capacity has yet to sign relevant customer contracts.

This large portfolio of power assets gives IREN greater strategic flexibility. The company can adjust the pace of AI cloud computing business expansion according to market demand and financing environment, and can also monetize assets by selling power-related assets, developing data center hosting services, or establishing partnerships with technology companies other than Nvidia.

Bernstein believes that with IREN's stock price falling sharply, investors need to re-examine the potential value of their electricity assets rather than just focusing on the short-term operating risks faced by the AI cloud computing business. The bank currently gives IREN an “outperforming the market” rating, with a target price of $55. Based on the stock's closing price of $18.17 on November 13, the potential for growth is over 200%.

Additionally, Bernstein also gave Cipher Digital, Core Scientific, and CleanSpark “outperform the market” ratings, with target prices of $15, $19, and $11, respectively.

However, the bank also warned that the AI infrastructure business is still a capital-intensive industry. Low-cost financing, access to land and power resources, equipment supply, and the ability to deliver data centers on schedule will still be key factors in determining whether related companies can turn long-term contracts into actual profits.