The Coming Crisis in AI Infrastructure: What About USD.AI?

1. Expansion of AI Infrastructure Investment and the Rise of Credit Risks

The artificial intelligence (AI) infrastructure market has been rapidly growing based on explosive computing demand. One of the companies that best exemplifies this growth is CoreWeave. CoreWeave is a cloud service provider that has secured NVIDIA GPUs on a large scale and provides computing services necessary for model training and inference to AI companies. Unlike large tech companies like Microsoft and Amazon that generate revenue from various businesses, CoreWeave focuses solely on the AI computing infrastructure business.

CoreWeave's growth rate has been extremely steep. Its annual revenue increased from $16 million in 2022 to $5.131 billion in 2025, a nearly 320-fold increase in just three years. As of the end of June 2026, the expected revenue backlog from future customer contracts is estimated to reach approximately $104 billion. By aggressively securing GPUs and data centers in line with the increasing AI demand, CoreWeave has rapidly scaled its business. CoreWeave's annual revenue. Source=TIRK

The foundation of this growth has been large-scale external funding. CoreWeave must secure GPU and data center infrastructure before providing computing services to customers. Investment costs are incurred upfront, but revenue is recovered over a long period as the infrastructure is operated and services are provided. Therefore, the faster the business expands, the more funding is required. In fact, CoreWeave recorded a net cash outflow of approximately $14.9 billion from investment activities in the first half of 2026, and its debt on the balance sheet at the end of June was about $35.1 billion.

The problem is that the financial costs associated with this expansion are also rising rapidly. CoreWeave's revenue in the second quarter of 2026 was $2.575 billion, more than double that of the same period last year, but its net interest expenses reached $640 million. During the same period, the net loss was recorded at $626 million. Of course, adjusted EBITDA was a profit of $1.51 billion, indicating that the business is not entirely incapable of generating cash. However, considering the massive capital investment and borrowing costs, it becomes crucial whether future revenues can be converted into actual cash to sufficiently recover investments and debts.

Recently, the financial market has begun to reassess these risks in earnest. In mid-October, CoreWeave's stock price and corporate bond prices fell simultaneously. In particular, the market yield of corporate bonds issued in April at about 9.7% rose to around 15% by mid-October. As bond prices fell, investors began to demand higher yields for the company's credit risk. This does not mean that actual defaults have occurred, but it shows that the financial market's assessment, which has supported the expansion of AI infrastructure, is changing.

The key point is that the growth of AI computing demand does not necessarily align with the profitability of AI infrastructure investments. While CoreWeave has rapidly grown by securing large customer contracts, it has also accumulated significant debt and interest burdens in the process. Moving forward, what matters is not how many GPUs and customer contracts have been secured, but whether actual revenues and cash flows can sufficiently support the funds needed for debt repayment and additional investments.

2. Democratic Victory and Strengthened Data Center Regulations: The Wall Faced by AI Infrastructure Companies

Recently, AI infrastructure companies that have relied on large-scale borrowing for growth have faced another burden: the tightening of regulations on data center construction. Even if they invest massive funds to secure GPUs and data centers, delays in operation due to permitting and power supply issues can hinder the recovery of investment funds.

In fact, there has been a growing backlash from local communities against data center construction in the United States. According to Data Center Watch, more than 75 projects worth approximately $130 billion were halted or delayed in the first quarter of 2026 due to resident opposition and other issues. Concerns that the massive power consumption of data centers could lead to rising electricity costs and strain on the power grid are the main reasons.

In particular, regulatory movements are intensifying in Texas, a key data center market in the U.S. There are currently 335 data centers operating in Texas, with at least 248 more planned. However, due to rapid expansion leading to power and water shortages and increased resident opposition, even Republican Governor Greg Abbott, who had been proactive in attracting the AI industry, ordered a temporary halt to the power grid connection review for new data centers in August and directed investigations into power and water usage. Status of opposition movements against data centers by state in the U.S. in Q1 2026. Source=Data Center Watch

The midterm elections on November 3 increased the likelihood that these regulatory movements will become long-term. With the Democratic Party securing a majority in both the House and Senate, Gina Hinojosa, a Democrat, defeated Governor Abbott in Texas. Hinojosa pledged to halt new data center approvals until resident protection laws are established and to impose the costs of power grid expansion on operators while reducing tax benefits. Although this has not yet been implemented as actual policy, there are concerns that the existing temporary approval restrictions could lead to long-term regulations. The day after the election, on November 4, CoreWeave's stock price fell by 9.9%, indicating that the financial market is sensitive to regulatory uncertainties.

The problem is that these regulations could disrupt the funding recovery plans of AI infrastructure companies. Delays in data center construction and operation will postpone revenue generation, but the interest burden on already borrowed funds continues. If the costs of power grid connection and expansion also increase, expected profit margins could decrease, and the burden of debt repayment could grow.

Ultimately, AI infrastructure companies face uncertainty not only from the financial costs associated with massive borrowing but also regarding whether they can operate the planned infrastructure on time. Even if AI computing demand continues to grow, if they cannot build the infrastructure to meet it, they may face difficulties in recovering their investments. This risk may be even greater for large-scale projects that require new power generation facilities and transmission networks.

3. USD.AI Focusing on the Mid-Sized GPU Financial Market: Will It Be Free from Risks?

As uncertainties in the AI infrastructure financial market expand, the on-chain GPU collateral loan platform USD.AI claims that new opportunities are opening up in the mid-sized computing financial market. Unlike large data centers, mid-sized GPU projects that have secured existing power infrastructure can relatively reduce risks associated with new construction and permitting.

There is some validity to this claim. In fact, Texas's data center regulations vary based on scale. Governor Abbott's suspension of power grid connection reviews in August primarily targets large projects over 75MW. It is reported that approximately 250 to 300 projects are affected. In contrast, mid-sized projects between 25MW and 75MW are not subject to this connection suspension and only require investigations regarding power and water usage and community impact. Facilities under 25MW are also exempt from this investigation.

However, not all regulations exclude small and medium-sized projects. Governor Abbott has also directed a halt to new environmental permits related to data centers in September, and the newly elected Hinojosa has pledged to halt new data center approvals regardless of scale. Therefore, being a mid-sized project does not guarantee avoidance of regulations, but the burdens associated with large-scale power grid connections and new infrastructure construction may be relatively smaller.

Reducing construction risks does not mean being free from credit risks in the AI infrastructure financial market. If financial institutions begin to conservatively assess AI infrastructure loans and reduce funding, borrowers from USD.AI may also face difficulties in new investments and funding. Furthermore, if a slowdown in AI investment leads to a decrease in actual computing demand or a deterioration in the creditworthiness of client companies, it could also impact the loan repayment ability of GPU operators. If GPU prices fall, both the cash flow of borrowers and the collateral value may deteriorate simultaneously.

Of course, USD.AI is also establishing mechanisms to manage these risks. Loans are limited to a maximum of 80% of the costs for GPU introduction and deployment, with principal repayments spread over 3 to 5 years and a reserve for three months' principal and interest payments. The creditworthiness of end customers and computing usage contracts are also important evaluation factors. This structure helps reduce the possibility of loan losses. However, if the creditworthiness of end customers deteriorates or the value of GPUs declines faster than expected, existing protective measures may not be sufficient to prevent all losses.

Another variable is the funding from institutional investors. USD.AI expanded its business by launching the GPU loan trading market (GLX) in October. It sold a previously executed $98.1 million DuoSetEdgeAI loan bond to institutional investors, and Silverpoint Capital and Victory Park Capital participated from the new loan stage in the $106 million NexGen Cloud financing. Combining the sale of existing loan bonds and new joint financing, the financial scale through GLX exceeded $200 million in the first month of its launch.

This is positive as it shows that USD.AI can expand GPU financing not only with its own deposited funds but also with external institutional funds. Particularly, utilizing funds recovered from the sale of existing loan bonds for new loans can enhance capital efficiency. However, if the credit risk assessment for AI infrastructure financing as a whole is strengthened, institutional investors may reduce their purchases of GPU loan bonds or demand higher yields. In this case, delays in the sale of loan bonds and a slowdown in the turnover of funds could create burdens for executing new financing.

Ultimately, the reassessment of risks in the AI infrastructure financial market is expected to be both a new opportunity for USD.AI and a chance to validate the soundness of its business model. If existing financial institutions reduce loan supply, it may secure a new market, but if the repayment ability of borrowers and the funding from institutional investors deteriorate simultaneously, USD.AI may also be affected. Moving forward, USD.AI's competitiveness will be revealed not by the growth potential of the mid-sized GPU financial market or the scale of loan execution, but by its ability to selectively identify quality loans and recover and circulate them stably, even in a deteriorating market environment.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.