“Compute Power and Credit: Nvidia’s Two Tests” — Dr Zhou Yuanfan, Chairman of the Board of Anrong (Hong Kong) Credit Ratings Co., Ltd, Gives an Exclusive Interview to Economic Observer

On August 28, Dr Zhou Yuanfan, Chairman of the Board of Anrong (Hong Kong) Credit Ratings Co., Ltd. (ARHK), gave an exclusive interview to Economic Observer, offering an in-depth analysis based on Nvidia’s latest earnings report and global AI industry trends. He presented a core judgment: the key constraint on AI industry growth has shifted from chip supply and demand to capital availability and credit boundaries. The financial transmission chain of AI capital expenditure is continuing to lengthen, and how far this AI capital cycle can go will ultimately depend on the race between rising capital costs and declining unit intelligence costs. The interview was subsequently published that day in Economic Observer and on eeo.com.cn in an article titled “Compute Power and Credit: Nvidia’s Two Tests.”

On August 26 local time, Nvidia released its results for the second quarter of fiscal year 2027: revenue of US$96.2 billion, up 106% year on year; data centre revenue of US$89 billion, up 117% year on year. It also forecast third-quarter revenue of US$108 billion and, unusually, provided long-term guidance of approximately 70% revenue growth for fiscal year 2028, temporarily dispelling market speculation that AI demand had peaked. Boosted by the positive news, Nvidia’s share price closed up 8.74% on August 27, with capital markets once again voting in favour of its growth prospects.

Behind the impressive computing performance, deeper constraints on AI industry expansion are gradually emerging. In addition, Dr Zhou provided a detailed analysis of how market attention on Nvidia had previously focused on the match between GPU capacity and AI demand, whereas the industry’s growth logic has now fundamentally changed: Nvidia’s revenue growth corresponds to the expanding AI capital expenditure of downstream customers. Such expenditure no longer relies solely on companies’ own cash flows, but is entering capital markets through diversified structures including corporate bonds, project financing, leasing, and SPVs, and is ultimately constrained by benchmark interest rates, credit spreads, and project risk premiums. According to Nvidia, capital expenditure by the world’s five largest cloud service providers is expected to rise from nearly US$800 billion in 2026 to approximately US$1.3 trillion in 2027, while the cloud industry’s order backlog has exceeded US$2 trillion.

To overcome the capital supply bottleneck, Nvidia’s role is expanding from that of a pure equipment supplier to a provider of credit support and a participant in infrastructure returns. On August 10, Nvidia announced that it would join Apollo, BlackRock, and other institutions in establishing an independent computing-power financing platform, with plans to gradually mobilise more than US$500 billion in third-party capital for AI infrastructure, while providing up to US$125 billion in support itself. Morgan Stanley described this model as “balance sheet as a service,” estimating that Nvidia’s related total credit exposure could approach US$200 billion by the end of 2028. S&P Global Ratings believes that the current arrangements have had limited negative impact on Nvidia’s credit metrics, but the continued rise in leverage across the AI ecosystem could increase volatility in its future operating and financial performance.

As AI infrastructure becomes more dependent on external financing, long-term U.S. Treasury yields have extended beyond technology stock valuations. Dr Zhou further noted that the recent rise in long-end U.S. Treasury yields reflects not only changes in inflation and monetary policy expectations, but also fiscal risk premiums and term premiums arising from expanding fiscal deficits and increased bond supply. When long-term risk-free rates exceed 5%, AI projects must demonstrate not only that they can generate revenue, but also that their returns on capital are sufficient to cover financing costs and remain consistently higher than the risk-free returns investors can obtain from holding long-term Treasury bonds.

AI infrastructure financing and U.S. fiscal financing are currently competing, at the margin, for global long-term capital. Morgan Stanley expects global AI-related debt issuance to approach US$570 billion in 2026, as leading technology companies that previously relied heavily on their own cash flows accelerate their shift toward debt financing. At the same time, the “efficiency rebound” from technological advances is offsetting cost pressures: Nvidia’s new-generation AI platform can increase inference throughput by up to 35 times, while the cost per token continues to decline, driving further expansion in total computing demand and overall capital requirements.

Dr. Zhou stated that if computing power can continue generating sufficiently high revenue and cash returns, US$500 billion in third-party capital may only be the starting point for AI infrastructure to become a new asset class. However, if core assumptions such as computing utilisation rates, revenue growth, and equipment residual values fluctuate, the credit arrangements currently used to unlock demand may become risk factors requiring repricing in the future. For now, capital markets remain optimistic about growth prospects, but Nvidia’s “credit test” has only just begun, and every movement in the price of capital is reshaping the true threshold for AI industry expansion.