The risks of this type of chips that are promising have increased significantly!

Morgan Stanley Securities (Morgan Stanley) released the "Greater China Semiconductor" industry report, pointing out that artificial intelligence (AI) is hot, and the annual compound growth rate (CAGR) of AI semiconductors will exceed 40% from 2023 to 2027, reaching US$290 billion, accounting for 35% of global semiconductor demand. In the Taiwan stock market, Morgan Stanley is bullish on TSMC, Mason Technology and JDT, and maintains an "outperform" rating.

Morgan Stanley pointed out that in the AI ​​semiconductor market, the CAGR of inferred AI semiconductors will be as high as 91%, which is higher than the growth rate of training AI semiconductors; the CAGR of edge AI semiconductors is 25%, and "hybrid AI" computing will appear; the CAGR of customized AI semiconductors is 100%, mainly from hyperscale cloud operators, electric vehicle operators and start-ups.

Morgan Stanley pointed out that the driver of AI semiconductor demand is generative AI, but it also faces four major growth constraints. The first is budget, because the budget of cloud and enterprises in AI is not unlimited, unless they can improve the return on investment through killer applications. The second is energy. AI computing must consume more electricity, which will eventually lead to ESG considerations. The third is production capacity. CoWoS production capacity will double by 2025. The fourth is regulations. The Chinese mainland market is driven by policies.

AI semiconductors face growth limitations, but there are solutions. Morgan Stanley pointed out that, for example, under Moore's Law, the evolution of chips to 3 nanometers and 2 nanometers can improve energy efficiency; advanced packaging CoWoS/SoIC increases information processing speed; co-packaged optical components (CPO) break through network speed limitations; high-bandwidth memory (HBM) expands memory capacity; and when the number of customized chips increases, costs can be reduced and energy can be saved.

Among the relevant Taiwanese companies, Morgan Stanley said that TSMC is an enabler of AI semiconductor production and a long-term beneficiary. Optimistic about the huge demand for AI, Morgan Stanley listed TSMC as the top choice of Greater China semiconductor stocks.

Meanwhile, the Financial Times said in an article that financial history is full of strange and wonderful examples of collateral. For example, in the 19th century, Peru used its future revenue from guano (a substance made from bat, bird and seal feces) to secure loans for large projects. This feces mixture is an effective fertilizer and is easily available in the nearby Chincha Islands. Fortunately, the pungent smell of securities is not as strong today, but they are not necessarily less toxic. Problematic mortgage-backed securities triggered the 2008 financial crisis. So what does the latest financial innovation: mortgaging artificial intelligence chips mean?

According to the Financial Times, Wall Street's largest financial institutions have loaned more than $11 billion to "neocloud" groups that hold Nvidia's AI chips. These companies, including CoreWeave, Crusoe and Lambda, provide cloud computing services to technology companies that develop AI products. They have obtained tens of thousands of Nvidia graphics processing units (GPUs) through cooperation with the chipmaker. As data center capital expenditures surge, the company's chips have become a valuable commodity in the rush to develop AI models.

Enthusiasm for new technologies often comes with, and in turn reinforces, financial innovation. Two centuries ago, during the railroad boom in the U.S. and U.K., some railroad companies got loans to lay more track, partly secured by existing lines. Today, Neoclouds are following their lead. They provide data storage infrastructure to AI developers through power purchase agreements. They get loans from firms like Blackstone, PIMCO, Carlyle, and BlackRock, secured by Nvidia chips, which then allow them to buy more. In the event of a default, the lenders get the chips and the lease.

The rapid growth of a new debt market in an industry that is still in its infancy requires caution. For one thing, chips are unlikely to retain their collateral value for long. While GPU demand remains high, supply has increased as hardware reserves are resold, and supply could increase further as lease contracts expire. New chips developed by Nvidia or its competitors, including Microsoft, Google and Amazon, could also erode the value of existing collateral.

Second, these deals could push up valuations in the sector. The exact details of the agreement between Nvidia and Neoclouds are unclear. But the chipmaker itself is an investor in some startups, which in turn are among its largest customers. Cloud providers that receive loans in exchange for Nvidia chips can use the funds to buy more chips from Nvidia. This dynamic could exaggerate Nvidia's earnings and means that Neocloud groups are also at risk of being overleveraged. Third, partnerships with cloud providers could allow Nvidia to maintain its chip dominance, which increases market concentration risks.

The chips-for-securities trend is still nascent, and based on current loan volumes, Wall Street’s biggest financiers may not be too concerned about their exposure just yet. But the development does shed light on some of the risky lending, revolving financing, and competitive dynamics that are underpinning the AI ​​boom. Investors should be wary of potential pitfalls. Nvidia might be wise to draw a clearer line between its commercial interests and its venture capital interests, which would support market transparency.

Financial innovation is often positive, and when done well, it can direct capital to projects that promote growth. But as billions of dollars continue to flow into AI infrastructure, pressure on developers to generate revenue is increasing. If risky and opaque financial engineering continues to fuel the frenzy, prices could drift further away from reality. In that case, if a correction occurs, the pain will be deeper and more widespread.

Recommended Reading