The AI computing power bottleneck is moving upstream! Behind Micron's (MU.US) investment in Global Wafer: The storage supercycle is moving from “missing cores” to “stealing silicon wafers”
Zhitong Finance App learned that as part of plans to expand investment in chip manufacturing in the US, Micron Technology (MU.US) has taken a stake in GlobalWafers (GlobalWafers), a leader in silicon wafer manufacturing headquartered in Taiwan, to provide key financial support for its large-scale wafer manufacturing facility in Sherman, Texas, USA. Wedbush Securities, a well-known investment agency on Wall Street, believes that the supply of silicon wafers, the core raw material needed to expand chip production capacity, may eventually become another “key AI computing power hardware” bottleneck facing the AI computing power industry chain and even the global technology industry.
According to senior stock market analysts from major Wall Street firms such as Goldman Sachs, Morgan Stanley, and Nomura, the main line of the global AI superbull market, including US stocks, is being upgraded from “a single point explosion in demand for Nvidia AI GPUs” to a “bullish wave of the entire AI computing power infrastructure industry chain”, which will also support the US stocks and global stock markets to continue to interpret the super bull market surrounding AI.
The most typical example is that Nvidia's latest performance clearly highlights that the global AI computing power infrastructure craze is far from over, and is expanding from AI GPU/AI ASICs to data center CPUs, high-performance network infrastructure, enterprise-grade HBM/DRAM/NAND storage, machine-level server clusters, AI gigafactories, and enterprise-grade large-scale AI cloud computing systems. On Wall Street, analysts are increasingly bullish on Nvidia, the “global leader in AI.” The Wall Street average target price means that Nvidia's market value is expected to exceed 7 trillion US dollars.
According to the latest research report released by IDC, a world-renowned research institute, the world's highest market capitalization company, AI chip superleader Nvidia (NVDA.US), became the largest supplier in the global data center Ethernet switch market based on revenue benchmarks. IDC's latest report is consistent with the views of Wall Street giants such as Morgan Stanley, Goldman Sachs, and Bank of America. That is, AI computing power industry chain leaders such as Nvidia are advancing the computing power chain coverage from “GPU/AI chip single-point control” to “GPU/CPU cabinet cluster+memory chip+Ethernet switch cluster+DPU+optical interconnect system+software ecosophy+data center power chain+AI server manufacturing”.
As a result, the “cake” of the AI super-bull market is expanding and continuing to grow from GPU core assets to hardware bottlenecks related to full-stack AI computing power infrastructure. Nvidia is still the pricing anchor that defines the AI bull market, but PCBs, MLCC, ABF, ODM, liquid cooling, power supplies, silicon wafers, CMP consumables, photoresists, glass substrates, SOI/InP, data center optical interconnect equipment, optical modules, and advanced packaging equipment are all likely to receive a new round of valuation and revaluation.
The core supply capacity of the world's advanced semiconductor silicon wafers is highly concentrated in the hands of a few manufacturers such as Japan's Shin-Etsu Chemical and Shenggao, Taiwan's Global Wafer, Germany's Shichuang Electronic Materials, and South Korea's SK Siltron. Technically, silicon wafer manufacturers need to first melt semiconductor-grade high-purity polysilicon, grow monocrystalline silicon rods with almost no lattice defects through direct drawing or zone melting, and then complete doping, slicing, grinding, etching, polishing, and epitaxial growth; the final silicon wafer is only about 0.7-0.8 mm thick, but it must meet extreme flatness, surface cleanliness, and trace metal contamination standards.
Silicon wafers are both the physical foundation of all transistors and storage units, and the “starting limit” of a fab's yield: any tiny crystal defect, warping, or particle contamination can be amplified in hundreds of subsequent processes, reducing the number of qualified chips a single wafer can produce. Therefore, without sufficient and certified high-specification silicon wafers, even if the wafer factory's lithography machine and advanced packaging production capacity are all in place, capital expenditure cannot be converted into actual chip production.
Wedbush: Micron invests in Global Wafer to lock in the lifeblood of silicon wafer production capacity
“We believe that Micron's investment in global wafers may indicate that Micron sees wafer supply as another potential AI computing power hardware bottleneck. Considering that demand for semiconductor-grade wafers (wafers) is likely to increase dramatically as investment in memory chips and logic chips is likely to continue to expand in 2028, 2029, and 2030, Micron management's judgment is very reasonable,” Matt Bryson, a senior financial marketing analyst at Wedbush Securities (Wedbush Securities), wrote in a report to clients.
Micron said in a statement that the company plans to invest up to $3 billion to strengthen the semiconductor supply chain ecosystem in the US, including a stake in Global Wafer in Taiwan, China.
The analyst team led by Bryson further analyzed that these latest promises by Micron largely indicate that “memory chip customers expect demand for AI computing power to be stronger in the future, which may greatly extend the current storage supercycle, and even additional production capacity will eventually bring additional risk of oversupply.”
He also pointed out that since the US federal government is actively cooperating with the chip manufacturing industry to jointly address the shortage of memory chips, the risk of related oversupply or supply chain shortages has been limited to a certain extent.
Driven by the surge in storage demand in the age of artificial intelligence, Micron currently estimates that the total investment by 2035 will exceed 250 billion US dollars, higher than the previous forecast of 200 billion US dollars. The memory chip giant expects that the latest increase in investment will support its long-term growth goal of producing 40% of the company's dynamic random access memory (DRAM) in the US while creating more well-paid direct and indirect jobs.
In June 2025, Micron management said at the time that it plans to invest about 200 billion US dollars in the US to meet the growing demand for HBM/DRAM/NAND memory chips in the US, and eventually manufacture 40% of the company's dynamic random access memory product line in the US.
Is the storage supercycle spreading to a “shortage of silicon substrates”?
Since this year, the global capital's grand investment narrative of “seeking silicon-based inflation and weakening the carbon base” is essentially a shift of capital from “carbon-based assets” that rely on population, resources, and linear economic growth, such as traditional manufacturing, automobiles, consumption, real estate, and energy, to a high-end manufacturing chain around silicon wafers related to AI computing power infrastructure. This is not simply a story about chasing technology stocks; rather, global capital is repricing “the core carrier of future growth”: whoever has control of AI computational power infrastructure resources related to AI training/inference will get a higher valuation premium, but the relative allocation of capital will shift from old economic assets that rely on population, oil and gas, real estate, and consumption cycles to infrastructure assets that can support AI training/inference and automated physical AI productivity expansion.
The core logic of current global capital embracing “silicon-based inflation” is unquestionably that what is scarce in the AI era is not traditional labor, real estate, or general production and manufacturing capacity, but “silicon-based production materials” such as GPU/ASIC, HBM/DRAM/NAND memory chips, data center CPU components, high-performance Ethernet infrastructure, advanced packaging capacity, EUV and other cutting-edge semiconductor manufacturing equipment, data center power chains, and data center optical interconnection/optical communication.
The 300 mm semiconductor-grade silicon wafers required for the world's advanced chip manufacturing processes, including AI chips and high-end DRAM memory chips, may become the latest structural bottleneck in the AI computing power industry chain. Micron plans to provide 500 million US dollars in financing to Global Wafer and sign a 10-year silicon wafer supply agreement. The core purpose is clearly not only financial investment, but also to lock in key inputs ahead of time for future US DRAM, NAND and high-bandwidth memory capacity expansion; this shows that Micron already regards silicon wafer supply security as a production expansion constraint at the same level as wafer manufacturing equipment, advanced packaging, and electricity.
If storage and advanced logic fabs expand production at the same time from 2028 to 2030, the increase in chip production must first be translated into higher “wafer input.” Dynamic random access memory and flash memory mainly use high-quality 300mm polished silicon wafers, while advanced logic chips rely heavily on epitaxial silicon wafers; advanced nodes require not only more silicon wafers, but also lower crystal defect density, higher flatness, more stringent metal contamination control, and more stable electrical resistivity.
It is worth noting that such materials require long-term customer certification, and fabs cannot switch suppliers as quickly as replacing ordinary industrial raw materials, so even if there is no shortage of global silicon wafers, high-specification products that meet the requirements of high-bandwidth memory and advanced logic may be partially in short supply.
Micron's move has undoubtedly greatly strengthened the judgment that AI supply chain bottlenecks are continuing to shift from GPUs, HBM, and advanced packaging to silicon wafers and other basic materials. Advanced 300 mm silicon wafer suppliers may gain longer order visibility, higher capacity utilization, and stronger long-term agreement bargaining power, thus becoming upstream beneficiaries in the AI capital expenditure diffusion market. However, the 10-year long Xiehe supplier financing may also be aimed precisely at avoiding future shortages; if various silicon wafer manufacturers simultaneously expand production during the most optimistic phase of demand, and the growth rate of chip capital expenditure falls around 2029, today's bottlenecks may also turn into tomorrow's excess.