$190 Billion AI Investment Sparks Competition in Custom Chips

Microsoft ($MSFT) has announced approximately $190 billion (about 265 trillion won) in capital investment by the calendar year 2026, expanding the competition among big tech companies in AI infrastructure to include custom chips and control over data centers. While demand for NVIDIA's ($NVDA) graphics processing units (GPUs) remains strong, major clients are increasingly developing their own semiconductors and enhancing their power and cloud operational capabilities.

During the FY26 Q3 earnings conference call, Microsoft reported that its Q3 capital expenditures were $31.9 billion (about 44.5 trillion won), with approximately two-thirds allocated to short-lived assets such as GPUs and central processing units (CPUs). The company also indicated that it expects capital investments to reach about $190 billion by the calendar year 2026.

Recent earnings have been cited as a backdrop for the expansion of AI infrastructure investments. On July 29, Microsoft disclosed that its FY26 Q4 revenue was $90 billion (about 125.6 trillion won), with operating income of $40.6 billion (about 56.6 trillion won) and net income of $35.8 billion (about 49.9 trillion won). Revenue from Azure and other cloud services increased by 43%.

The expansion of investments does not solely imply increased reliance on NVIDIA. Microsoft has stated that its Maia 200 AI accelerators are operational in data centers in Iowa and Arizona, and that Cobalt server CPUs have been deployed in nearly half of its data center regions. This indicates that big tech companies are entering a phase where they seek to control costs and supply chains through their own silicon while purchasing GPUs.

Google and Amazon ($AMZN) are also moving in the same direction. In investor materials from June 2026, Google reported that 75% of its cloud customers are using AI products, and that its cloud backlog has exceeded $460 billion (about 641.7 trillion won). Amazon disclosed in its Q1 2026 earnings report that its chip business, including Graviton, Trainium, and Nitro, has annualized revenue exceeding $20 billion (about 27.9 trillion won).

Custom chips are typically designed to meet the needs of specific cloud services or internal workloads. Rather than replacing all general-purpose GPUs, they are often used first for repetitive AI inference, customized computations for cloud customers, and reducing internal service operating costs. Therefore, the expansion of custom chips is interpreted more as a strengthening of negotiation power and infrastructure control for large customers rather than an immediate reduction in NVIDIA demand.

NVIDIA's performance remains overwhelmingly strong. In FY2026 Q4, NVIDIA reported revenue of $68.1 billion (about 95 trillion won) and annual revenue of $215.9 billion (about 301.2 trillion won). Data center revenue was $62.3 billion (about 86.9 trillion won) in Q4 and $193.7 billion (about 270.2 trillion won) annually.

Jensen Huang, founder and CEO of NVIDIA, stated, "The demand for AI computing is growing exponentially." Even as large cloud companies develop their own chips, a significant portion of current demand for AI training and high-performance inference still relies on NVIDIA's platform.

NVIDIA is also expanding its footprint beyond chip sales into infrastructure financing. On August 10, the company announced plans to create an AI computing infrastructure financing platform worth over $500 billion (about 697.5 trillion won) in collaboration with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. Previously, we reported on the trend of AI infrastructure financing becoming a burden off the balance sheet.

The bottleneck in the AI infrastructure competition cannot be resolved solely by chip purchases. Data center sites, power, cooling facilities, packaging capacity, and network equipment must all be interconnected to lead to actual service revenue. Even if chips are secured, delays in operational timelines can occur if power and building capacity are not prepared.

Market interpretations are divided. Reuters reported on August 17 that investors are moving to a stage where they are distinguishing between companies that will generate long-term profits and those that will not, amid fears of AI capital investment. Optimists believe that the current investments by hyperscalers could translate into future revenue and cash flow.

On the other hand, some view the speed of monetization as a concern. Since it typically takes 12 to 18 months for data centers to go from construction to revenue generation, if the pace of capital investment increases does not keep up with revenue, valuation pressures could mount. From this perspective, both custom chips and financing platforms are interpreted as responses to lower cost control and capital procurement burdens.

This trend also serves as an indirect variable in the crypto market. AI infrastructure investments are intertwined with power, data centers, semiconductors, and cloud demand, affecting the cost structures of mining companies and AI cloud service providers. The previously reported expansion of AI infrastructure agreements and long-term burdens on big tech also fall within this context.

For domestic investors, the interplay between semiconductor and power infrastructure and cloud demand is a key observation point. While it is difficult to definitively state the impact of the NVIDIA and Microsoft dynamics on specific asset prices, the speed of AI capital investment and its monetization are likely to continue influencing the valuations of tech stocks and related infrastructure companies.

The next confirmed schedule is NVIDIA's Q2 FY2027 earnings announcement. NVIDIA will hold an earnings conference call on August 26 at 2 PM Pacific Time, where it will provide explanations regarding revenue, data center demand, and next-generation platforms.