Industry Structure of the U.S. Semiconductor Market

1. Industry Structure of the U.S. Semiconductor Market

Currently, the U.S. semiconductor market is being reshaped by three major trends: 'multipolarization and de-dependency from NVIDIA's dominance,' 'domestic reconstruction of foundries (Intel's fierce pursuit),' and 'the intersection of hyperscaler Capex expansion and ROI verification.'

2. Four Major Factors Accelerating Structural Change

① Full-scale start of 'de-NVIDIA (multi-vendor strategy)'

NVIDIA has dominated the market for a long time, but hyperscalers and AI model developers are diversifying their procurement sources to avoid supply shortages, high prices, and vendor lock-in.

AMD's counterattack: Anthropic announced a strategic investment of up to $5 billion in AMD and the introduction of the next-generation AI accelerator 'Instinct MI450' and new infrastructure 'Helios.'
The wave of proprietary silicon (ASIC) development: While partnering with companies like Broadcom, Anthropic, Meta, and Alphabet are rapidly accelerating the development of 'in-house chips (2nm/custom ASICs)' optimized for specific models.

② Revival of Intel Foundry and strengthening of U.S. manufacturing capabilities

In its Q2 2026 earnings report on July 23, 2026, Intel announced results that significantly exceeded market expectations, with revenue of $16.1 billion (up 25% year-on-year) and Data Center and AI (DCAI) segment revenue of $6.3 billion (up 59% year-on-year).

Practical application of 18A process and High-NA EUV: Yields for the cutting-edge process (Intel 18A) using ASML's next-generation 'High-NA EUV' have improved, and commercial shipments are accelerating.
Upward revision of Capex (capital expenditure): To meet strong customer demand, Intel has raised its 2026 capital expenditure outlook to over $20 billion, strengthening advanced packaging and foundry capabilities within the U.S.

③ Continued massive Capex by giant IT companies (hyperscalers)

The trend of expanding capital investment in AI infrastructure continues.

Alphabet's investment expansion: In the Q2 2026 earnings report, Google Cloud revenue surged 82% year-on-year to approximately $24.8 billion. The annual capital expenditure forecast for 2026 was raised to approximately $200 billion.
Continued excess demand: Demand for server CPUs, AI accelerators, and HBM (High Bandwidth Memory) continues to exceed supply, and supply chain constraints remain a bottleneck for performance.

④ 'Concerns over over-investment' and 'strict selection' in the stock market

As a whole, the market remains sensitive to geopolitical risks (Middle East situation and rising oil prices) and interest rate trends. The phase of buying based solely on the 'AI-related' theme is over, and we have entered a phase where it is strictly evaluated whether that capital investment is actually generating profit (ROI).

'Multipolarization and de-dependency from NVIDIA's dominance'

NVIDIA has monopolized the market until now, but to avoid supply shortages, high prices, and dependence on a specific vendor (vendor lock-in), 'multi-vendor' strategies are rapidly advancing across the entire market.

Strong counterattack by AMD and the formation of a massive ecosystem: Major AI model developer Anthropic has made a strategic investment of up to $5 billion in AMD.
With this, the introduction of the next-generation AI accelerator 'Instinct MI450' and new infrastructure 'Helios' was announced, solidifying its foothold as a rival to NVIDIA.
The wave of in-house chips (custom ASICs): Hyperscalers and AI companies such as Alphabet (Google), Meta, and Anthropic are rapidly accelerating the development of 'in-house chips (2nm/custom ASICs)' optimized for the computational processing of their specific models.
Broadcom and Marvell are generating high revenue in this design support.
'Return to on-premises (in-house environment)' as governance: Due to the rise of open-weight models represented by China's Moonshot AI 'Kimi K3,' companies are strengthening moves to run AI in their own data centers (on-premises) rather than relying on specific companies' cloud APIs to protect confidential data.
As a result, 'which chip to use' itself has become a corporate security and governance strategy.

'Domestic reconstruction of foundries (Intel's fierce pursuit)'

Intel, which advocates for strengthening manufacturing capabilities within the U.S., is showing technological breakthroughs and a rapid recovery in performance.

Mass production and shipment of next-generation semiconductors using "High-NA EUV": Intel Foundry has succeeded in improving the yield of its cutting-edge process (Intel 18A, equivalent to approximately 1.8nm) using ASML's next-generation "High-NA EUV" lithography equipment, and has begun full-scale mass production and shipment for the first time in the world.
This physically increases transistor density.
Disruption of AI inference costs and the evolution of edge AI: This next-generation semiconductor dramatically reduces heat and power consumption during AI model processing.
As a result, not only are data center operating costs (TCO) and AI API usage fees falling, but the foundation is being laid for running massive AI models at high speed on laptops and smartphones without relying on the cloud.
Rapid earnings growth and massive investment: In the second quarter of 2026, Intel's Data Center and AI (DCAI) division saw revenue surge 59% year-over-year to $6.3 billion.
To meet strong demand, the company has raised its 2026 capital expenditure outlook to over $20 billion, strengthening its packaging and foundry capabilities within the United States.

A deep dive into "The intersection of hyperscaler Capex expansion and ROI verification"

Infrastructure investment supporting the AI revolution has ballooned to its largest scale ever, but in the stock market and corporate settings, the perspective on cost-effectiveness has become extremely strict.

Bottomless infrastructure investment (Capex) and supply shortages: Following a 82% year-over-year surge in Google Cloud revenue in its Q2 earnings report, Alphabet (Google) has raised its annual capital expenditure forecast for 2026 to approximately $200 billion.
Furthermore, demand for HBM (High Bandwidth Memory), which is essential for extracting performance from server CPUs and GPUs, continues to outstrip supply, and Micron and SK Hynix are maintaining high levels of sales.
Transition from "buying on expectation" to "focusing on practical benefits and ROI": In the stock market, the PoC (Proof of Concept) phase of "buying because it's AI-related" is over, and we have entered a "practicality and productivity verification period" to measure whether those massive investments are actually contributing to revenue growth (Top-line) or cost reduction (Bottom-line).
Polarization of software and service companies: Corporate AI budgets have become severe, and funds are concentrated on solutions that can achieve a payback within 3 to 6 months of implementation.
Companies (such as Palantir and AppLovin) that incorporate "AI agents" that autonomously complete practical tasks instead of just chatbots into their core operations are being highly valued by the market for significantly improving their actual profit margins.

1. From "NVIDIA dominance" to "multipolarization (multi-vendor)"

• In-house development (ASIC) and AMD's pursuit: The shift away from NVIDIA dependence is in full swing, driven by hyperscalers (Alphabet, Meta, etc.) developing their own 2nm-class custom ASICs and large-scale investments in AMD (MI450/Helios) by companies like Anthropic.

• Breakthroughs by design partners: With the rapid increase in proprietary chip development, infrastructure foundation companies such as Broadcom and Marvell, which handle ASIC design, are increasing their presence.

2. "Intel Foundry's revival" and strengthening of domestic manufacturing in the US

• Mass production of 18A process & High-NA EUV: Intel succeeded in the mass production and shipment of the 18A (equivalent to 1.8nm) process, significantly exceeding market expectations in its 2026 Q2 earnings (DCAI division +59% year-over-year).

• Reshoring of physical infrastructure to the US: Against the backdrop of geopolitical risks and supply chain concerns, advanced packaging bases for companies like Intel, TSMC, and Amkor are being rapidly expanded within the US mainland.

3. "AI investment (Capex)" moves from thematic buying to the "practicality and ROI evaluation" phase

• Continuation of massive investment and excess demand: Hyperscalers like Alphabet are carrying out capital investments on the scale of hundreds of billions of dollars annually. Demand for HBM (Micron / SK Hynix) and AI server chips continues to significantly exceed supply.

• Stricter return on investment (ROI): The stock market has finished buying into mere "AI-related" stocks and is concentrating (selecting) funds into companies building practical AI agents that generate profits in 3 to 6 months, or semiconductor companies that can produce substantial infrastructure revenue.

The US semiconductor market in the summer of 2026 is entering a very healthy and mature second stage, where "coexistence of diverse AI chips," "rebuilding of manufacturing bases within the US," and "strict practical verification of investments" are proceeding simultaneously.