The AI Supply Shock

Theme and Background

This chapter discusses the speed of AI’s transition from the experimental stage to large-scale deployment, as well as the core contradiction that infrastructure expansion may lag behind demand growth. The author believes that the current market is in an early stage similar to the iPhone moment in 2007, where long-term opportunities have emerged but the full picture is still taking shape.

Core Thesis

The author’s core judgment is that AI demand exhibits viral diffusion characteristics, with enterprise penetration still below 2%, while key suppliers operate in capacity-constrained, high-barrier markets. This means that some companies in the AI value chain may convert scarce supply into sustained profit growth. The counterintuitive point is that the market may underestimate the speed at which AI evolves from a tool to a “digital employee” (Agentic AI), steepening the adoption curve through human-machine interaction and machine-to-machine activities.

Key Arguments and Data

1. Viral Demand Diffusion: Anthropic’s annualized revenue surged from approximately $1 billion to over $47 billion in about 12 months, indicating that the demand curve is difficult to capture with conventional charts.

2. Extremely Low Enterprise Penetration: AI penetration among global knowledge workers remains in the low single digits and is concentrated in early use cases such as coding. Historical comparisons show that television, personal computers, and mobile phones took decades to achieve widespread penetration, while enterprise AI agents are only in the first few months of the adoption curve.

3. Supply Bottlenecks: Although the industry has installed substantial additional computing power to support early AI adoption, penetration remains low. If penetration rises from current levels to 10%, 25%, or higher, the required infrastructure scale will far exceed current levels. Key suppliers (semiconductors, high-bandwidth memory, networking equipment, data center infrastructure) operate in capacity-constrained, high-barrier consolidated markets.

Indicator	Data
Anthropic annualized revenue growth	Approximately $1 billion → Over $47 billion (within about 12 months)
Enterprise AI penetration	Below 2% (global knowledge workers)
Historical technology penetration cycle	TV/PC/mobile phones: decades; Enterprise AI agents: only months

Companies/Assets Involved

• Anthropic: As a typical case of viral AI demand diffusion, its revenue trajectory is cited as evidence of a steep demand curve.
• Zoom Communications: Used as an analogy for viral technology spread, rapidly expanding users through meeting invitations during the pandemic.
• Meta Platforms: Used as an analogy for network-effect-driven technology diffusion, where the expansion of social networks increases platform value.
• Semiconductor and Infrastructure Suppliers (not specifically named): Including advanced logic chips, foundry capacity, high-bandwidth memory, networking equipment, power systems, data center infrastructure, etc. These suppliers have pricing power in capacity-constrained, high-barrier markets.

Investment Implications

Investors should focus on companies in the AI value chain with scarce supply capabilities, especially those operating in capacity-constrained, high-barrier markets. These companies may convert demand growth into sustained profit growth. It is important to note that while current infrastructure investment has increased significantly, penetration remains low. If enterprise adoption accelerates, supply bottlenecks may further intensify, strengthening the pricing power of leading companies. At the same time, attention should be paid to business quality and valuation, avoiding the pursuit of profitless growth.

---

Theme and Background

This chapter focuses on the supply constraints of AI infrastructure, particularly semiconductors. The report argues that the market may underestimate the duration and investment value of AI infrastructure bottlenecks, as demand exhibits viral diffusion, enterprise penetration rates remain extremely low, and key supply-side components—chips, memory, networking, power, and foundry capacity—are all in a state of scarcity.

Core Views

• Semiconductor opportunities may be larger and more persistent than the market expects. The core issue is not just short-term chip demand, but the possibility that the entire AI infrastructure stack could become a bottleneck for digital growth in the years ahead.
• Scarcity can create attractive growth, but also brings risks (capacity may eventually catch up, customers may optimize usage, new architectures may reduce costs, and valuations may outpace fundamentals). It is essential to distinguish between temporary shortages and persistent bottlenecks.
• The AI opportunity will not remain confined to a single link. Early value flows to infrastructure "enablers," but long-term value creation will shift to AI application "builders" and "beneficiaries." AI should not be viewed as a semiconductor cycle or a narrow tech trade, but as a broad transformation in how work is done, products are built, and competition unfolds.
• The current phase is akin to the early days after the iPhone launch: the platform is in place, but the most valuable applications, enterprises, and beneficiaries may still be years away.

Key Arguments and Data

• Demand characteristics: The simultaneous presence of viral demand, low penetration, and constrained supply can support years of value creation.
• Bottleneck areas:
• Memory: Agents require storage, retrieval, and cross-task use of information
• Networking: Models and workloads scale across data centers
• Power: AI workloads make the economy more electricity-intensive
• Foundry capacity and semiconductor equipment: Advanced chip manufacturing is difficult, and capacity buildout takes time
• AI chips themselves: Still a critical component
• Historical analogy: The current stage is compared to the early period after the iPhone launch—the platform is ready, but the most valuable applications and business models have yet to emerge.

Companies/Assets Involved

This chapter does not specifically mention individual stocks, but implies the following directions:

• Enablers: Chips (e.g., NVIDIA), cloud capacity, power, memory, data center infrastructure providers—these remain key bottlenecks, and the opportunity will not disappear as the market matures.
• Builders: Enterprises creating new AI-driven products and services (not specifically named)
• Beneficiaries: Existing enterprises leveraging AI to enhance productivity, customer engagement, or generate new revenue streams (not specifically named)

Investment Implications

• Do not narrow AI investment to a semiconductor cycle trade. Instead, allocate across the value chain in layers: in the short term, focus on supply-constrained infrastructure (chips, foundry, memory, power); in the medium to long term, focus on the "builder" and "beneficiary" opportunities arising from the diffusion of AI applications.
• Be wary of the risk that valuations outpace fundamentals. Distinguish between temporary shortages (e.g., short-term capacity tightness) and persistent bottlenecks (e.g., barriers in advanced manufacturing, power infrastructure limitations). Only enterprises that can convert scarce supply into sustained profit growth are worth holding for the long term.