AI Stocks: Can Investors Still Win Big or Is The Crash Ahead?

Key Concepts

• AI Industry Layers: Applications, Foundational Models, Hyperscalers, Hardware & Infrastructure.
• Compute: Processing power required for AI, including inference (serving answers) and training (improving models).
• Inference Compute: The processing power used to serve AI model outputs to users.
• Training Runs: The computationally intensive process of improving AI models.
• Hyperscalers: Companies that build and operate massive data centers, renting out computing power.
• Unit Economics: The profitability of selling a single unit of a product or service.
• Capital Spending: Investment in long-term assets, such as data centers and equipment.
• Free Cash Flow: Cash generated by a company after accounting for capital expenditures.
• Picks and Shovels: A metaphor for investing in companies that provide the essential tools and infrastructure for a booming industry.
• Existential Threat: A perceived risk to a company's survival if it fails to adapt to a new technological paradigm.
• Bottlenecks: Constraints in the AI ecosystem that limit growth and create investment opportunities.
• Edge Infrastructure: Computing resources located closer to the end-user or device, enabling local AI processing.
• Data Quality/Moat: The proprietary and high-quality datasets that give foundational model builders a competitive advantage.
• Circular Nature of AI Spending: A feedback loop where companies invest in suppliers who then provide them with the components needed for their operations.

AI Industry Breakdown and Investment Landscape

The AI industry is experiencing unprecedented growth, characterized by rapid innovation, multi-billion dollar deals, and soaring stock prices. Tracking this dynamic sector requires understanding its complex structure and financial flows. This analysis breaks down the AI ecosystem into four distinct layers: Applications, Foundational Models, Hyperscalers, and Hardware & Infrastructure, and examines the financial dynamics and investment opportunities within each.

1. Applications Layer

This is the end-user layer where individuals and businesses interact with AI. Examples include AI-powered tools like Perplexity's notebook, Cursor for coding, and popular chatbots such as ChatGPT, Google Gemini, and Anthropic's Claude.

• Business Model: These applications typically generate revenue through subscriptions.
• Profitability: Currently, most AI applications are not profitable. The high cost of compute power required for each user action results in a net loss for these companies. They have a long path to profitability.
• Relationship with Foundational Models: AI applications are built on top of foundational models. They pay the model creators for access to AI intelligence. In cases where the application and model are built by the same company (e.g., ChatGPT by OpenAI), the application layer likely incurs costs that exceed its revenue, effectively subsidizing the foundational model.

2. Foundational Models Layer

This layer comprises the core AI models developed by major AI companies. Key players include OpenAI (ChatGPT), Anthropic (Claude), XAI (Grok), Alphabet (Gemini), and Meta (Llama).

• Costs: Building and running these models is extremely expensive, involving significant costs for both inference compute and massive training runs to improve model performance.
• Financials: These companies are not typically profitable.
  • OpenAI: Reported $4.3 billion in revenue in the first half of 2025 but incurred a loss of approximately $2.5 billion.
  • XAI: Reportedly burning about $1 billion per month.
  • Alphabet and Meta's AI model costs are also substantial, likely impacting their overall financials.
• Funding: For private firms like OpenAI and XAI, funding comes primarily from venture capital. For large tech companies like Alphabet and Meta, their massive free cash flows from advertising businesses subsidize these AI projects.

3. Hyperscalers Layer

These are the companies that build and operate massive data centers, providing the essential computing power for training and running AI models.

• Key Players: Traditionally dominated by Amazon, Microsoft, and Alphabet. The market has expanded to include specialized cloud providers like Coreweave and Oracle, which are increasing their compute capacity for AI companies. Model builders themselves are also becoming hyperscalers (Meta, OpenAI, XAI).
• Profitability: This layer is where profits begin to emerge. Hyperscalers earn significant margins by renting out compute power, demonstrating profitable unit economics.
• Investment: Despite current profitability, hyperscalers are reinvesting all profits back into expanding their infrastructure for future growth. For example, Coreweave has positive cash flow from operations but negative free cash flow due to substantial capital spending.
• Funding: Large tech companies can self-fund much of their hyperscaler growth through existing cash flows. However, debt financing is also becoming increasingly prevalent. Meta, despite having significant cash reserves, sought to raise $26 billion in private debt. Coreweave has $14 billion in private debt. The private credit industry is a major financier of AI spend, with private lending to the tech sector reaching $450 billion, an increase of $100 billion year-over-year.

4. Hardware & Infrastructure Layer

This layer represents the companies providing the physical components and infrastructure necessary for AI data centers.

• Key Areas: This includes companies manufacturing GPUs (Nvidia), networking equipment (Cisco), and components for data center construction such as wiring, cooling systems, and electrical power solutions.
• Boom Mode: Companies in this layer are experiencing a significant boom, with high revenues, substantial profits, and corresponding stock performance.
• Examples:
  • HVAC: Comfort Fix
  • Power Generation: Constellation Energy
  • Fiber Optics: Corning
  • High-Speed Interconnects: Amphenol
• Investment Thesis: These are often referred to as "picks and shovels" plays, providing essential tools for the AI gold rush.

Funding Dynamics and the AI Ecosystem

A crucial observation is that AI is not yet self-sustaining. The revenue generated from AI end-users does not cover the immense spending on data centers and infrastructure. Funding for the AI ecosystem comes from various sources at different layers:

• Applications: Primarily funded by venture capital.
• Foundational Models:
  • Private firms (OpenAI, XAI): Venture capital.
  • Large tech companies (Alphabet, Meta): Free cash flows from existing businesses.
• Hyperscalers: A mix of internal cash flows and debt financing.

The Circular Nature of AI Spending and Future Outlook

The AI ecosystem exhibits a complex and circular funding mechanism. Companies with capital are investing back into other layers of the AI stack, creating a feedback loop.

• Example: Nvidia is investing $100 billion into OpenAI, which will then use that capital to purchase chips from Nvidia. Nvidia also holds a significant stake in Coreweave, a customer that uses Nvidia chips.
• Risk and Opportunity: This circularity, while creating a strong sense of momentum and potentially de-risking by locking in suppliers and resources, also introduces non-linear risk. A swift and severe crash could occur if faith in the AI future falters.
• Long Runway for Spending: Despite the risks, the spending on AI is expected to continue for a considerable period. This is driven by the perception of AI as an existential threat by major tech CEOs (Satya Nadella, Mark Zuckerberg, Larry Page), who are committed to investing heavily to avoid being left behind.
• Strategic Spending: Current spending is not panic-driven but strategic, aimed at securing scarce resources like power capacity, chip allocation, and data center real estate.

Constraints and Investment Opportunities

Expert Josh Balin highlights three key constraints that amplify both risk and opportunity in the AI space:

1. Power Capacity: The traditional grid struggles to meet the massive electrical demands of data centers. This creates an ancillary investment thesis around electrical infrastructure, transformers, grid modernization, and efficient power transformation within data centers.
2. Data Quality: Proprietary and high-quality datasets are becoming a competitive moat for foundational model builders. Deals with platforms like Reddit, Stack Overflow, and The New York Times are crucial for training advanced models.
3. Edge Infrastructure: AI will increasingly run locally on devices like cars, factories, cell phones, and wearables. This necessitates compute infrastructure for these proliferating edge devices, presenting a significant investment opportunity.

Investment Strategies

For investors, opportunities lie within these constraint points and in companies demonstrating profitability:

• "Picks and Shovels" within Constraints:
  • Micron: High bandwidth memory (HBM) is a scarce resource for both data centers and edge devices.
  • Power and Cooling: Companies like Vertiv, which provide solutions for power and cooling in data centers, are well-positioned.
• "Profitability Now" Bucket:
  • Oracle: Possesses stable cash flows from other businesses and pays a dividend, offering a more traditional investment profile.
• Edge Expansion:
  • Qualcomm: Poised to benefit from the proliferation of AI-enabled devices.
  • Broadcom: Continues to be a strong player in the semiconductor space.

Conclusion

The AI industry is in a phase of rapid, albeit currently unprofitable, expansion. The ecosystem is characterized by significant investment, a circular flow of capital, and strategic spending driven by major tech players who view AI as an existential imperative. While risks exist due to the non-linear nature of the market, the identified constraints and the ongoing commitment to AI development suggest a long runway for continued investment and potential returns, particularly in companies providing essential infrastructure, addressing power and data bottlenecks, and enabling edge AI. The circular funding mechanism, rather than being a flaw, can be seen as a sign of maturity, ensuring continued revenue and support within the ecosystem.