Foundation Models: Building Blocks of AI Investment

Foundation models—large AI systems trained on broad data that can be adapted to many tasks—have emerged as the technological core of the AI transformation. The US has dominated AI funding, with $159 billion—79% of global funding—going to US-based companies, with the San Francisco Bay Area alone raising $122 billion. Foundation models from companies like OpenAI, Anthropic, Google, and Meta underpin applications ranging from chatbots and coding assistants to enterprise automation and scientific research.

Understanding foundation models is essential for AI investors. These models represent both direct investment opportunities and the enabling layer that determines the value potential of downstream applications. This analysis examines the foundation model landscape and investment strategies for this foundational AI technology.

---

Foundation Model Architecture

What Are Foundation Models?

Foundation models are distinguished by several characteristics:

Scale: Trained on massive datasets (billions to trillions of tokens) using substantial compute resources

Generality: Capable of many tasks without task-specific training, demonstrating emergent capabilities

Adaptability: Can be fine-tuned or prompted for specific applications with relatively little additional data or compute

Transfer Learning: Knowledge from broad pre-training transfers to new domains and tasks

Multi-Modal Potential: Increasingly capable across text, images, code, audio, and other modalities

The Scaling Paradigm

Foundation models have improved through scaling:

Parameter Scaling: Larger models with more parameters tend to perform better on diverse tasks

Data Scaling: More training data improves model capabilities and generalization

Compute Scaling: More computational resources for training enable larger, better models

Scaling Laws: Predictable relationships between scale and performance guide investment in training

Recent developments suggest additional sources of improvement:

Architectural Innovation: Better model architectures improve efficiency and capability Data Quality: Curated, high-quality training data matters as much as quantity Training Techniques: Improved training methods (RLHF, DPO, etc.) enhance model behavior Inference-Time Compute: Spending more compute at inference time improves reasoning

---

Market Landscape

Frontier Labs

The leading foundation model developers:

OpenAI:

• GPT series: Market-leading capabilities, massive developer adoption
• Microsoft partnership providing capital and distribution
• $100B+ valuation, potential for very large outcomes
• Product diversification (ChatGPT, API, Enterprise)

Anthropic:

• Claude series: Strong safety focus, competitive capabilities
• Amazon and Google investments providing resources
• Constitutional AI approach differentiating technology
• Enterprise focus with growing commercial traction

Google DeepMind:

• Gemini series: Multimodal capabilities, Google integration
• Massive resources and data access
• Distribution through Google products and cloud
• Research depth and talent concentration

Meta AI:

• Llama series: Open-source approach, broad distribution
• No direct revenue model but strategic value
• Enabling ecosystem while reducing competitor advantages
• Research leadership in certain areas

Open Source and Open Weights

An alternative paradigm gaining momentum:

Key Players:

• Meta (Llama): Most capable open weights models
• Mistral: European startup with strong open models
• Alibaba (Qwen): Chinese open models
• Together, Hugging Face: Infrastructure for open models

Investment Implications:

• Open models reduce differentiation for closed providers
• Create opportunities for fine-tuning and deployment companies
• Lower barriers for vertical applications
• Potential commoditization of base capabilities

Emerging Foundation Model Companies

Startups pursuing differentiated approaches:

Model Developers:

• Cohere: Enterprise-focused foundation models
• AI21 Labs: Text generation and understanding
• Adept: Action-focused AI models
• Character.AI: Conversational AI focus

Investment Considerations:

• High capital requirements for training
• Competitive pressure from leaders
• Differentiation through verticals or specific capabilities
• Potential acquisition targets

---

Investment Thesis

Direct Foundation Model Investment

Investing directly in foundation model companies:

Opportunities:

• Massive market opportunity as AI adoption accelerates
• Platform economics with potential for very large outcomes
• Essential infrastructure position in AI value chain
• Network effects from developer ecosystems

Challenges:

• Extreme capital intensity for training frontier models
• Concentrated competition with deep-pocketed participants
• Open-source commoditization pressure
• Regulatory and safety uncertainties

Investment Approach:

• Leaders (OpenAI, Anthropic) offer concentrated exposure to frontier development
• Specialized players may offer differentiated risk-return profiles
• Consider through venture, growth, or secondary market access

Application Layer Investment

Foundation models enable downstream applications:

Horizontal Applications:

• Writing assistants, coding tools, customer service
• Leverage foundation models without training from scratch
• Faster time-to-market and lower capital requirements
• Risk of commoditization as models improve

Vertical Applications:

• Domain-specific solutions (legal, healthcare, finance)
• Combine foundation models with proprietary data and workflows
• Deeper integration may create sustainable moats
• Domain expertise as differentiator

Investment Approach:

• Identify applications with sustainable differentiation
• Prefer companies with proprietary data or workflow integration
• Consider both foundation model risk and application execution risk

Infrastructure Investment

Supporting foundation model development and deployment:

Training Infrastructure:

• GPU/TPU providers and cloud platforms
• Training optimization tools
• Data preparation and curation

Deployment Infrastructure:

• Inference optimization
• Model serving platforms
• Edge deployment

Investment Approach:

• "Picks and shovels" strategy may offer more predictable returns
• Less dependent on which specific models win
• Growing market regardless of foundation model competition

---

Financial Analysis

Market Sizing

Foundation model market exhibits extraordinary growth:

Current Market (2025):

• Foundation model API revenue: $10-15 billion
• Infrastructure for training: $50+ billion
• Application layer: $30-40 billion

Growth Trajectory:

• API revenue growing 100%+ annually
• Infrastructure investment accelerating
• Application layer diversifying rapidly

Unit Economics

Foundation model economics vary by approach:

API Providers:

• Revenue per API call or token
• Gross margins 50-70% depending on efficiency
• High customer acquisition costs but strong retention
• Usage-based expansion within customers

Open Source/Open Weights:

• Indirect monetization through services
• Infrastructure and deployment revenue
• Enterprise support and customization
• Strategic value rather than direct revenue

Application Layer:

• SaaS subscription or usage-based pricing
• Gross margins 60-80% typical
• Unit economics depend on foundation model costs
• Value capture vs. commoditization tension

---

Investment Framework

Portfolio Construction

A diversified foundation model investment strategy:

Frontier Labs (20-30%):

• Direct exposure to leading foundation model companies
• High risk/high reward position
• Access through venture, growth, or secondary markets

Infrastructure (30-40%):

• Semiconductor and chip companies
• Cloud providers with AI focus
• Training and inference optimization

Applications (30-40%):

• Horizontal applications with strong positioning
• Vertical specialists with deep domain integration
• Developer tools and platforms

Public Market Exposure

Direct Foundation Model Exposure:

• Microsoft (MSFT): OpenAI partnership and integration
• Alphabet (GOOG): DeepMind and Gemini
• Meta (META): Llama and open AI strategy
• Amazon (AMZN): Anthropic investment and Bedrock

Infrastructure:

• NVIDIA (NVDA): Essential AI training hardware
• AMD (AMD): Alternative GPU provider
• TSMC (TSM): Manufacturing for AI chips

Applications:

• Salesforce (CRM): Einstein and enterprise AI
• Adobe (ADBE): Creative AI applications
• ServiceNow (NOW): Enterprise workflow AI

Private Market Exposure

Venture/Growth:

• Foundation model companies (OpenAI, Anthropic, Cohere, etc.)
• Application layer startups
• Infrastructure providers

Secondary Markets:

• Access to foundation model company equity
• Typically at premium valuations

---

Risk Assessment

Technology Risks:

• Rapid capability improvements may obsolete current models
• Architectural shifts could advantage new entrants
• Safety incidents could slow adoption or trigger regulation

Market Risks:

• Open-source commoditization pressure
• Hyperscaler competition and bundling
• Concentration in few winners

Regulatory Risks:

• AI governance regulations evolving
• Potential capability restrictions
• Liability frameworks uncertain

Execution Risks:

• Capital requirements exceeding funding
• Talent competition and retention
• Scaling challenges for deployment

---

Monitoring and Evolution

Foundation model investing requires ongoing monitoring:

Capability Developments: New model releases, benchmark performance Competitive Dynamics: Market positioning, pricing, partnerships Open Source Evolution: Capability parity, adoption patterns Regulatory Developments: Governance frameworks, restrictions

Workflow automation tools like n8n can systematize tracking of foundation model developments, with Swfte providing templates for monitoring AI companies and market trends.

---

Conclusion

Foundation models represent the technological core of the AI transformation, with investment opportunities spanning direct model development, infrastructure, and applications. The market exhibits extraordinary growth potential alongside significant risks from competition, commoditization, and regulatory uncertainty.

Successful foundation model investing requires understanding both technical dynamics and market positioning. A diversified approach across the value chain—from infrastructure through models to applications—may offer the best risk-adjusted returns while capturing upside from this transformative technology.

Interested in AI foundation investments? Contact FundXYZ to learn about our technology programs providing exposure to companies building the foundation of artificial intelligence.

---