Building an AI Investment Thesis: Where the Real Value Accrues

Artificial intelligence has become the defining investment narrative of the 2020s, and for the first time in a technology cycle, the hype may actually be understated. The capabilities of large language models, multimodal AI systems, and domain-specific machine learning have advanced at a pace that has surprised even the researchers building them. Companies across every industry are racing to integrate AI into their products, operations, and decision-making processes.

For investors, this creates both an extraordinary opportunity and a serious hazard. The opportunity is obvious — AI will create trillions of dollars in economic value over the coming decades, and investors who position correctly will capture meaningful returns. The hazard is equally clear — in every technology revolution, the majority of investors lose money by buying into the narrative without understanding where value actually accrues. The dot-com boom created Amazon and Google, but it also created Pets.com and Webvan. The AI boom will follow a similar pattern.

The central question for investors is not "should I invest in AI?" — the answer to that is unambiguously yes. The question is "where in the AI ecosystem should I invest?" And the answer requires understanding the layered architecture of the AI industry and the distinct economics at each layer.

The AI Stack: Four Layers of Value

The AI ecosystem can be understood as a four-layer stack, each with different competitive dynamics, margin profiles, and investment characteristics:

Layer 1: Infrastructure (Compute and Cloud)

At the foundation of the AI stack sits the hardware and cloud infrastructure that powers AI training and inference. This includes semiconductor companies (Nvidia, AMD, Intel), cloud providers (AWS, Azure, Google Cloud), and the emerging category of AI-specific cloud platforms (CoreWeave, Lambda Labs).

The bull case for infrastructure: AI workloads require enormous and growing amounts of compute. Nvidia's data center revenue has grown from $3 billion in 2022 to over $80 billion in 2025, driven by insatiable demand for GPUs. Cloud providers are investing hundreds of billions in data center capacity. The infrastructure layer captures value from every AI application, regardless of which application layer companies win.

The bear case: Infrastructure is a commodity business with high capital intensity and cyclical demand. Nvidia's margins will face pressure as competitors (AMD, Intel, custom chips from Google and Amazon) gain traction. Cloud providers are engaged in a capacity arms race that may lead to overcapacity and margin compression. And the current pace of infrastructure spending may reflect a bubble in AI investment that is not sustainable.

Investment implications for HNW investors: The infrastructure layer is dominated by public companies, and the investment opportunity is primarily through public market equity. For private market investors, the emerging AI cloud providers (CoreWeave recently IPO'd) and AI chip startups represent the primary opportunity, but these are capital-intensive businesses that require significant scale to compete.

Layer 2: Foundation Models (The Platforms)

The foundation model layer — companies building the large language models and multimodal AI systems that serve as platforms for AI applications — has attracted the most attention and the most capital. OpenAI, Anthropic, Google DeepMind, Meta AI, Mistral, and others are engaged in a fierce competition to build the most capable and efficient AI models.

The bull case: Foundation models are potential platform businesses — if one or two models become the default infrastructure for AI applications, they could capture enormous value, similar to how operating systems and cloud platforms captured value in previous technology eras. The network effects from data, users, and developer ecosystems could create durable competitive moats.

The bear case: Foundation model training is extraordinarily expensive ($100 million to $1 billion+ per model), model capabilities are converging (the gap between the best and second-best model is narrowing), and open-source models are providing free alternatives that may limit commercial pricing power. The economics of foundation model companies remain unproven — most are burning capital at rates that require sustained venture funding or corporate backing.

Investment implications: Foundation model companies have attracted valuations that assume platform-level dominance — OpenAI's valuation exceeded $300 billion in its most recent round. For private market investors, the entry prices at these valuations require extraordinary conviction in the winner-take-most hypothesis. A more prudent approach is to gain exposure through diversified VC funds that have positions in multiple foundation model companies.

Layer 3: Application Layer (Vertical and Horizontal AI)

The application layer encompasses companies that build AI-powered products and services for specific use cases — either horizontal (applicable across industries) or vertical (specific to a single industry). This is where we believe the most attractive risk-adjusted investment opportunities exist for private market investors.

Horizontal AI applications include tools for code generation, content creation, customer service automation, data analysis, and workflow automation. Companies like Jasper, Copy.ai, Writer, and Glean are building products that augment or replace human workers across multiple industries.

Vertical AI applications build AI solutions for specific industries — healthcare diagnostics, legal document analysis, financial risk assessment, manufacturing quality control, agricultural optimization, and dozens of other domains. Companies like Viz.ai (radiology), Harvey (legal), and Eigen Technologies (financial services) are applying AI to specific professional workflows.

The bull case for applications: Application layer companies build on top of foundation models (reducing their R&D costs), add domain-specific data and workflows (creating differentiation), and sell to end customers who value solutions to specific problems (enabling premium pricing). The application layer historically captures the most value in technology platform shifts — Salesforce captured more value from cloud computing than most cloud infrastructure companies.

The bear case: Application layer companies face the constant risk of "platform risk" — the foundation model providers could move into their vertical and commoditize their offering. They also face the challenge of building durable moats when the underlying technology (the foundation model) is available to any competitor.

Investment implications: We believe the application layer offers the best risk-adjusted opportunity for angel investors and early-stage VCs. The capital requirements are modest compared to infrastructure and foundation model companies, the market opportunity is enormous, and the key success factors — domain expertise, customer relationships, data assets, and workflow integration — are the same factors that have historically created durable businesses in previous technology cycles.

Layer 4: AI-Enabled Services and Workflows

The outermost layer of the AI stack consists of companies that use AI as an enabling technology within broader service or workflow offerings — consulting firms, managed service providers, and companies that combine AI with human expertise to deliver outcomes. This layer also includes the growing ecosystem of AI implementation, integration, and management services.

Investment implications: This layer is less attractive for venture-style investment because the businesses tend to be services-oriented with lower margins and scalability constraints. However, for investors interested in lower-risk, cash-flow-generating businesses, AI-enabled services companies can be attractive buyout or growth equity targets.

Our Investment Framework

Based on this analysis, here is how we recommend HNW investors construct an AI-focused portfolio:

What to Own

Vertical AI applications with domain moats. Companies that combine AI capabilities with deep domain expertise, proprietary data, and embedded customer relationships are best positioned to build durable, high-margin businesses. Prioritize verticals where the AI solution replaces expensive human expertise (legal, medical, financial), where data is proprietary and difficult to replicate, and where switching costs are high.

AI infrastructure picks-and-shovels in the private markets. Companies building the tooling, monitoring, security, and governance infrastructure that enterprises need to deploy AI at scale — MLOps platforms, AI observability tools, data labeling services, model evaluation frameworks — are relatively capital-efficient and benefit from the growth of AI regardless of which models or applications win.

AI-first companies in regulated industries. Healthcare, financial services, insurance, and government represent enormous markets where AI adoption is accelerating but where regulatory requirements, compliance needs, and domain complexity create barriers to entry. Companies that navigate these barriers successfully build durable competitive positions.

What to Avoid

Me-too AI wrappers. Companies whose entire value proposition is a thin interface on top of a foundation model API — without proprietary data, unique workflows, or domain expertise — are extremely vulnerable to commoditization. If the product can be replicated by a competent engineer in a weekend using the OpenAI API, it is not a defensible business.

Pre-revenue foundation model companies at hyper-valuations. Unless you are investing through a top-tier VC fund with deep AI expertise, the risk-reward of investing in foundation model companies at current valuations is unfavorable.

AI hardware startups without massive capitalization. Building competitive AI chips requires billions in R&D and manufacturing investment. Startups competing against Nvidia, AMD, and the hyperscaler custom chip programs face an extraordinarily difficult competitive landscape.

What This Means for Investors

AI is real, transformative, and investable. But the distribution of returns within the AI ecosystem will be highly uneven, and investors who allocate indiscriminately to "anything AI" will likely be disappointed.

1. Focus on the application layer. The best risk-adjusted returns for private market investors will come from AI application companies with domain expertise, proprietary data, and clear customer value propositions. This is where your angel investing and early-stage VC capital should be concentrated.

2. Evaluate AI moats critically. Ask every AI company: "What happens when the foundation model you use gets 10x better or 10x cheaper?" If the answer is "our product becomes commoditized," the company does not have a moat. If the answer is "our product becomes 10x more valuable because our domain data and workflow integration become more powerful," the company has a real moat.

3. Diversify across AI sub-sectors. Build a portfolio that spans multiple vertical applications, horizontal tools, and infrastructure plays. The AI ecosystem is evolving rapidly, and the winners in specific segments are not yet determined.

4. Be patient with valuations. AI startup valuations are elevated across the board, reflecting the enormous narrative momentum. Be willing to wait for reasonable entry points rather than chasing deals at peak hype valuations. The best AI companies will be valuable regardless of whether you invest in 2026 or 2027.

5. Invest in what you understand. If you have domain expertise in healthcare, apply it to evaluating healthcare AI companies. If you understand financial services, focus on fintech AI applications. Your domain knowledge is a genuine edge in evaluating AI companies, because the technology alone is not what creates value — the application of technology to real problems is.

The AI revolution is here, and it will create generational wealth. The question is whether that wealth will accrue to you or to someone who understood the stack better. Choose your layer wisely.