The Great AI Divergence: How Frontier Models Concentrate Wealth at the Top

In 2025, the combined market capitalization of the six largest AI-related companies — NVIDIA, Microsoft, Apple, Alphabet, Amazon, and Meta — exceeded $15 trillion. The AI industry raised over $80 billion in venture capital. NVIDIA alone reported $215.9 billion in revenue with an 84% gross margin — the highest of any industrial company in modern history.

Meanwhile, the typical AI API developer earned approximately $175,000 per year. The median US household income was $80,610. The global median wage was approximately $11,000.

The gap between those who control AI capital and those who work with AI tools is not an accident of market dynamics. It is a structural feature of how the AI industry is organized — and it is about to get worse.

The IMF's Warning

The International Monetary Fund, not typically an alarmist institution, issued a stark warning in 2025: artificial intelligence will worsen inequality, both within and across nations. The IMF's analysis found that AI would primarily benefit higher-income workers and capital owners, while displacing or reducing wages for workers in routine cognitive occupations.

The mechanism is straightforward. AI automates tasks previously performed by white-collar professionals: writing, analysis, coding, translation, customer support. It complements senior professionals (who can leverage AI to become more productive) while replacing junior professionals (whose entry-level tasks are automated away). The result is a compression of career ladders and a widening gap between experienced professionals and those at the beginning of their careers.

The Stanford HAI Artificial Intelligence Index confirms this trend. The 2025 report found that wage polarization is accelerating alongside AI adoption, with the gap between high-skill and low-skill wages widening faster than at any point since the 1980s.

Where the Money Goes

The MSCI AI Exposure Index provides a revealing snapshot of where AI revenue actually flows. Approximately 92% of AI industry revenue is concentrated in just 6 companies: NVIDIA, Microsoft, Google, Amazon, Meta, and Apple. The next tier — AMD, Intel, IBM, Salesforce — accounts for perhaps another 5%. The vast ecosystem of AI startups, API providers, and application layer companies competes for the remaining 3%.

This is not an accident. The AI stack has a natural monopoly structure at each layer:

• Hardware: NVIDIA controls approximately 80-90% of the AI GPU market. Creating a competitive alternative requires years of design, billions in capital, and access to the same advanced fabrication nodes that are themselves capacity-constrained.

• Cloud platforms: AWS, Azure, and GCP together control over 65% of cloud infrastructure. AI workloads require dense GPU clusters, specialized networking, and high-bandwidth interconnects that only the hyperscalers can provide at scale.

• Frontier models: OpenAI, Anthropic, and Google DeepMind control access to the largest and most capable models. Training a competitive frontier model now costs $500 million to $1+ billion, creating a prohibitive barrier to entry.

Each layer of concentration reinforces the others. NVIDIA's GPU dominance shapes which models can be trained. The hyperscalers' cloud dominance shapes how models are deployed. The frontier model labs' API dominance shapes how applications are built. Capital flows from each layer to the next in the circular pattern described in our first article.

Labor Market Effects

The World Economic Forum's "Future of Jobs" report estimated that AI would create 97 million new jobs while displacing 85 million — a net positive of 12 million jobs. But this aggregate figure masks significant distributional effects.

The new jobs — AI engineers, prompt specialists, model trainers — are concentrated in technology hubs and require specialized skills. The displaced jobs — writers, translators, customer service representatives, data entry clerks — are distributed broadly across the economy and require skills that are not easily transferred.

More concerning is the "hollowing out" of middle-skill occupations. The polarization effect means that workers at the top of the income distribution benefit from AI (higher productivity, higher wages) while workers at the bottom are somewhat insulated (many physical-service jobs are harder to automate). Workers in the middle — the administrative, analytical, and creative professionals who formed the backbone of the knowledge economy — face the greatest displacement risk.

The Stanford HAI index found that AI-related job postings grew by 200% from 2021 to 2024, but the skills required shifted dramatically toward AI-specific expertise. A writer or analyst displaced from their role cannot simply "learn AI" and return to the workforce at the same wage level.

What Reskilling Actually Costs

Corporate and government commitments to AI reskilling are dramatically insufficient relative to the scale of the disruption. A 2025 analysis found that for every $1 of AI valuation gains (increased market capitalization of AI companies), approximately $0.02 is spent on worker reskilling and transition support.

At NVIDIA's valuation increase of roughly $2 trillion over two years, that implies approximately $40 billion in reskilling spending. Actual spending is below $1 billion globally.

The mismatch is not trivial — it is structural. The companies generating AI-driven profits have no incentive to fund the retraining of displaced workers in other sectors. And governments, facing tight budgets and competing priorities, have not stepped in to fill the gap at anywhere near the required scale.

The Concentration Spiral

The AI industry's wealth concentration creates a self-reinforcing dynamic. Capital generates AI advances, which generate more capital, which flows to the same concentrated nodes. NVIDIA's $181.6 billion in gross profit funds R&D for the next generation of chips. Microsoft's $40 billion in Azure AI revenue funds deeper AI integration across its product suite. The frontier model labs raise hundred-billion-dollar valuations that attract the best talent, further entrenching their lead.

This concentration is not inherently illegal or even unusual by technology industry standards. But the rate of concentration — and the scale of the wealth being accumulated at the top — is historically extraordinary. The AI industry is on track to generate trillions in market value while employing fewer than 50,000 people directly.

For comparison, the automotive industry employs approximately 10 million people globally. The retail industry employs over 150 million. The AI industry's productivity gains accrue to a tiny fraction of that workforce.

The great AI divergence is not a future possibility. It is happening now. And the data suggests it will accelerate before it slows.

Sources: International Monetary Fund (AI and inequality working paper, 2025); Stanford HAI Artificial Intelligence Index 2025; MSCI AI Exposure Index; World Economic Forum (Future of Jobs report); U.S. Census Bureau (median household income); Bloomberg and SEC filings (market cap data); Freeland, "The New Concentration: AI and the 0.1%" (SSRN, 2024); PwC Global AI Jobs Analysis.