How Will AI Monetization Experiments Look Like?
As we enter into the final quarter of 2026, the question on investor’s minds is “Can the revenue generated by AI justify this extraordinary infrastructure buildout?”
GREY RHINO
Since the beginning of 2026, AI investment momentum has continued to accelerate, with the spending increasingly concentrated in data centers, GPUs/accelerators, networking, power and cooling. The five major U.S. hyperscalers—Microsoft, Alphabet, Amazon, Meta and Oracle—are now expected to spend roughly $730 billion of capex in 2026, up sharply from about $485 billion estimated at the start of the year; Goldman Sachs puts global AI investment at more than $1 trillion in 2026, while Gartner estimates total worldwide AI spending at $2.7 trillion, up 49.5% year over year.
As we enter into the final quarter of 2026, the question on investor’s minds is “Can the revenue generated by AI justify this extraordinary infrastructure buildout?” And some concerns have surfaced questioning the speed and sustainability of revenue generation from AI investment. Goldman estimates that hyperscalers need roughly $300 billion of annual AI-related revenue just to break even on the investment over the next few years. To earn a more conventional ~30% return on invested capital, the required annual revenue rises to approximately $636 billion. A more aggressive Gartner analysis suggests that if hyperscalers ultimately spend $6.3 trillion on AI infrastructure through 2030, they would need roughly $1.5 trillion of average annual revenue to achieve a 25% ROIC. Michael Burry, the asset manager famous for his “big short” bet, has been consistently criticizing the circular financing made between AI companies and Nvidia, with his most recent position pointing to the likely outcome that compute becomes commoditized faster than AI applications can generate enough high-margin revenue. The result is underwhelming returns and substantial asset write-downs.
So, as the investor community is starting to ask smarter questions, the urgency for the AI industry to offer convincing facts and numbers on financial returns increases. We are entering a period of rapid experimentation by hyperscalers who are trialing new offers and testing new pricing on both consumer and business sides. There are four types of monetization routes, each is in its respective early stages.
The Direct-charge model: Target users directly with usage-based or flat-rate price plan for using AI compute. Currently, the most successful use case for this model is the coding/programming, which is driving the revenue growth behind Anthropic and OpenAI. Meta opened another front with its launch of Muse a month ago. Muse acts as an AI assistant, targets consumers, and follows a freemium model, hoping to drive adoption first and monetization later. OpenAI followed with its Dots debut recently, although Dots appears targeting power users and more in a business setting than ordinary consumers.
The Bundle model: Microsoft was the first to experiment with this model via its CoPilot offer. For $30/month, its Microsoft 365 license can get a productivity boost via Copilot. This is also behind Gemini’s model as Gemini is offered at different tiers bundled with additional Google products and services.
The Ecosystem model: Longer term, this is the more attractive and more sustainable monetization model but we are at the early stage of it. In its simplest form, the best example is Apple Intelligence that makes Apple’s hardware smarter and keeps Apple’s ecosystem more attractive. A bit more sophisticated is Meta’s plan. Muse is an Agent to lure Meta users to start using its AI compute, and subscription revenue will be dwarfed by future earning power in delivering more targeted and outcome-driven ads. The commission revenue is perhaps just a nice-to-have piece, in my opinion. Google, on the other hand, has all the pieces to build an Apple-like ecosystem to monetize its Gemini in a variety of forms. Google designs the Gemini models, owns TPUs/data centers, owns the consumer distribution through Search/Android/Chrome/YouTube, owns Workspace, and owns Google Cloud. That means a Gemini request can potentially create economic value in several different businesses without Google having to pay another model provider. Lastly, Microsoft is doing something similar as Google but with a significant distinction–its Copilot is becoming an enterprise AI orchestration layer–whether their enterprise customers adopt the ChatGPT or Claude frontier model, Copilot helps drive various business AI agent’s usage of computing resources such as cloud, data, security, CRM, where Microsoft has built customer base already.
The Gateway model: Microsoft’s approach is now adopted by some of the leading enterprise data companies such as Snowflake, Datadog, and still private Databrick. These companies offer tools and services that assist enterprises to better use their AI investment. For instance, Snowflake's newer Cortex AI Gateway can dynamically route a workload to different models according to quality, speed and cost, while tracking AI usage and controlling spending. Databricks is taking an even broader version of this approach. Its Agent Bricks platform lets enterprises build agents using OpenAI, Anthropic, Gemini and open-source models while Databricks provides the data, security, memory, evaluation, deployment and governance infrastructure. These companies are agent-agnostic and their business model’s success is built on AI model’s success but with them, enterprises may have a harder time to benefit from their AI investment. Thus they are equally important to AI investment’s success.
The need for common metrics
All the AI front runners are experimenting with different monetization models, and their financial reporting varies in breadth and depth about AI related revenue. The investor community needs to find more commonly defined financial metrics in order to evaluate/compare AI companies’ ability to monetize their AI investment. The revenue per XXX is quite important, similar to the P/E ratio or PEG ratio commonly used by investors to compare valuation. In the short term, it can be an AI revenue to AI investment ratio as a very rough metrics. Longer term, more robust and more relevant metrics need to be agreed on to better assess the AI industry’s earning power and valuation.
Trivia Economics © 2024-2028
Trivia Economics broadens/deepens your understanding of economics and investment opportunities.
Coming Soon
Subscribe to our newsletter and never miss a trivia.
We care about your data in our privacy policy.