AI quarterly revenue surpasses depreciation costs for the first time—has the trillion-dollar computing power gamble entered the return validation phase?

```

Over the past two years, the core debate in capital markets has remained unchanged: is the AI boom driven by real demand or is it a capital expenditure bubble?

Exponential View, a research organization founded by renowned investor and entrepreneur Azeem Azhar, may have provided an answer in its latest "State of the AI Economy" report.

The report shows that the actual annualized commercial revenue of the global generative AI industry, excluding China, has reached about $175 billion. After deduplication, in the first quarter of this year, the AI industry’s quarterly revenue surpassed infrastructure depreciation costs for the first time. This means that after two years of explosive investment, the AI industry is finally beginning to cover the rapidly expanding infrastructure investment with real customer income.

However, this does not mean AI investment has entered the "harvest period." The report believes that whether this trillion-dollar AI infrastructure investment can deliver returns ultimately depends on whether falling AI costs will continue to stimulate demand, creating enough token consumption and business revenue as prices keep dropping.

In a Bloomberg TV interview, Azeem Azhar said that in the past, the supply side of the market was "almost fully visible," but the demand side "remained shrouded in fog," and his team released the report to clarify the real context of the AI economy.

The Truth About AI Demand: Annualized Revenue Reaches $175 Billion

Over the past year, the supply side of the AI industry has been precisely quantified—Nvidia GPU shipments, capital expenditures by Microsoft and Meta, and global data center construction progress all have mature data systems for ongoing tracking.

The real blind spot is on the demand side. Leading AI companies such as OpenAI and Anthropic are not publicly listed, and cloud providers like Microsoft, Google, and Amazon never disclose AI business revenue separately. The market has never been able to answer a key question: how many businesses and consumers are actually paying for AI?

Over six months, Exponential View broke down the public disclosures, financial data, industry chain information, and cloud procurement records of more than 1,000 companies. By eliminating double counting in the value chain, they built an independent, bottom-up revenue estimation model.

According to their estimate, as of June 2026, the actual annualized revenue of the global generative AI industry (excluding China) has reached about $175 billion, with about $110 billion of real revenue generated in the past 12 months.

AI Revenue Covers Depreciation for the First Time, but Break-Even Is Still Far Off

For capital markets, the most closely watched data in this report is that AI revenue has crossed an important threshold for the first time.

By the first quarter of 2026, quarterly AI industry revenue exceeded AI infrastructure depreciation costs for the first time. This means current AI business cash flow can already cover the accounting depreciation of servers, GPUs, and data centers.

However, in the context of the entire investment cycle, real payback is still a long way off. The report estimates that by the end of 2026, cumulative AI-related capital expenditure by global hyperscale cloud vendors and emerging AI cloud platforms will reach about $2 trillion, with AI-related capital spending rising $535 billion above the original trend.

Meanwhile, AI infrastructure annual depreciation is expected to approach $111 billion in 2026. While current quarterly revenue can now cover depreciation, cumulative revenue still does not yet cover historic accumulated depreciation pressure from capital investments.

In other words, the AI industry has crossed the first threshold of being able to "feed itself," but proving that the entire capital cycle can deliver reasonable returns is still some way off.

AI Growth Rate Is Over Three Times That of the Internet

AI is moving from a technological revolution into the stage of commercial realization, and this process is occurring much faster than previous IT platform shifts.

The report shows that generative AI revenue is still growing at about 200% year-over-year—a rate about three times faster than any historical IT platform upgrade. The overall growth trajectory now exceeds that of the internet, cloud computing, and smartphones at early stages. Based on revenue growth curves, in 2023 it took about 180 days for the AI industry to add $1 billion in cumulative revenue; now, this process takes less than two days.

The surging revenue is driven by booming inference demand, pushing the entire computing industry into a new supercycle. The report shows that since 1971, global computing power has maintained a compound growth rate of about 66%, but in the AI era this has climbed to 80%. Meanwhile, U.S. electricity demand, dormant for more than a decade, is growing again, and the size of large AI data centers has expanded nearly 50-fold in four years.

The AI wave is also reshaping data center cost structure. The report predicts that chip costs as a proportion of data center costs will rise from about 40% in 2021 to 60% in 2026. The biggest change is not in GPUs but high-end memory such as HBM, whose share has jumped from about 2% to 18%, making it an important element of AI infrastructure investment.

In the interview, Azhar said his team originally expected AI revenue growth to gradually slow this year, but the actual result far exceeded expectations. "We originally thought growth would begin to cool off, but Anthropic’s explosive growth kept industry-wide revenue near a 200% year-over-year pace."

The Real Key to AI Investment Returns Is Price

The report notes that the biggest variable for the AI industry in the coming years is not leaps in model capabilities but whether sustained demand can be unleashed as prices fall.

With continuous improvements in model performance, ongoing inference efficiency gains, and significant increases in GPU utilization, the cost per million tokens called has plummeted from about $17 in 2023 to around $2 now. At the same time, token consumption is growing exponentially—up nearly 14-fold year-over-year.

Google, OpenAI, and other leading companies have noticed a similar pattern: every 10% drop in token prices typically leads to a 12–18% increase in demand, meaning demand elasticity already exceeds the rate of price decline. For this reason, the AI industry is moving towards a pattern similar to internet advertising. Just as Google’s cost-per-click model fueled the digital ad ecosystem, token-based pricing is becoming the new unit of value in the AI era.

The cheaper the model gets, the more use cases arise and the larger the market grows—lower cost may become the true catalyst for industry expansion.

AI Value Is Shifting to the Application Layer

Another structural change worth noting: AI industry profits are migrating downstream.

Since 2025, application layer revenue growth has significantly outpaced that of the model layer and cloud infrastructure. Over the past year, the share of application layer in overall AI revenue has risen from about 7% to 11%, while the model layer’s share has edged down from 11% to about 9%, and cloud infrastructure has dropped below 80% from about 82%. This shows commercial value is concentrating toward the application end.

But the report also points out there is still a limited window for premium pricing for cutting-edge models. With open-source models quickly catching up, even the most advanced large models are rapidly commoditized after release. In the future, to maintain profitability, AI labs will not only need to continually launch new frontier models, but also extend business into vertical domains like legal and programming, rather than merely relying on API charges.

Risk DisclaimerThe market involves risk; invest cautiously. This article does not constitute personal investment advice and does not take into account individual users’ specific investment goals, financial situation, or needs. Users should assess whether any opinions, views, or conclusions in this article suit their individual circumstances. Investments made based on this information are at your own risk. ```