Research: AI global revenue surpasses depreciation costs for the first time, but the niche is dangerously thin.

According to research firm Exponential View’s analysis of data from over 1,000 companies, global AI revenue (excluding China) reached $25 billion in Q1 2026, for the first time surpassing the $21 billion depreciation and amortization costs for the same period.
(Background recap: To prioritize AI, Apple is reportedly cutting M6 Pro/Max; high-end Macs will go straight to the “M7 generation”)
(Background supplement: Anthropic accuses Alibaba of launching the “largest cloning attack in history,” repeatedly scraping Claude 28.8 million times)

Research firm Exponential View built an AI spending dataset spanning over 1,000 companies, tracking end-to-end costs from chip procurement to model training, and then comparing them against AI-related revenue in the same period. Ultimately, it arrived at a conclusion: global AI revenue (excluding China) reached $25 billion in Q1 2026, for the first time surpassing the $21 billion depreciation and amortization costs for the same period.

According to Bloomberg’s analysis, after the tech industry has splurged on building data centers for hundreds of billions of dollars over the past few years, this is the first signal that “AI revenue may catch up to infrastructure costs.”

However, it also emphasizes that this line is both thin and fragile, because the entire analytical framework is built on the assumption that “the hardware depreciation period is six years.” Tech and cloud companies currently spread the costs of AI chips and other equipment over an approximate six-year usage cycle, which directly affects the quarterly depreciation expense figures.

In simple terms: if the actual lifespan of these GPU clusters is shorter than six years—for example, if the performance leap from next-generation chips causes existing devices to be retired early—then accelerated depreciation would push the $21 billion baseline higher, and the $25 billion revenue figure would immediately go from “surpassing” back to “falling short.”

DeepSeek appears on the other side of the equation

On the other hand, even if revenue has crossed the depreciation line for the first time, cost awareness is eating away at this margin.

Some users have started switching to cheaper, or even free, models such as China’s DeepSeek. Once enterprises shift on a large scale to low-cost models, hyperscale cloud providers will be forced to lower their AI service unit prices. Even if the number of users continues to grow, revenue contributed per user could also be diluted at the same time—making the depreciation line that has just been crossed difficult to sustain again.

Commitments of $850 billion

Bloomberg’s data for the same period shows that Meta and Microsoft alone, from newly added data center lease commitments, reached $79 billion and $41 billion, respectively. The total future data center lease obligation across the cloud industry has now accumulated to $850 billion.

The $850 billion infrastructure commitments correspond to $25 billion in quarterly revenue. Put another way, just to keep the depreciation line consistently exceeded for several years—so this construction boom can enter a true payback period. Let alone the fact that OpenAI recently partnered with Broadcom to launch a self-developed AI chip codenamed Jalapeno, which claims to save about 50% in costs compared with existing GPU solutions. It is expected to be integrated into Microsoft’s data centers later this year, and competition on the cost side from the supply side is only just beginning.

A $4 billion lead, supporting an answer that relies on a six-year assumption, corresponds to an infrastructure repayment cycle of more than a double-digit number of years. Crossing the depreciation line is a fact, but whether it is the starting point of a new era or just a temporary number that this construction wave uses to convince itself may need data from the next few quarters to find out.