According to Bloomberg on June 25, citing an analysis of AI spending data from over 1,000 companies by research firm Exponential View, global AI revenue (excluding China) reached $25 billion in the first quarter of 2026, surpassing for the first time the depreciation and amortization cost of $21 billion for the same period. However, $850 billion in infrastructure commitments still await a return.
The Vulnerability of the Six-Year Depreciation Assumption: If GPU Lifespan Shortens, the $21 Billion Baseline Will Rise
Tech and cloud providers currently spread the cost of AI chips and other equipment over a useful life of roughly six years, which directly determines the quarterly depreciation figure. Exponential View's $21 billion depreciation baseline is entirely based on this six-year assumption.
If the actual lifespan of GPU clusters is shorter than six years — for example, if next-generation chips deliver a performance leap that renders existing equipment obsolete sooner — depreciation accelerates, pushing the $21 billion baseline higher, and the $25 billion quarterly revenue would fall back from "surpassing" to "chasing."
The article notes that the Jalapeno AI chip, jointly developed by OpenAI and Broadcom, claims to reduce costs by approximately 50% compared to existing GPU solutions. It is expected to be deployed in data centers of partners like Microsoft later this year. Such supply-side cost competition is just beginning.
Potential Impact of Low-Cost Models Like DeepSeek on AI Service Pricing
On the revenue side, some users have already begun shifting to cheaper or even free Chinese models like DeepSeek. Once enterprises migrate en masse to low-cost models, hyperscale cloud providers will be forced to follow suit in lowering AI service unit prices. Even if user numbers continue to grow, revenue per user may be diluted simultaneously, making it difficult to maintain the newly crossed depreciation line.
$850 Billion in Infrastructure Commitments vs. $25 Billion in Quarterly Revenue
Bloomberg data for the same period shows that Meta has added $79 billion in new data center lease commitments, Microsoft $41 billion, and the cumulative future data center lease obligations for the entire cloud industry have reached $850 billion.
$850 billion in infrastructure commitments corresponds to $25 billion in single-quarter revenue. Just the depreciation line alone requires sustained exceeding for several years before this construction wave enters a true payback period. The article concludes: "Crossing the depreciation line is a fact, but whether it marks the beginning of a new era or is a self-justifying temporary number for this construction surge will likely be answered by data in the coming quarters."
Common Questions
What does "depreciation and amortization cost" mean, and why is this comparison meaningful?
Depreciation and amortization is an accounting method that spreads large capital expenditures (such as purchasing GPUs) over their useful life as periodic expenses. Comparing depreciation costs rather than actual purchase amounts is more meaningful because it reflects the periodic "consumption" of capital and is the standard way companies assess whether investments are beginning to generate returns. AI quarterly revenue exceeding depreciation costs means that, from an accounting perspective, the AI business is starting to "cover" the cost amortization of deployed infrastructure.
Is the six-year depreciation assumption reasonable?
According to the article, six years is the current common practice among tech and cloud providers for depreciating AI equipment, making it an industry convention. However, AI hardware evolves extremely fast. If next-generation chips significantly outperform current GPUs within three to four years, the actual effective lifespan of existing equipment may be shorter than six years, resulting in higher actual depreciation costs. Therefore, the six-year assumption is both the current industry standard and the biggest uncertainty variable in the analysis.
How does OpenAI's Jalapeno chip affect this equation?
According to the article, Jalapeno is an in-house AI chip jointly developed by OpenAI and Broadcom, claiming to reduce costs by approximately 50% compared to existing GPU solutions. It is expected to enter the data centers of partners like Microsoft later this year. If more efficient and lower-cost chips are widely deployed, on one hand, they may lower the future depreciation baseline (benefiting the revenue side); on the other hand, they may accelerate the premature obsolescence of existing GPUs, increasing short-term depreciation pressure.
