Global AI Revenue Exceeds Data Center Depreciation Costs for Two Consecutive Quarters for the First Time
Alina Collins
Global AI sales hit $25 billion in Q1 2026, clearing the $21 billion depreciation line for a second consecutive quarter — but depreciation still consumes over two-thirds of that revenue, and the report warns the margin for error is "razor-thin."
What does "barely clearing the bar" actually mean?
AI revenue now exceeds the cost of wearing out the hardware that runs it — but only just: $25 billion vs. $21 billion in estimated depreciation.
This means → AI is no longer pure cash burn; it can cover its own equipment bill. But "covering depreciation" is a long way from true profitability — power, labor, and financing costs haven't been subtracted yet.
Founder Azeem Azhar put it plainly: "If you massively cleared that bar, it would mean you left too much opportunity on the table." In plain terms = at this stage, running close to the line is normal, not alarming.
Is the six-year depreciation assumption realistic?
The report assumes a six-year useful life for GPUs and IT equipment. Some investors call that too generous — chip innovation could render older hardware worthless faster.
Investor Michael Burry once described underestimating depreciation as "one of the most common frauds in modern times."
Yet the data tells a different story: Nvidia's H100 rental price still sits at roughly 80% of its launch rate, and over the past year H100 prices actually *rose* because the newer Blackwell chips couldn't ship fast enough to meet demand.
AWS CEO Matt Garman said in February that Nvidia A100 servers — six years old — are still in service because demand remains strong. This means → older chips are not losing value as quickly as skeptics expected.
Are users migrating to cheaper models?
Data from the OpenRouter platform shows Google, OpenAI, and Anthropic's combined share of token requests fell from 72% in June 2025 to 33% in June 2026.
Open-source and Chinese AI models — notably DeepSeek — are picking up the traffic.
In plain terms = power users are routing simple tasks to cheaper, faster models. That doesn't necessarily mean the leading foundation-model companies are in trouble, but it raises the bar for them to sustain premium pricing.
Can the $725 billion capex bet pay off?
Meta, Alphabet, Microsoft, and Amazon plan a combined $725 billion in AI-infrastructure capital spending this year.
This reflects a massive collective wager: that AI revenue growth can keep outrunning an unprecedented spending surge.
The report also warns that financing risk is shifting to capital markets through leases, debt, and equity — a trend especially pronounced among neoclouds, the new breed of cloud providers built specifically to rent out AI compute.
In plain terms = the AI industry just proved it can cover its depreciation bill, but that bill is ballooning fast. Whether revenue growth keeps pace with spending growth is the central question the market will test over the next several quarters.
Content is for reference only, not financial advice.