Key Takeaways:
Global AI revenue from hyperscale cloud providers has exceeded data center depreciation costs for a second straight quarter, a milestone that bolsters the economic case for the industry's $200B-plus annual infrastructure buildout.
Global AI revenue from hyperscale and emerging cloud providers reached $25 billion in the most recent quarter, surpassing estimated depreciation costs of $21 billion tied to data centers and chips, according to a report Wednesday from research firm Exponential View. The gap marks the second consecutive quarter that AI-generated revenue has covered the capital-intensive depreciation burden, a threshold the firm said signals the industry is approaching financial sustainability.
"The data shows that AI revenue is no longer just a promise — it's covering real costs," said Azeem Azhar, founder of Exponential View, in the report. "This doesn't mean every dollar spent is profitable, but it does mean the aggregate economics are starting to work."
The $25 billion in revenue came from a mix of hyperscale providers — Amazon Web Services, Microsoft Azure, Google Cloud — and emerging cloud services that offer AI-specific compute. Depreciation costs of $21 billion reflect the rapid amortization of graphics processing units from Nvidia and Advanced Micro Devices, along with data center construction expenses that have pushed combined capital spending at the four largest US hyperscalers past $230 billion annually. The 19% revenue cushion over depreciation is thin but represents a marked improvement from early 2025, when AI revenue consistently fell short of depreciation by 10% to 15%.
Why the breakeven matters for investors
The milestone addresses a central question that has hung over the AI trade since the launch of ChatGPT in late 2022: whether the massive upfront spending on GPUs and data centers would ever generate sufficient returns. Microsoft, Amazon, Google and Meta Platforms have collectively committed more than $200 billion in annual AI-related capital expenditures, with Wall Street analysts at Goldman Sachs and Morgan Stanley warning earlier this year that a "productivity payoff" was needed by mid-2026 to sustain investor confidence.
Exponential View's data suggests that payoff is beginning to materialize, albeit with narrow margins. The $4 billion surplus — revenue minus depreciation — represents roughly 2% of the hyperscalers' combined annual AI CapEx, meaning profitability from AI operations remains minimal even as top-line growth accelerates. Nvidia, whose data center revenue hit $36 billion in its most recent quarter, remains the primary beneficiary of the buildout, while companies such as AMD, Broadcom and Marvell Technology are competing for secondary supply positions.
What comes next
The sustainability of the trend hinges on whether AI revenue growth can outpace the depreciation curve. Hyperscalers are expected to continue raising CapEx through 2027 as they build out next-generation data centers equipped with Nvidia's Blackwell Ultra and Rubin architectures, as well as in-house chips from Amazon's Trainium and Google's TPU families. If AI revenue maintains its current trajectory, the cushion over depreciation could widen to 30% or more by late 2027, Exponential View projected.
For investors, the report provides a data-driven counterweight to concerns that AI infrastructure spending is a bubble. Microsoft shares trade at 33x forward earnings, Amazon at 38x and Google at 24x — multiples that embed expectations of a multiyear AI-driven revenue acceleration. The breakeven milestone does not eliminate downside risk, but it shifts the burden of proof onto skeptics who argue the spending is unjustified.
This article is for informational purposes only and does not constitute investment advice.
