AI Revenue Beats Depreciation for the First Time: Exponential View Says Data-Center Spend Starts to Pencil Out
A report from Exponential View says global AI revenue has hit a turning point: hyperscaler and neocloud AI sales reached $25B, beating an estimated $21B in data-center and chip depreciation for a second straight quarter. That means AI revenue is starting to cover capital-expenditure costs—though margins remain thin.
Bottom line
The biggest question hanging over AI—'can the hundreds of billions poured in ever be earned back?'—just got a mildly positive signal. Exponential View reports that global AI revenue exceeded the related depreciation cost for a second straight quarter, the first sign the AI business model may be 'economically sustainable'.
Key numbers
- Hyperscaler + neocloud global AI sales: $25B.
- Estimated depreciation tied to data-center and chip investment: $21B.
- $25B revenue > $21B depreciation—and it's the second consecutive quarter ahead.
Why depreciation is the key yardstick
Many fixate on AI revenue growth, but what really decides 'is this a viable business' is whether revenue can cover the amortization of massive capex:
- Building data centers and buying GPUs is enormous one-time capex;
- Accounting depreciates those assets over their lifetime, creating an ongoing cost;
- Only when AI revenue consistently exceeds that depreciation does the investment pencil out—otherwise it's 'burning cash for scale', hard to sustain.
The 'revenue vs depreciation' lens answers 'bubble or business' better than revenue alone.
But don't celebrate too soon
The report also flags: margins remain thin. Beating depreciation isn't high profit:
- Covering depreciation is just the 'break-even' bar—power, ops, staff, and R&D aren't yet counted;
- Fast chip iteration risks upending the depreciation assumptions (new GPUs devalue old assets faster);
- 'Two quarters' is short—whether it becomes a lasting trend is unproven.
What it means for different players
- Investors: a counterpoint to the 'AI bubble' thesis—at least at the cloud layer, revenue is starting to self-fund rather than rely purely on financing.
- Cloud providers: validates that the 'capital-heavy AI infrastructure' path is starting to work economically, though the profit inflection needs a thicker revenue cushion.
- Downstream companies: healthier underlying economics means more sustainable AI compute supply—good for long-term price and availability stability.
In a line: AI is moving from 'telling a story' to 'making the numbers work'—but it's still a way from 'making big money'.