AI revenue will need to reach $1.5 trillion to $2 trillion a year by the end of 2028 to justify the money going into data centers, according to an analysis from Apollo. That is roughly 5 to 7 times today’s level. By 2030, the figure could climb to $3 trillion to $5 trillion, depending on how much more capacity gets built.
A $3.5 trillion buildout
Hyperscalers, chipmakers and AI developers have committed large sums to expand computing capacity. Apollo estimates cumulative capital spending could reach about $3.5 trillion through 2028, based on announced plans.
So far the signals are favorable. Demand for compute still exceeds supply, GPU rental prices are rising, and some infrastructure providers report capital payback periods under a year. The open question is how much revenue the whole chain has to bring in to earn a return.
The $20 billion per gigawatt rule
Apollo starts with one gigawatt of AI computing capacity. Running it, including depreciation and operating costs, runs about $7 billion a year. Hyperscalers, model builders and application companies each need a margin on top, at least over the long run.
Stacking those margins, each gigawatt needs about $20 billion in annual end-user revenue. With roughly 100 GW of AI capacity deployed by the end of 2028, the total lands in the $1.5 trillion to $2 trillion range.
The assumptions matter. Apollo uses $30 billion of IT spending per gigawatt and a six-year useful life, which roughly matches Blackwell-generation chips. Vera Rubin chips would push annual depreciation and operating costs closer to $9 billion per gigawatt, raising the revenue each gigawatt must earn. It also applies a 50% gross margin to the app layer, which covers the roughly 25% of applications served by standalone app companies. Hyperscalers and model builders deliver the rest directly.
Where revenue stands now
Most of the money is expected to come from businesses and consumers paying for AI products, including subscriptions and enterprise software. Annualized AI application revenue was about $200 billion in the second quarter of 2026, up from $11 billion in early 2024. By September it had reached roughly $276 billion, around 3.75 times annual growth since 2023.
To hit the 2028 target, revenue would need to grow 2.3 to 2.6 times a year for the next two years. Recent growth rates make that look achievable. Still, each doubling is harder when the base is already large.
What could break the math
The model assumes spending arrives on schedule. Several things could slow it:
- Power constraints and chip shortages
- Construction delays
- Restrictive policy after the midterm elections
Capital markets provide a check too. If revenue disappoints, lower valuations and higher financing costs would likely slow spending, which would reduce the revenue needed to justify it. Apollo sees that feedback loop as a brake on excess capacity and on a bubble.
It also points to a risk flagged in Anthropic’s IPO filing: slower revenue growth could constrain investment in compute and model development, which would weaken a company’s competitive position.
Why the 5 to 7 times figure matters
The target is a multiple of today’s base, not a forecast. Reaching it means AI has to spread into much larger markets and show measurable economic value across industries. Subscriptions from consumers help, but the bulk of the required AI revenue has to come from businesses that save money or earn more by using the tools.
The exact figure also moves with the deployment total. Fewer gigawatts means a lower bar, and a higher margin requirement at any layer means a higher one.
What to track through 2028
Three numbers decide whether the gap closes: the pace of AI revenue growth against the 2.3 to 2.6 times annual hurdle, the gigawatts actually switched on, and the cost per gigawatt as Vera Rubin ships. If power and chips slow deployment, the required revenue shrinks with it. If deployment holds and revenue lags, the pressure falls on the companies that financed the capacity.
For now, the figures show the bet is large but not out of reach on the current growth trend. Investors will test the 5 to 7 times target one quarter at a time.
Source: apollo.com