Achieving a 10% return on current AI investments could require $650 billion in annual revenue, or $35 from every iPhone user monthly.
The hyperscalers are set to spend $533 billion in capex this year as the race to build AI infrastructure ensues. With some estimates suggesting the AI buildout could eclipse $5 trillion over the next several years, investors are rightfully questioning whether this scale of investment can ultimately generate acceptable returns.
AI monetization is already underway, but it remains concentrated in the infrastructure layer. Elsewhere, end-user monetization is still early, uneven and opaque. This pattern is common for an early general-purpose technology, but the stakes are high in today’s world, and how monetization ultimately materializes across the value chain will have significant implications for markets.
The easiest place to look: Infrastructure
Capital spending on semiconductors, data centers, networking equipment and power infrastructure has surged as companies race to deploy AI at scale. Two forces are driving this scramble for compute:
- Rising adoption. 55% of American adults now use GenAI each week, an adoption rate that took the internet 16 years to achieve1, and 17% of U.S. firms report adoption.
- Intensifying workloads. Compute intensity per task is rising as models scale and become more complex, particularly with the rise of reasoning models.
Despite hiking capex 170% over the last two years, the hyperscalers still expect a very tight demand-supply environment through 20262. Of the 125 GW of data centers globally, only ~20 GW can currently handle AI workloads3, and at least 100 GW of new power generation will be needed to support projected data center demand4. These constraints pose real risks to companies servicing AI, but they also create unique opportunities across the infrastructure value chain.
The jagged frontier: Applications
Three years into generative AI, there are already unprecedented success stories. AI-native companies are reaching massive scale with lean, capital-efficient teams that would have been unthinkable a decade ago.
That said, the economics of AI services are still under construction. Many leading models lose money on each interaction, and while that’s not unusual for a new service, it's unclear how long it can persist. Subscription revenue has dominated so far, but advertising and alternative pricing models are likely to emerge.
Enterprise software companies are also testing a range of pricing strategies, from bundling AI into existing products to charging per token. Roughly 60% of firms expect to expand AI budgets significantly, but the open question is whether this represents incremental spend or substitution within IT budgets. Large firms already spend roughly $1,000-$3,500 per employee annually on software5, and productivity gains will ultimately determine how far spend can climb.
The tollkeepers: Hyperscalers
The hyperscalers sit at the intersection of infrastructure and applications, and many have already monetized AI indirectly through higher cloud demand6. But the return hurdles ahead are substantial. Achieving a 10% return on current AI investments could require $650 billion in annual revenue, or $35 from every iPhone user monthly7. While that may be feasible, it's a high bar that assumes sustained, broad-based adoption.
Finally, an important reminder for investors: even if the aggregate AI returns disappoint, individual winners can still do extremely well, particularly if winner-take-all dynamics emerge. And even if aggregate monetization exceeds expectations, there will almost certainly be losers. This is why diversification remains key—across companies, the AI value chain and asset classes—as portfolios benefit not just from exposure to innovation, but from balance.