AI Infrastructure Spending Tops $3 Trillion as Investors Demand $3 Trillion Returns

The AI industry’s central question has sharpened into a $3 trillion problem as AI infrastructure spending races past $3 trillion and investors press for returns rather than promises.

“Skip to content Investing Up Massive YTD: Here Are Nvidia, Palantir & AMD’s Price Predictions For 2028”

24/7 Wall St.

TechCrunch traces the math to Sequoia partner David Cahn, who in 2023 started from Nvidia’s reported annual GPU revenue of $50 billion to argue that $200 billion in revenue would be required to pay back the upfront investment.

Bitcoin World

Fast-forwarding to 2026, Cahn’s updated estimate is that total AI infrastructure spending will reach $1.5 trillion and that the AI industry will have to earn $3 trillion to justify the chips and data center expenditures.

TechCrunch adds that the $3 trillion figure is “probably an underestimate” because rising costs of memory and the increasing use of exotic or inference-specific chips will drive the number up.

On the other side of the ledger, TechCrunch says Anthropic is thought to have hit $60 billion in ARR and OpenAI reportedly earned $13 billion in 2025 (and said it was at $20 billion ARR in November 2025), but the gap to justify the spending remains large.

Apollo Global Management chief economist Torsten Slok is warning that the payback from hyperscaler chip investments may be at risk if cash-flow targets slip, with the consequences extending beyond the tech sector.

In a research note highlighted by Bitcoin World, Slok points to the four major hyperscalers—Google, Meta, Microsoft, and Amazon—projecting massive accelerations in their free cash flow by 2028.

https

Bitcoin World frames the downside as organizations increasingly turning to cheaper open-weight AI models, often developed by Chinese companies, rather than relying on expensive frontier models built by leading labs.

The same source also ties the risk to falling token prices, citing that OpenAI’s latest model is “54% more token-efficient on coding tasks than its predecessor,” which could reduce the token volume needed to deliver results.

Slok’s warning, quoted by Bitcoin World, is that “a slower payoff wouldn’t just be a sector problem, it would risk tipping the economy into recession and the S&P 500 into a correction.”

As the ROI debate intensifies, Nvidia is moving to lock in demand by shifting from pure GPU sales toward financing structures that keep cloud providers tied to its broader stack.

“Nvidia has rolled out a new financing model targeting artificial intelligence (AI) cloud providers”

https

The West Asian report says Nvidia rolled out a new revenue-sharing financing model for AI cloud providers, extending beyond GPU sales by re-leasing unused GPU capacity from cloud operators at a predetermined price.

The report adds that in return Nvidia receives a share of cloud service revenue generated by those GPUs, and it cites SharonAI planning to deploy up to 40,000 Nvidia Grace Blackwell GB300 GPUs and Firmus planning to deploy up to 170,000 Nvidia GPUs in Batam, Indonesia.

The same article describes how Nvidia’s approach is meant to reinforce ecosystem lock-in once cloud providers build infrastructure around Nvidia’s architecture and GPUs, including networking, software, and operating environment.

Separately, TradingView reports that Nvidia stock fell 4% in early trading even as major tech companies announced a sharp rise in spending on AI infrastructure, underscoring investor caution about the sustainability of Nvidia’s dominance amid growing emphasis on in-house chip development.