Returns are from AI are being seen. The smaller spending from 2 years ago sees a lag in returns but adjusting for lag shows profits.

Pascal's Wager Applied (CEOs Believe in Godlike AI)

Invest and AI Succeeds: Huge payoff—competitive advantage, innovation leadership, and returns on investment (e.g., breakthroughs in compute-intensive tasks like drug discovery or climate modeling). The cost of building data centers is dwarfed by the benefits.
Invest and AI Fails: Loss of capital—money spent on infrastructure, energy, and maintenance yields little return. The cost is significant but finite.
Don't Invest and AI Succeeds: Missed opportunity—competitors or other nations dominate, and you're left behind in a potentially world-altering technological shift. The regret and strategic disadvantage could be catastrophic.
Don't Invest and AI Fails: Minimal loss—you save resources and avoid a sunk cost, but there's no gain either.

The BG2 podcast (with semiconductor , AI analyst Dylan Patel) describe the AI investment.

This was an AI scaling and spending discussed by multi-billion fund managers Bill Gurley and Brad Gerster with top AI analyst Dylan Patel.

Free cashflows from Meta, Microsoft, Amazon, Google and other are going into AI data centers and AI research.

There is a discussion of capex spending of hyperscalers ( the capex investment part of the discussion of BG2).

Gurley's broader narrative aligns with analyses from sources like Morgan Stanley and Dell'Oro Group, which project hyperscaler CapEx reaching $300–$335 billion in 2025, largely driven by AI. For instance, Morgan Stanley's updated forecasts (reported February 20, 2025, via Datacenter Dynamics) estimate Amazon and Microsoft leading with $96.4 billion and $89.9 billion, respectively, in 2025 CapEx, with AI infrastructure as a primary driver. Gurley's discussions in prior years suggest he views this as a strategic necessity.