A Google DeepMind leader just threw out a staggering number. AI-related capital spending will hit $1 trillion by 2026. The math checks out when you look at what Alphabet, Microsoft, Amazon, and Meta have already committed. Their combined capex guidance adds up to roughly $725 billion for next year alone, a 77% jump from the $410 billion they're expected to spend this year. Alphabet itself raised its forecast to between $195 billion and $205 billion.

But the real driver isn't just more chips or faster servers. It's recursive self-improvement, or RSI. This is where an AI system becomes capable of optimizing itself without waiting for engineers to tweak the dials. The DeepMind exec sees this as the actual path to AGI, not just incremental scaling of existing models.

The trillion-dollar thesis

Goldman Sachs layered its own analysis on top of these corporate commitments and pegged annual AI capex at around $765 billion for 2026. The investment bank's cumulative forecast stretches to $7.6 trillion from 2026 through 2031. That's not hyperbole. It's the largest capital allocation cycle in technology history, and it's already reshaping how infrastructure companies think about power, cooling, and networking.

RSI sits alongside multi-agent systems and traditional scaling as a pathway toward Artificial Superintelligence, or ASI. That's intelligence that surpasses human capability across every domain, not just matching it. DeepMind CEO Demis Hassabis flagged RSI as a key focus for the lab's development roadmap.

Data centers powering AI are already competing with Bitcoin miners for power capacity in places like Texas and the Nordic countries. A 77% year-over-year increase in hyperscaler spending means more facilities being built, more demand for cooling infrastructure, and more strain on regional power grids. For crypto investors, that $7.6 trillion cumulative capex forecast creates sustained demand for the raw inputs of computation: energy, semiconductors, and networking hardware. The infrastructure play, not necessarily the AI models themselves, is where the structural demand sits.

This article is informational and should not be construed as financial or investment advice.