AI, semiconductors, robotics & biotech: the most capital-intensive investment cycle in history

AI capex has become an infrastructure cycle of utility-like scale , comparable to railroads or the electrical grid , with a readable value chain: semiconductors (the cluster of megacaps that the market now groups under acronyms like “MANBRIC”: Microsoft, Apple, Nvidia, Broadcom, and others), data centers, energy, and now physical robotics. Biotech remains the pocket of relative value in the complex. Facts. AI capex: ~$765B in 2026 in the base case, ~$1,600B/year by 2031, i.e., ~$7,600B cumulative for 2026-2031 (Goldman Sachs, May 2026). Global data center spending exceeds $1,000B in 2026, up 78% year-over year in Q1 for the four largest hyperscalers (Dell’Oro, July 2026); the five hyperscalers exceed $600B in capex, with ~75% AI focused (~1.9% of U.S. GDP , more than the Apollo program). The stated bottleneck is no longer silicon but electrical grid connection capacity. Humanoid robotics: Tesla has deployed 1,000+ Optimus Gen 3 internally; Figure AI (Helix VLA) operates at BMW and a second Fortune 500 client; 2026 valuations: Figure ~$39B, Apptronik ~$5.5B, 1X ~$10B; projected market: $38B by 2035 (Goldman Sachs), $152B by 2040 (Morgan Stanley). The dominant model is emerging as RaaS (Robotics as a Service): recurring revenue rather than hardware sales. Analysis. Three takeaways for an allocator. (1) The cycle’s return migrates downstream: after foundries and GPUs, the next winners are energy (PPAs, grids), cooling, and RaaS integrators. (2) The financing question becomes central: these assets (data centers, robot fleets) are contractual cash flow assets , natural candidates for securitization and then tokenization (cf. idea 3, part 4). (3) The dispersion in humanoid valuations (×30 among private leaders) signals a pre-standard market: position via “picks and shovels” suppliers rather than betting on a single integrator. Risks. Overcapacity if application monetization disappoints (forecasts rely on assumptions about silicon lifespan and token demand); tension on DRAM/HBM memory (server price increases); electrical constraints delaying commissioning; in humanoids, the gap between demos and profitable deployments remains real (prediction markets: ~21% probability of a consumer robot in 2026). Implication. Favor infrastructure (energy, cooling, connectivity) over final,stage integrators; monitor the capex/revenue ratio of hyperscalers (45-57%, industrial level) as a warning indicator; in robotics, wait for proof of profitable RaaS before deploying growth capital.