AI and Economics

Will enterprises use the cloud providers' AI, or build their own?

The cloud providers' expansion thesis assumes enterprises will use cloud AI, but data-sensitive enterprises (IP, defense, regulated) tend to build their own local AI to prevent data outflow. Public cloud primary inference 56%→41%, private cloud 56%. The expansion thesis is right for 'most enterprises' and wrong for 'data-sensitive enterprises'.

Where does the cost AI saves go?

The cost AI saves does not vanish; it moves to the price of shared memory components, then to consumer electronics: phones 13% more expensive and PCs 17% more expensive in 2026. This is a regressive transfer — those who don't get AI's benefit pay most for AI's side effect.

What has AI's current state actually helped with?

AI's current state really solves one thing: software. In this low-verification-cost domain, AI's effect is measurable and real, but the magnitude depends on task complexity. Companies that use AI for routine software work gain; the unit value of software engineers' routine work is squeezed.

What funds the AI investment loop?

The AI investment loop is funded by B2B/infrastructure capture (memory rent, NVIDIA pricing power, hyperscaler cloud revenue), not B2C. It is currently a bet: hyperscalers put 31–83% of revenue into capex and need $1.5 trillion in new debt over 2025–2027. If buyers don't see ROI, they cut capex and the loop breaks.

The real bottleneck is not GPUs, it is electricity

AI is in a crazy expansion, but the real bottleneck is electricity. GPUs are available (NVIDIA has capacity), but grid interconnection takes 5 years and transformers take 5 years. 2026: 12 GW announced, only 5 GW actually under construction (about one-third). The crazy expansion is partly a paper boom.