ONE AI BOOM · THREE DIFFERENT PRICE REGIMES

Physical Scarcity Rises. Digital Intelligence Gets Cheaper.

The investment mistake is treating chips, model APIs and electricity as one trade. Their supply curves move at radically different speeds.

HARDWARE · CYCLICAL BOTTLENECK

Scarce now

$75.2B data-center revenue

Nvidia's fiscal Q1 2027 data-center sales rose 92% year over year as accelerated-computing demand outran near-term supply.

Watch: lead times, HBM and packaging capacity, cloud rental prices and gross margins.

MODEL APIS · COMPETITIVE DEFLATION

Price down

$0.10 input · $0.40 output

Google lists Gemini 2.5 Flash-Lite at these prices per million tokens. Efficiency and competition keep pushing the floor lower.

Watch: price cuts, benchmark-adjusted cost, customer switching and application gross margins.

ELECTRICITY · PHYSICAL BOTTLENECK

Demand up

4.4% → as much as 12%

U.S. data centers' share of electricity use could rise from 2023's level to the top of Berkeley Lab's 2028 scenario range.

Watch: capacity auctions, interconnection queues, rate cases, transformers and new generation.

The model gets cheaper. The bottleneck gets paid.Value migrates toward layers whose supply cannot expand as quickly as AI demand.

Sources: Nvidia Q1 FY2027 results; Google Gemini API pricing; U.S. Department of Energy and Lawrence Berkeley National Laboratory. Prices and projections are snapshots, not forecasts of investment returns.