What supports the chain. Energy, water, fabs, materials, skilled trades, the inputs the entire stack consumes.
Why it matters
Before chips, before data centers, before models, there's power, water, foundries, rare earths, and the humans who build it all. When AI demand explodes, the bottleneck isn't algorithms. It's megawatts and skilled trades.
The Ground Itself, Land, Power, Materials
Before the gold rush, you need land, water rights, ore deposits, and the miners who work the seams. In AI: power generation, cooling water, foundry capacity, rare earths, and the electricians and technicians who physically build the boom. When demand spikes, this layer is the real bottleneck.
The 5 sublayers
L-1a
Energy & Grid Interconnect
Power generation, PPAs, transmission, and the multi-year grid-interconnect queue, the megawatts the stack consumes and the wait to get them switched on
L-1b
Thermal & Water Management
Cooling systems, water access, immersion/liquid cooling, heat reuse, the thermodynamic ceiling on every GPU cluster
L-1c
Fabrication & Foundry
Leading-edge chip fabrication capacity, EUV lithography, advanced packaging (CoWoS), the physical floor of L0
L-1d
Critical Materials & Supply Chain
Rare earths, lithium, cobalt, gallium, specialty substrates, and the refining, logistics, and geopolitical chokepoints that gate them
L-1e
Skilled Trades & Human Capital
Electricians, HVAC techs, data-center builders, fab process engineers, robotics technicians, the labor pool no model can synthesize
, Layer diagnostic card · SCOI v1
Is a company really at L-1?
Physical inputs the entire AI stack consumes, power, water, fabs, materials, the trades that build them.
Inclusion tests · include if ALL
Owns or directly contracts physical capacity (megawatts, water rights, fab lines, ore deposits, trade crews).
Lead time to add supply is measured in years, not quarters.
Demand from the AI stack shows up as a line item in their P&L or order book.
Exclusion tests · exclude if ANY
Only consumes power/compute, does not produce it.
Sells software that schedules or monitors physical assets but does not own them.
Brokerage or marketplace with no take-or-pay exposure to capacity.
The L-1 removal test
Remove L-1 and the entire stack stops within hours. Nothing above L-1 can substitute on the timescale that matters.
Economic work this layer does
Converts capital and time into physical capacity that AI demand cannot manufacture on-demand.
Canonical examples
NextEra
Owns generation. Sells the megawatts every hyperscaler is now competing for.
TSMC
Owns leading-edge fab capacity. No L0 silicon ships without it.
Bechtel
Owns the trade workforce and EPC capability to actually build data centers.
Anti-examples · look-alikes that fail
Energy-monitoring SaaS
Measures power. Does not produce it. L7 dashboard on someone else's L-1.
Carbon-offset marketplace
Brokers credits. No physical capacity, no real bottleneck exposure.
DCIM software
Manages a data center. Owning the building is L0; the software is L6/L7.
