Energy

Where the power comes from

10 options, ordered by nothing — they are not ranked, because the right answer is entirely decided by when you need power and what you will accept. Lead time is the field to read first.

• Grid interconnection — 4-5 years typical, approaching 5 on average. Whatever the local grid mix is
• Nuclear PPA (existing plants) — 1-3 years to contract; the plant already exists. Effectively zero operational carbon
• Small modular reactors (SMR) — 2030s for meaningful capacity. Effectively zero operational carbon
• On-site natural gas turbines — 12-24 months, the fastest firm power available. High. This is the trade being made.
• Behind-the-meter generation — Depends on the generation, but it skips the queue. Whatever you build
• Solar + battery storage — 18-36 months. Near zero operational
• Wind PPA — 24-48 months. Near zero operational
• Fuel cells — 9-18 months, among the fastest on this list. Lower than turbines on natural gas; near zero on hydrogen, which is rarely available
• Enhanced geothermal — 3-5 years. Near zero operational
• Demand response and curtailable load — Months. Neutral to positive

How the heat leaves

The crossover where liquid cooling beats air sits near 30 kW a rack on ten-year cost and 45–50 kW on thermals. Above that it stops being a choice: NVIDIA has confirmed liquid cooling is a mandatory architectural requirement for GB200 NVL72.

• Air cooling — Up to ~30 kW per rack, PUE 1.4-1.6 typical
• Rear-door heat exchangers — ~30-50 kW per rack, PUE 1.2-1.4
• Direct-to-chip liquid cooling — 50-200+ kW per rack, PUE 1.1-1.2
• Single-phase immersion — 100+ kW per rack equivalent, PUE 1.02-1.10
• Two-phase immersion — Highest available, PUE Approaching 1.02
• Free and evaporative cooling — Depends on the paired system, PUE Can push annualised PUE below 1.2 in the right climate

The numbers that decide

• PUE — power usage effectiveness — Total facility power divided by IT power. 1.0 is perfect; every 0.1 above it is overhead you pay for and never
• WUE — water usage effectiveness — Litres of water per kilowatt-hour of IT load. The number local communities and regulators increasingly care ab
• Tokens per watt — The metric that actually matters in 2026: useful model output per unit of power. It joins the chip layer to th
• Rack power density — Kilowatts drawn by one rack. The number that decides your cooling method, your floor loading and whether an ex
• The interconnection queue — The waiting list to connect new load or generation to the grid. In 2026 it is the single most important number

Head to head

• Nuclear PPA (existing plants) vs On-site natural gas turbines — Clean and slow against fast and dirty. The defining 2026 trade, and most operators are quietly choosing gas.
• Small modular reactors (SMR) vs Grid interconnection — Two answers to the same question, both arriving in the 2030s.
• Direct-to-chip liquid cooling vs Single-phase immersion — The production method against the thermally superior one. Serviceability decides it, not physics.
• Air cooling vs Direct-to-chip liquid cooling — Where the crossover actually sits: about 30 kW a rack on cost, 45-50 kW on thermals.
• Solar + battery storage vs On-site natural gas turbines — Intermittent and clean against firm and fast. Almost always paired rather than chosen.
• Behind-the-meter generation vs Grid interconnection — Skip the queue and take the regulatory risk, or wait five years and take none.
• Fuel cells vs On-site natural gas turbines — The premium bridge against the default bridge.
• Enhanced geothermal vs Nuclear PPA (existing plants) — The two clean firm options. One is geology-limited, the other supply-limited.
• Two-phase immersion vs Single-phase immersion — Marginal thermal gain against a live PFAS regulatory question.
• Demand response and curtailable load vs Behind-the-meter generation — Two ways to get connected sooner: agree to use less, or avoid the wires entirely.