What is a megawatt of AI compute?
Translate power into potential compute capacity.
← Back to Compute College
Compute College
Power Sourcing for AI: PPAs and Behind-the-Meter
AI data centers secure electricity through PPAs, behind-the-meter generation, and firm low-carbon sources like nuclear and SMRs.
AI power sourcing definition
Power sourcing is how an AI data center secures the electricity it needs. Because AI compute is power-hungry and grid capacity is constrained, operators increasingly sign long-term power purchase agreements (PPAs), build behind-the-meter generation (power produced on-site, bypassing the public grid connection), and explore firm low-carbon sources including nuclear and small modular reactors (SMRs).
Memory trick: For AI, chips wait on watts — how power is sourced often sets when capacity actually arrives.
Why it matters
Why power sourcing gates AI capacity
Power, not just chips, is now a binding constraint on AI capacity. How an operator sources power affects how fast it can bring capacity online, its cost and carbon profile, and whether it depends on a congested grid-interconnection queue.
- A PPA is a long-term contract to buy electricity at agreed terms, giving operators price certainty and developers financing.
- Behind-the-meter power is generated on-site and used directly, bypassing the public grid connection and its queue.
- Firm low-carbon sources, including nuclear and SMRs, are being explored to meet round-the-clock AI demand.
Simple example
Suppose two identical data-center plans differ only in power: one waits in a multi-year grid-interconnection queue, while the other adds behind-the-meter generation plus a PPA for firm supply. The second can energize capacity sooner, even if its power costs more per megawatt-hour — turning a power decision into a time-to-capacity decision.
- Power sourcing often decides time-to-capacity, not just cost per megawatt-hour.
- A PPA trades price certainty for a long commitment; behind-the-meter trades capital for grid independence.
- Any cost or timeline here is illustrative; real terms vary by site, region, and source.
Example figures are illustrative calculations, not current quoted market prices.
Market signal
How to read the market signal
Power-sourcing announcements — PPAs, behind-the-meter builds, and nuclear or SMR deals — are leading signals of where and how fast AI capacity can grow. Read them alongside interconnection-queue length: operators that secure firm power independently can move faster than those waiting on the grid.
- A PPA or behind-the-meter deal can signal capacity that bypasses the interconnection queue.
- Firm low-carbon commitments signal round-the-clock demand and a siting strategy.
- Power deals are a leading indicator for data-center buildout timing.
Market read: how an operator sources power is a leading signal of capacity timing and cost. Evidence discipline: separate announced PPAs and reactor plans from energized, delivering capacity, and date every power claim. Figures here are illustrative unless explicitly sourced and dated — see our methodology.
Common mistake
Treating a power or reactor announcement as delivered capacity. A signed PPA, a planned behind-the-meter plant, or an SMR agreement is a commitment, not megawatts on the floor — permitting, construction, and interconnection still gate real supply.
Practical takeaway
What you can do with this
Read power-sourcing news as a timing-and-cost signal for AI capacity, separating announced deals from energized supply.
- Buyers: ask providers how their capacity is powered and whether it depends on the grid queue.
- Analysts: track PPAs, behind-the-meter builds, and nuclear or SMR deals as leading capacity indicators.
- Separate announced power deals from energized, delivering capacity before counting supply.
- Note that firm low-carbon power (nuclear, SMR) is being pursued for round-the-clock AI loads.
- Keep announced timelines separate from observed energization.
Decision check: a power announcement matters for supply only after the capacity is permitted, built, and energized.
Compute College
Turn the lesson into a number
Use the GPU-Hour Cost Calculator, AI Training Cost Calculator, or Model Serving Cost Calculator.
Compute College track
Power & Data Centers
Step 13 of 17: Power sourcing for AI