Facts-only context brief. This is source analysis, not a live energy-price feed, a physical-watt measurement, an investment recommendation, or an energy-market product.
The reporting question
AI systems purchase more than model tokens. Their operating economics also depend on power availability, grid interconnection, cooling, utilization, and—in many electricity markets—the fuels that set marginal generation cost.
That makes energy a material reporting context for CTW (Cost per Token per Watt). It does not make every energy scenario a routing instruction or a tradable market.
SKENAI's current Energy–CTW slice is deliberately limited:
- Measurement tier:
inferred_heuristic - Mode: report only
- Live power-price data: unavailable
- Physical node watts: unavailable for this public brief
- Settlement, XP, model routing, and prediction-market use: excluded
What the sources say
1. AI economics increasingly encounter infrastructure limits
SemiAnalysis' AI Tokenomics Model examines AI economics through infrastructure, utilization, and operating-cost constraints. Its relevance to CTW is directional: token economics are not separable from the infrastructure required to produce tokens.
This brief does not convert that analysis into a SKENAI cost estimate. It establishes a reporting discipline: cost-per-token narratives should say whether their energy component is directly measured, estimated, or absent.
2. Grid interconnection can constrain compute deployment
SemiAnalysis' US Grid Constraints: Towards 40GW discusses the difficulty of supplying and interconnecting very large data-center loads. Transmission, transformers, generation, and local interconnection capacity can all become deployment constraints.
The capacity figures in that report are an external scenario, not a SKENAI forecast. The practical reporting implication is narrower: geography and grid access can change the cost and availability conditions under which compute operates.
3. Fuel scenarios add uncertainty, not an oracle
Patrick O'Shaughnessy's July 21 post relays Matthew Smith's scenario of a US natural-gas shortage beginning around 2028. The same thread includes direct disagreement from energy-industry participants, including assertions that US gas resources remain abundant.
That disagreement is the point. A public social post—even from a sophisticated analyst—is not an electricity-price oracle. SKENAI records it as contested scenario context, not as input truth.
The CTW boundary
| Layer | Current status |
|---|---|
| Model cost and estimated watts | Heuristic CTW risk telemetry |
| Public grid and fuel research | Facts-only reporting context |
| Real-time LMP or utility tariff | Not integrated |
| Physical energy telemetry | Not used in this public brief |
| Compute routing or settlement | Not driven by Energy–CTW context |
| Energy prediction market or trading | Not built |
The distinction protects the system from a common category error: a well-cited macro thesis can motivate measurement work, but it cannot settle a price, dispatch a workload, or back an economic claim by itself.
What would change the tier
Energy could become a stronger CTW input only with a separately auditable source contract: a licensed public price feed or trusted meter, explicit geography and timestamp, freshness limits, unit normalization, outage behavior, and a test proving that unavailable data fails closed.
Until then, this is the honest posture: energy belongs in the CTW conversation as provenance-rich context, not as a hidden oracle.
Sources are linked above. This brief distinguishes source claims, competing scenarios, and SKENAI's current implementation boundary.