Theory of Flames: environmental impact of AI decor and LED screens
A greenhouse-gas inventory of the AI decor of the opera Theory of Flames (Dutch National Opera & Ballet, 2025–2026 season): electricity for AI content creation, LED screens in operation, inbound transport and the materials embodied in the LED hardware, calculated in five reporting scenarios with eco-costs as a supplementary indicator. The calculator behind the study is available as an open, anonymised model for the performing arts sector.
What the study did
Dutch National Opera & Ballet measures the environmental impact of new productions. For Theory of Flames, an opera in which AI-generated images and video on LED screens form the decor, the opera house asked for a substantiated estimate of the impact of that decor. The central research question: what is the environmental impact of the decor of Theory of Flames, in kg CO₂e, including AI computing and LED screens (production and use), and which measures reduce that impact?
The study is a greenhouse-gas inventory of the current production. The unit of analysis runs from September 2025 to March 2026 and covers AI content creation by DoubleA, LED decor build and testing, rehearsals and 5 performances. The final calculation with scenario analysis and improvement priorities is reporting model v2.0 of April 2026. The public edition of the report and the open calculator date from September 2026.
Scope and method
The system boundary is the current production (current-production boundary). The inventory is informed by the GHG Protocol Product Life Cycle Accounting and Reporting Standard, without a formal conformance claim: it is not a full cradle-to-grave product inventory. No offsets, credits or avoided emissions are included. Eco-costs are reported separately as a supplementary single-indicator screening; the circularity bridge follows the wording of Theatre Green Book v2 Basic and applies only to the on-stage LED asset.
Unit of analysis and system boundary: the current modeled Theory of Flames AI decor production boundary for Sep 2025 – Mar 2026, covering AI content creation, LED decor build/testing, rehearsals and 5 performances. This model includes in Scenario B the re-use of the LED-decor across the following confirmed rehearsals and performances of the production Theory of Flames, but excludes the re-use of the LED-decor for new productions beyond Theory of Flames.
Clarification on allocation: Scenarios A, B and D attribute 20 per cent of the durable LED hardware to this production as a shared-use proxy, equal to one of five expected production-equivalent uses of the asset over its reference service life. No electricity, transport or operation of any production beyond Theory of Flames is included in the inventory. The fully dedicated single-production case is reported separately as Scenario C.
The model keeps three layers apart: operational electricity use, purchased transport and embodied hardware and materials. Scope 1 is zero (no direct combustion modelled); Scope 2 is the electricity for LED operation in the venue; Scope 3 covers the external electricity for AI content creation, the LED hardware as a capital good, packaging and transport. In scenario A the split Scope 1 / 2 / 3 is 0.0 / 288.5 / 7,526.0 kg CO₂e.
| Component | What is calculated | Indicator |
|---|---|---|
| Operational electricity – AI content creation | kWh of the local AI workstations × the applied electricity factor | CO₂ and eco-cost |
| Operational electricity – LED operation | Power per m² at the brightness used × area × active hours per phase, plus standby | CO₂ and eco-cost |
| Purchased transport – inbound new LED asset | Counted only when the purchase status is ‘new purchase’ | CO₂ and eco-cost |
| Embodied materials – packaging and durable LED hardware | Proxy per m² until a supplier PCF or EPD is available; allocation of the hardware to this production | CO₂ and eco-cost |
| AI local hardware and cloud or API services | Lower-bound proxies | Eco-cost only |
| Circularity and end of life | Previous and future life of the LED asset; bridge to Theatre Green Book v2 Basic | Status (meets / does not meet) |
Source: report, ‘Method basis and comparison restrictions’. Eco-costs follow the eco-costs method (ecocostsvalue.com): a prevention-cost indicator in euros, not project spend.
The five scenarios
The model calculates the same decor in five reporting scenarios. Two settings make the difference: the applied electricity factor (grid electricity 0.258 kg CO₂e/kWh or renewable electricity 0.000 kg CO₂e/kWh, applied to all project electricity) and the allocation of the newly purchased LED asset to this production (20% for shared use, 100% for a dedicated asset, 0% for existing stock). Scenario D adds a 20% embodied-materials completeness adjustment as a conservative sensitivity case. The LED reference service life is 7.5 years in all scenarios.
| Scenario | Description | Total CO₂ (kg CO₂e) | Eco-cost (€) | Theatre Green Book Basic (LED asset) |
|---|---|---|---|---|
| A | Grid electricity / shared-use new LED hardware asset (20% allocated) | 7,814.5 | 1,843.6 | Does not meet – previous life below threshold |
| B | Renewable electricity / shared-use new LED hardware asset (20% allocated) | 7,373.7 | 1,739.2 | Does not meet – previous life below threshold |
| C | Renewable electricity / dedicated new LED hardware asset (100% allocated) | 30,597.6 | 7,976.8 | Does not meet – previous life and future life below threshold |
| D | Renewable electricity / shared-use new LED hardware asset + embodied-materials completeness adjustment (+20%) | 8,549.0 | 2,052.3 | Does not meet – previous life below threshold |
| E | Renewable electricity / existing reusable LED hardware asset (0% allocated) | 0.0 | 16.9 | Potentially meets thresholds for on-stage LED asset; evidence incomplete |
Source: report, table ‘Scenario comparison overview’ (reporting model v2.0). CO₂ in kg CO₂e; eco-cost in € as an environmental indicator, not project spend. The scenarios are comparable only with each other: same activity base, same unit of analysis, same model structure.
Key results
The hotspot is the hardware, not the AI. In all new-purchase scenarios the LED hardware itself (embodied materials) is the dominant item: 5,806.0 kg CO₂e in scenarios A, B and D (74.3%, 78.7% and 67.9% of the scenario total) and 29,029.8 kg CO₂e in scenario C (94.9%). Inbound transport of the new asset weighs 1,497.6 kg CO₂e (19.2% in A). Electricity for AI content creation weighs 152.2 kg CO₂e (1.9%) in scenario A and LED operation 288.5 kg CO₂e (3.7%).
| Component | Scenario A: kg CO₂e (share) | Scenario B: kg CO₂e (share) |
|---|---|---|
| Operational electricity use – AI content creation | 152.2 (1.9%) | 0.0 (0.0%) |
| Operational electricity use – LED operation | 288.5 (3.7%) | 0.0 (0.0%) |
| Purchased transport – inbound new LED asset | 1,497.6 (19.2%) | 1,497.6 (20.3%) |
| Embodied materials – LED purchase packaging | 70.2 (0.9%) | 70.2 (1.0%) |
| Embodied materials – durable LED hardware asset | 5,806.0 (74.3%) | 5,806.0 (78.7%) |
| Total | 7,814.5 | 7,373.7 |
Source: report, ‘CO₂ emissions breakdown’ tables for scenarios A and B.
Renewable electricity helps, but does not remove the materials hotspot. Scenario B isolates the electricity-factor change versus scenario A: −440.7 kg CO₂e (−5.6%) and −104.4 eco-cost. The hotspot remains the LED hardware; renewable electricity does not neutralise the critical-material question of LED decor.
Allocation drives the spread. When the asset is attributed entirely to one production (scenario C), the total rises to 30,597.6 kg CO₂e and 7,976.8 eco-cost. In the Theatre Green Book bridge this case fails both the previous-life and the future-life criteria. Scenario D shows what a conservative completeness adjustment does: +1,175.2 kg CO₂e versus B, shown as a separate line so the base component values remain visible.
Model zero is not impact-free. Scenario E (existing, reusable asset) scores 0.0 kg CO₂e within this boundary because no new-purchase burden is attributed; eco-cost remains 16.9 because electricity, AI local hardware and cloud or API services still carry burden. If a newly purchased asset is used more often, the per-production burden falls through allocation; that logic is closer to scenario B than to E.
No comparison with a non-electronic decor. The study does not model a non-AI counterfactual. Whether a craft-oriented, more durable, repairable or transferable set could have delivered the same artistic service remains a design discussion, not a quantified result.
Source: report, ‘Primary decision messages’, the interpretation per scenario and ‘Craft-oriented, durable, repairable and transferable alternatives’.
Electricity in perspective
The electricity note in the report places the model values next to household references. One active hour of the LED decor uses 4.10 kWh; one AI machine-hour 1.25 kWh; with both AI workstations active at the same time, one clock hour is about 2.50 kWh. One performance of 1.75 hours uses 7.18 kWh and the 5 performances together 35.88 kWh. One active LED hour therefore sits above one AI machine-hour and above one clock hour with both AI workstations active. For reference: 1 kWh equals 16.7 hours of a 60 W incandescent bulb.
| Use case | Time | Electricity (kWh) |
|---|---|---|
| Laptop, normal work | 1 hour | 0.005–0.020 |
| Laptop, heavy work / local AI / rendering | 1 hour | 0.030–0.070 |
| Electric kettle, 0.5 litre | 3 min | 0.10 |
| Vacuum cleaner | 10 min | 0.15 |
| Electric oven | 30 min | 2.00 |
| AI content creation for the decor | 1 machine-hour | 1.25 |
| Both AI workstations at the same time | 1 clock hour | 2.50 |
| LED decor in build, testing or rehearsal | 1 active LED hour | 4.10 |
| One performance | 1.75 hours | 7.18 |
| 5 performances in total | 5 × 1.75 hours | 35.88 |
Source: report, ‘Electricity note’. The comparison is directional, not metered; the rehearsal row uses the active LED-hour rate as a proxy. Across the whole unit of analysis the reporting model counts 590 kWh for AI content creation and 1,118.3 kWh for LED operation.
Critical raw materials
The LED panels contain critical raw materials. The model screens five material families qualitatively (traffic light), cross-checked against European Commission and JRC sources on critical and strategic raw materials. Outcome: 2 red, 3 orange. A quantified, mass-based assessment is not yet possible while trace masses, sourcing region and recovery route are missing.
| Material family | Status | Interpretation | Next step |
|---|---|---|---|
| Gallium | Red | Functional semiconductor input in the LED chip layer; exact trace mass and sourcing are unknown in the current model. | Ask supplier for chip chemistry, trace mass, sourcing region and recovery route. |
| Rare earth elements | Red | Rare-earth family is present, but exact phosphor chemistry and mass share are not documented. | Ask supplier which REEs are used in phosphors and in what trace quantities. |
| Copper | Orange | Likely meaningful mass share and potentially recoverable, but recycled content and mass split are not evidenced. | Request copper mass share, recycled content and formal recovery route. |
| Aluminium / alumina family | Orange | Large mass family with strong embodied relevance; circularity and recycled content matter more than trace presence alone. | Confirm alloy mass, recycled content and durable re-use / recycling route. |
| Indium | Orange | Trace display material with relevance for LED and electronics chains; current EU criticality status is weaker than gallium and rare earths. | Ask supplier for ITO trace mass and coating chemistry. |
Source: report, table ‘Critical materials screening’. More on the approach: Critical raw materials.
Limitations and terms of use
The report carries the release gate Conditional – internal use only / not audit-ready and the fit-for-use status Directional / internal decision support, with a weighted QA score of 0.766 on a scale of 0 to 1 and 1 blocking, 2 major and 2 medium open data points. The QA score weighs the completeness and evidence quality of the inputs; the open flags relate to the first three improvement priorities below. The public edition is shared for learning in the performing arts sector, not as an audit-ready external disclosure.
In concrete terms: the embodied LED layer is proxy-based because no supplier PCF or EPD and no reviewed bill of materials with masses is available; AI local hardware and cloud or API services are lower-bound proxies; the Theatre Green Book bridge is limited to the on-stage LED asset and is not a full production-level assessment; the scenarios are comparable only within this set and this unit of analysis. The study makes no broader environmental superiority claim.
The same applies to anyone using the calculator: results are directional for design, procurement and re-use decisions. External claims require stronger supplier and provider evidence.
Source: report, ‘Governance and evidence context’ and ‘Limitations, allowed use and next improvements’.
Six improvement priorities
Within the unit of analysis, the first priority is better supplier evidence for the LED hardware, followed by stronger AI hardware and cloud evidence and a documented re-use or WEEE route.
| Priority | Topic | Main data still needed |
|---|---|---|
| 1 | Supplier PCF / EPD / BOM | Supplier PCF or EPD, or a panel-level bill of materials with material masses. |
| 2 | AI hardware and cloud evidence | Job logs, provider region, machine lifetimes, utilisation and PUE evidence. |
| 3 | Circularity and end-of-life route | Formal re-use plan, return route, WEEE or recycling route and any refurbishment assumptions. |
| 4 | LED operational electricity metering | Measured power and active hours by phase. |
| 5 | Critical material trace masses | Trace masses, sourcing region and recovery evidence for gallium, indium, rare earths, copper and the aluminium family. |
| 6 | AI provider and machine choice | Provider region, model type, machine class, datacentre efficiency and the quality of the renewable-electricity evidence. |
Source: report, table ‘Limitations, allowed use and next improvements’ (without the owner column).
Origin and licence
Developed by New Economy for Dutch National Opera & Ballet for the production Theory of Flames (season 2025–2026). The AI content was created by DoubleA. The study used the GHG Protocol Product Standard as guidance (Scope 1/2/3), Theatre Green Book v2 Basic for the circularity bridge, European Commission and JRC sources for critical raw materials and the eco-costs method of ecocostsvalue.com. The public edition is based on the final reporting model v2.0 (April 2026); personal data, contact details and supplier details have been left out, while the calculation logic and the A–E scenario structure are unchanged.
Report: © New Economy, commissioned by Dutch National Opera & Ballet. Sharing and citing with attribution permitted. The calculator behind the study is available as an open model: AI decor and LED calculator: calculate the CO₂ impact of a digital set.
Report and calculator
The report is in English; the Dutch project page summarises the key points in Dutch.
Insights from the study
Source-bound insights from the greenhouse-gas inventory of the AI decor: scenario figures, hotspots, eco-costs and critical raw materials.
Frequently asked questions
In the main reference scenario (B: renewable electricity, shared use of a new LED asset) the decor comes to 7,373.7 kg CO₂e and 1,739.2 eco-cost. With grid electricity (scenario A) the total is 7,814.5 kg CO₂e; when the asset is attributed entirely to one production (scenario C) it is 30,597.6 kg CO₂e. An existing, reusable asset (scenario E) counts 0.0 kg CO₂e and 16.9 eco-cost within this boundary. Source: report, table ‘Scenario comparison overview’.
Because the materials in the LED asset weigh more than the electricity use. In scenario A the durable LED hardware is 5,806.0 kg CO₂e (74.3% of the total) and inbound transport 1,497.6 kg CO₂e (19.2%), against 152.2 kg CO₂e (1.9%) for AI content-creation electricity and 288.5 kg CO₂e (3.7%) for LED operation. In scenario C the hardware share rises to 94.9%. Source: report, ‘CO₂ emissions breakdown’ tables for scenarios A and C.
Scenario B isolates the electricity-factor change versus scenario A: total CO₂ falls by 440.7 kg CO₂e (−5.6%) and eco-cost by 104.4. The hotspot remains the LED hardware; a better electricity factor does not remove the materials question. Source: report, ‘Primary decision messages’ and the scenario B interpretation.
It is a model result, not an impact-free reality. In scenario E the LED asset is already present, so no new-purchase burden (hardware, packaging, inbound transport) is attributed within this boundary. Eco-cost remains 16.9 because electricity, AI local hardware and cloud or API services still carry burden. If a newly purchased asset is used more often, the per-production burden falls through allocation; that logic is closer to scenario B than to E. Source: report, scenario E interpretation.
Eco-costs are a supplementary environmental indicator in euros: the prevention cost of avoiding the environmental burden, not project spend and not a carbon footprint. The report splits them into human health, nature, resource scarcity and carbon; in scenario B respectively 51.3 / 175.3 / 537.1 / 975.4 of a total of 1,739.2. Renewable electricity still carries eco-cost burden while its CO₂ factor is 0.000 kg CO₂e/kWh. Source: report, ‘Reading guide’ and scenario B headline results.
The screening puts gallium and rare earth elements on red and copper, aluminium/alumina and indium on orange. For all five, trace masses, sourcing region and recovery route are still missing; the first step is to request those data from the supplier. Source: report, table ‘Critical materials screening’.
The report carries the fit-for-use status ‘Directional / internal decision support’ and the release gate ‘Conditional – internal use only / not audit-ready’, with a weighted QA score of 0.766 on a scale of 0 to 1. The embodied LED layer is proxy-based because no supplier PCF or EPD is available, AI hardware and cloud are lower-bound proxies and the scenarios are only comparable with each other. The study makes no broader environmental superiority claim. Source: report, ‘Governance and evidence context’ and ‘Limitations, allowed use and next improvements’.
Companies, producers, theatres and set designers who want to estimate the CO₂ impact of a digital set. The calculator is the anonymised version of the reporting model behind this study, is shared with the performing arts sector on the initiative of Dutch National Opera & Ballet and is free to use under Creative Commons Attribution 4.0 (CC BY 4.0). The workbook is an Excel file in English, with a bilingual explanation on the first tab.
Frequently asked questions about the report and its scope
No. The report is written in English; the Dutch project page summarises the key points in Dutch. The calculator has a bilingual first tab (Dutch and English).
The unit of analysis runs from September 2025 to March 2026 and covers AI content creation, LED decor build and testing, rehearsals and 5 performances.
The system boundary is the current production: what was made, purchased and used for this decor. It is not a full cradle-to-grave product inventory, which is why the report makes no formal conformance claim to the GHG Protocol Product Standard.
No. Offsets, credits and avoided emissions are not included. The figures show attributed emissions and eco-costs within the chosen boundary.
The bridge to Theatre Green Book v2 Basic tests only the on-stage LED asset for a previous life and a future life. A newly purchased asset has no previous life in the bridge and therefore does not meet the threshold; an existing asset (scenario E) potentially meets it, but the evidence is incomplete. It is not an assessment of the whole production.
Yes, with attribution: © New Economy, commissioned by Dutch National Opera & Ballet. The report is directional and intended for scenario comparison and decision support, not as an audit-ready disclosure.
Related footprint and materials work
Model structure, formulas, texts and questionnaire: New Economy, Creative Commons Attribution 4.0 (CC BY 4.0). Idemat 2026 factor values: © Sustainable Impact Metrics Foundation / TU Delft, CC BY-NC (not for commercial LCA or CSRD software), source ecocostsvalue.com. Attribution: ‘New Economy, AI decor and LED impact model v2.0 (CC BY 4.0)’. Case data: with the consent of Dutch National Opera & Ballet, as an example only. Report: © New Economy, commissioned by Dutch National Opera & Ballet; sharing and citing with attribution permitted.
Questions about the study or a calculation of a production
Questions about the study, the report or a calculation for a production go to New Economy.
Email New Economy
