{"id": 1119130, "name": "Cost (inflation-adjusted)", "unit": "constant 2023 US$", "createdAt": "2025-10-28T12:01:40.000Z", "updatedAt": "2025-10-28T12:01:40.000Z", "coverage": "", "timespan": "24441-27853", "datasetId": 7254, "shortUnit": "$", "columnOrder": 0, "shortName": "cost__inflation_adjusted", "catalogPath": "grapher/artificial_intelligence/2025-10-22/epoch_compute_cost/epoch_compute_cost#cost__inflation_adjusted", "descriptionShort": "The cost of developing the AI system expressed in 2023 US dollars, adjusted for inflation.", "descriptionFromProducer": "The cost of developing the AI system was adjusted for inflation using the Consumer Price Index (CPI) for the United States. The CPI was used to account for the changes in the price level of consumer goods and services over time. The inflation-adjusted cost is presented in 2023 US dollars.", "type": "float", "datasetName": "The rising costs of training frontier AI models", "datasetVersion": "2025-10-22", "nonRedistributable": false, "display": {"name": "Cost", "unit": "constant 2023 US$", "shortUnit": "$", "numDecimalPlaces": 0}, "schemaVersion": 2, "processingLevel": "minor", "presentation": {"titlePublic": "Hardware and energy cost to train notable AI systems", "topicTagsLinks": ["Artificial Intelligence"]}, "descriptionKey": ["Hardware costs account for depreciation. The cost reflects only the portion of hardware value used during training, not the full purchase price, since hardware loses value as newer chips become available.", "Training costs include both hardware expenses and electricity consumed during training, plus data center cooling and power distribution.", "These figures show only the final training run. Total development costs are typically higher because they include many experimental runs before the final model.", "Exact training times are often unknown, so costs are estimated using \"chip-hours\"\u2014the number of chips multiplied by hours of training."], "dimensions": {"years": {"values": [{"id": 24497}, {"id": 25128}, {"id": 25127}, {"id": 25869}, {"id": 27298}, {"id": 27219}, {"id": 27456}, {"id": 27564}, {"id": 24447}, {"id": 27276}, {"id": 27634}, {"id": 26644}, {"id": 24740}, {"id": 26080}, {"id": 26994}, {"id": 27101}, {"id": 27337}, {"id": 27816}, {"id": 27526}, {"id": 27367}, {"id": 27438}, {"id": 26639}, {"id": 27618}, {"id": 27806}, {"id": 26550}, {"id": 24441}, {"id": 27353}, {"id": 25027}, {"id": 26520}, {"id": 27597}, {"id": 27853}, {"id": 25959}, {"id": 25826}, {"id": 26581}, {"id": 27449}, {"id": 24780}, {"id": 27558}, {"id": 25913}, {"id": 26756}, {"id": 27157}, {"id": 25323}, {"id": 25748}, {"id": 26308}, {"id": 25862}, {"id": 24751}, {"id": 26582}, {"id": 26225}]}, "entities": {"values": [{"id": 257029, "name": "AlphaGo Lee", "code": null}, {"id": 257033, "name": "AlphaGo Master", "code": null}, {"id": 257039, "name": "AlphaGo Zero", "code": null}, {"id": 257060, "name": "AlphaStar", "code": null}, {"id": 370131, "name": "Amazon Titan", "code": null}, {"id": 367637, "name": "Claude 2", "code": null}, {"id": 369330, "name": "Claude 3 Opus", "code": null}, {"id": 370244, "name": "Claude 3.5 Sonnet", "code": null}, {"id": 365389, "name": "DeepSpeech2 (English)", "code": null}, {"id": 369508, "name": "Falcon-180B", "code": null}, {"id": 371480, "name": "GLM-4-Plus", "code": null}, {"id": 368065, "name": "GLaM", "code": null}, {"id": 257030, "name": "GNMT", "code": null}, {"id": 354864, "name": "GPT-3 175B (davinci)", "code": null}, {"id": 367084, "name": "GPT-3.5", "code": null}, {"id": 363052, "name": "GPT-4", "code": null}, {"id": 369003, "name": "GPT-4 Turbo", "code": null}, {"id": 371369, "name": "GPT-4.5", "code": null}, {"id": 369519, "name": "GPT-4o", "code": null}, {"id": 369506, "name": "Gemini 1.0 Ultra", "code": null}, {"id": 369515, "name": "Gemini 1.5 Pro", "code": null}, {"id": 367255, "name": "Gopher (280B)", "code": null}, {"id": 371231, "name": "Grok-2", "code": null}, {"id": 371366, "name": "Grok-3", "code": null}, {"id": 369563, "name": "HyperCLOVA 204B", "code": null}, {"id": 306064, "name": "Inception v3", "code": null}, {"id": 368751, "name": "Inflection-2", "code": null}, {"id": 257037, "name": "JFT", "code": null}, {"id": 257090, "name": "Jurassic-1-Jumbo", "code": null}, {"id": 370155, "name": "Llama 3.1-405B", "code": null}, {"id": 371514, "name": "Llama 4 Behemoth (preview)", "code": null}, {"id": 257064, "name": "Meena", "code": null}, {"id": 257055, "name": "Megatron-BERT", "code": null}, {"id": 368027, "name": "Megatron-LM (8.3B)", "code": null}, {"id": 257092, "name": "Megatron-Turing NLG 530B", "code": null}, {"id": 369880, "name": "Mistral Large", "code": null}, {"id": 257031, "name": "NASv3 (CIFAR-10)", "code": null}, {"id": 369982, "name": "Nemotron-4 340B", "code": null}, {"id": 257061, "name": "OpenAI Five", "code": null}, {"id": 273167, "name": "PaLM (540B)", "code": null}, {"id": 365387, "name": "PaLM 2", "code": null}, {"id": 306104, "name": "ResNeXt-101 32x48d", "code": null}, {"id": 365388, "name": "RoBERTa Large", "code": null}, {"id": 257078, "name": "Switch", "code": null}, {"id": 257059, "name": "T5-11B", "code": null}, {"id": 240142, "name": "Xception", "code": null}, {"id": 257093, "name": "Yuan 1.0", "code": null}, {"id": 368329, "name": "mT5-XXL", "code": null}]}}, "origins": [{"id": 8930, "title": "The rising costs of training frontier AI models", "descriptionSnapshot": "This data is specifically from Figure 1 of the paper.", "description": "This dataset provides a comprehensive analysis of the costs associated with training frontier AI models, with a focus on estimating the magnitude and growth of these expenses. It includes detailed cost components for various AI models, considering hardware, energy, cloud rental, and staff expenses. The dataset is based on a detailed cost model developed to fill the gap in public data on AI training costs. The data covers the period from 2016 to the projected costs for 2027.", "producer": "Epoch AI", "citationFull": "Ben Cottier, Robi Rahman, Loredana Fattorini, Nestor Maslej, and David Owen. 'The rising costs of training frontier AI models'. ArXiv [cs.CY], 2024. arXiv. https://arxiv.org/abs/2405.21015.", "versionProducer": "2025-10-22", "urlMain": "https://epoch.ai/blog/how-much-does-it-cost-to-train-frontier-ai-models", "dateAccessed": "2025-10-22", "datePublished": "2025-10-22", "license": {"name": "CC BY 4.0"}}]}