The rise of artificial intelligence is significantly contributing to increased energy consumption in data centers. According to the International Energy Agency (IEA), this consumption could escalate from 485 TWh in 2025 to 950 TWh by 2030. In response to the growing energy demands and associated emissions, a new market segment known as "Regenerative AI" is emerging, which aims to offset the ecological impact of AI operations.
One such player in this space is the cross-platform AI assistant EcoGPT. This startup has devised a system whereby it funds the planting of one tree for a specified number of queries made by users.
ForkLog reached out to EcoGPT's developers for comments on two occasions but had not received a response by the time of writing. Based on publicly available technical documentation and scientific data, we examined the project's architecture to understand its claimed climate model.
Environmental Mission as a Business Model
The service is developed by ECOGPT, INC., a company incorporated in Delaware. According to the official Delaware Division of Corporations, it was established on February 24, 2026, as a Benefit Corporation, with registration number 10522699. Interestingly, the EcoGPT app lists an address in Miami Beach, Florida, in app stores.
Source: Delaware Division of Corporations.EcoGPT was founded by Ethan Leonard, who left college to create a "more efficient AI". The startup is currently self-funded.
Being classified as a Benefit Corporation allows EcoGPT to integrate its public benefit goals alongside shareholder interests into its corporate structure.
In its basic free version, the company funds the planting of one tree for every 200 queries generated, while the paid Pro subscription offers one tree for every 100 queries.
The official website's counter indicates that over 263,000 trees have been planted so far.
Source: EcoGPT website.In the App Store, the application has a rating of 4.8 based on 8,000 reviews, ranking 45th in the "Productivity" category. On Google Play, with 1 million downloads claimed, it maintains an average rating of 3.7, although some users report issues with lag during generation and poor response quality from the models.
According to a version of the EcoGPT for Business page archived on August 22, 2026, the company was developing a separate offering for organizations, including corporate access to the service, an API, and a client-branded program. This page also displayed a live dashboard showing the number of queries, active users, and trees planted. However, as of writing, the original page redirects users to the main EcoGPT interface, leaving the current status of this offering unclear.
Limitations of the "Green" Architecture and Hidden Carbon Debt
EcoGPT's primary ecological focus is on enhancing the efficiency of its AI systems and infrastructure. The company does not disclose the specific solutions underpinning its service, only stating that it employs small open-source models with active parameters "more than ten times fewer than leading AI companies".
Source: EcoGPT website.Previous third-party reviews described EcoGPT's technical stack as utilizing models from families such as Llama, DeepSeek, Qwen, and Mistral, operated on Groq LPU infrastructure.
As per the current Terms of Service, EcoGPT's primary dedicated infrastructure is located in Google Cloud data centers in Hamina, Finland. According to Google's data, its infrastructure in Finland achieved a 98% carbon-free energy rate in 2023, with seawater used for cooling the data center.
However, the ecological impact of such infrastructure is not limited to inference energy consumption. Training large language models requires substantial resources; for instance, the energy cost for training GPT-3 was estimated at about 1.3 GWh, with water consumption during training reaching 5.4 million liters in some calculations.
Source: Devera Team research.Training is just one facet of an AI's ecological footprint; with widespread use, the total energy consumption for inference becomes significant and, according to some estimates, may even dominate the overall impact.
Alex Petrov, co-founder of Hyperfusion, argues that labeling EcoGPT as wholly "green" would be a marketing exaggeration.
"The project merely shifts the ecological debt onto the developers of foundational models," he commented to ForkLog.
Petrov emphasized that the impact of such a service should be evaluated considering the entire lifecycle of model creation and use, rather than just the energy costs of individual queries.
Determining the carbon and water footprint of a single API request is impossible without detailed data from the provider's infrastructure, according to Petrov. In such cases, one can only construct an estimate based on average coefficients and assumptions; tools like CodeCarbon and Green Algorithms utilize calculation methodologies rather than direct measurement of each computation.
An additional concern is the efficiency of architectures and equipment. Petrov described MoE architectures and specialized chips as "fundamental optimizations" regarding inference but warned that reducing costs per operation does not necessarily lead to a decrease in overall energy consumption within the industry.
"As soon as the cost and energy consumption per token decrease, the industry tends to generate significantly more tokens — this is known as the Jevons Paradox," he noted.
Overall, Petrov does not view EcoGPT's approach as an engineering innovation; rather, he sees it as a product that promotes the use of smaller models where larger systems are excessive.
"From a machine learning engineering perspective, it lacks new patents, groundbreaking algorithms, or architectural solutions — it’s more about marketing," he concluded.
Another aspect of the infrastructure is that when the main system is unavailable, EcoGPT utilizes OpenRouter, which routes requests to model providers. In such cases, execution may take place on different platforms, depending on the chosen provider. Public documents do not allow for establishing the full physical route of each request, including the equipment and source of electricity.
Carbon Math: Why Trees Don't Offset Servers
EcoGPT supports reforestation efforts in Kenya, Tanzania, and Uganda through the nonprofit organization Trees for the Future. The fund confirms its collaboration with EcoGPT and issued a certificate on September 7, 2026, for 100,000 trees planted.
Source: EcoGPT website.However, EcoGPT's ecological model of "200 queries = 1 tree" faces a challenge due to the time lag between energy consumption and carbon absorption. Energy is consumed during prompt processing, while trees gradually remove CO₂ from the atmosphere.
According to an article published in Nature Climate Change involving the German Research Centre for Geosciences, naturally regenerating secondary forests in most studied ecoregions reach their maximum carbon removal rate at around 20-40 years of age. While young forests do absorb CO₂, their average carbon accumulation rate is lower; the specific age at which this peak occurs varies by region and forest type.
The mortality rate of newly planted trees in the initial years can also be high. Research in East Africa shows significant variability in survival rates; for example, in one experiment in arid regions of Kenya, seedling survival rates varied from 36.1% to 47%, while a project aimed at restoring agroforestry lands in northeastern Uganda reported a 75% survival rate of 14,500 seedlings by project completion. Factors influencing survival include watering, planting and care quality, drought, pests, and livestock damage.
Mikhail Yulkin, founder and CEO of CarbonLab, commented to ForkLog that simply counting the number of trees planted isn't sufficient to assess the project's impact. He emphasizes that it is far more crucial for companies to reduce their own carbon footprint during infrastructure operation — particularly by selecting data centers and infrastructure partners that utilize low-carbon energy and monitor their environmental impact.
Yulkin described tree planting as an ineffective primary means of offsetting AI impact:
"Tree planting is not a solution to the problem at all. It’s merely an imitation of a solution."
He also noted that different types of environmental impact cannot be automatically offset against one another. For instance, tree planting in one region does not compensate for a data center's local impact on water resources in another.
Who Gets the Credit for Environmental Impact?
Trees for the Future, in addition to its charitable programs, is developing a separate carbon market initiative: its projects involve measuring, monitoring, and independently verifying climate effects, followed by issuing Verified Carbon Units under Verra standards. The fund also offers corporate clients options for regular project financing and future carbon unit purchases.
Meanwhile, Trees for the Future lists EcoGPT as a corporate sponsor financing tree planting, without mentioning carbon credits. The Terms of Service of the EcoGPT service also do not state that these plantings correspond to a specific volume of certified carbon units.
According to Yulkin, a corporate client of EcoGPT can report that part of the fees paid for the service goes towards an environmental project, but such decisions cannot be automatically represented as reducing or offsetting their own emissions.
"They chose a responsible supplier, and that’s the end of it. But they did not reduce anything themselves," he added.
Additionally, if multiple parties claim the same climate outcome, there is a risk of double counting. Rights to such outcomes must be clearly defined and documented in advance.
Yulkin pointed out that issuing carbon units is not mandatory for simple funding of plantings. What is more critical is independent verification and regular instrumental monitoring of the project, including through remote sensing.
"A reliable system of instrumental control over the entire array of trees they planted is needed. Otherwise, it’s just words," he concluded.
Payments and Transparency of Environmental Expenses
EcoGPT's public materials do not mention blockchain, tokens, or on-chain accounting for environmental impact. The service's paid version operates on a subscription model: payments are processed via Stripe on the website and through Apple and Google Play in mobile applications. However, the company does not disclose what percentage of revenue is allocated to tree planting or the methodology for calculating that amount.
Denis Smirnov, a leading researcher at Basilisk Labs and a blockchain and Web3 consultant, argues that integrating blockchain or ReFi protocols does not inherently protect projects from greenwashing. Such tools could enhance transparency in the flow of funds and environmental assets, and a Proof-of-Planting concept could serve as an additional audit tool.
"If we record in the blockchain 'a thousand trees have been planted', it does not prove that a thousand trees were actually planted," the expert told ForkLog.
He stated that a proprietary token for such a model is also not necessary: the company can accept regular payments and then publicly demonstrate what portion of revenue was directed to environmental projects and what carbon assets were acquired and retired.
"Blockchain in such a scheme is primarily needed for record transparency, while a token only makes sense if there is a separate function that cannot be realized without it," Smirnov added.
Privacy and Third-Party Providers
EcoGPT claims it does not use chats to train foundational models and prefers AI providers with a Zero Data Retention policy.
Source: EcoGPT website.However, in official documents, the company explicitly states that routing through OpenRouter does not guarantee zero data retention: third-party providers may store requests for service delivery or security purposes.
Moreover, in EcoGPT, the history of conversations is accessible to users until they delete it themselves. Data deletion starts immediately and is completed across all systems within 30 days. When using OpenRouter, requests are sent to third-party model providers, so the handling rules for these requests also depend on those companies' terms.
Smirnov clarified to ForkLog that OpenRouter offers a separate ZDR mode, where requests can be directed only to providers claiming zero data retention. However, he noted that even this scheme relies on policies and agreements rather than technical impossibilities to store information.
"Once a request has gone to someone else's server, we ultimately have to trust the owner to some extent," Smirnov remarked.
No Breakthrough in Sight
EcoGPT presents a straightforward scheme: reduce computational load using smaller models while linking service use to funding reforestation. However, public materials do not fully clarify how the ecological impact is calculated, what portion of funds is directed to planting, and what climate result one such "green" generation ultimately yields.
Experts consulted pointed out several limitations of this model. The efficiency of inference does not negate the resources expended in developing and training models, and the sheer number of trees planted does not determine the volume of CO₂ removed. Transparent reporting, independent verification, monitoring of plantings, and clear definitions of who has the right to claim such results are crucial for climate claims.
In terms of technology, EcoGPT employs existing approaches — small models, efficient infrastructure, and third-party computing power. Therefore, at this stage, the project appears more as a product model that connects an AI service with an environmental contribution rather than an innovative "green" technology. Its future climate value will largely depend on how transparently the company can substantiate its environmental claims.