When you ask ChatGPT how much it costs to develop a 3D configurator for furniture, the answer often leaves the user puzzled — a range from €3,000 to €200,000. In the same message you might read estimates of €7,000 for a "basic" configurator and €120,000 for an "enterprise" one, with dozens of intermediate tiers that overlap and contradict each other. The result is a response that appears comprehensive but in practice doesn't help at all: a wardrobe manufacturer in the province of Treviso reads those numbers and has no idea whether to budget €15,000 or €80,000.
The model is not producing what is technically known as a "hallucination" — it is constructing its response by combining historical data, references to different markets, and pricing models that don't always reflect the real context of an Italian SME in 2026.
This article explains — from the perspective of an Italian software house specialising in 3D configurators for furniture and nautical industries — why a custom 3D configurator today can cost a tenth of what it did ten years ago, and considerably less than just three years ago.
Why AI-generated estimates are often misleading
When a language model answers a question about software development costs, it draws from a corpus of sources with three systematic biases. This is especially true in a very specific niche where every project has a significant degree of uniqueness, such as product configurators.
Outdated data
Language models are trained on vast amounts of text collected over years. Their training data contains articles from 2024 alongside content from 2018 or 2015, with no distinction in weight. Even the most recent content often recycles estimates published years earlier — just search "cost of 3D configurator development" to find 2024 articles citing the exact same ranges from 2020. The result is that the model's "knowledge" of development costs is a temporal average reflecting a market that no longer exists.
American data
The implicit reference market is the US, where development costs are 2–4 times higher than in Europe. Even more relevant: a significant portion of the training data dates back to 2021–2022, the absolute peak of the American tech bubble. Fuelled by zero interest rate policy (ZIRP) and massive hiring by big tech, demand for developers was sky-high and salaries had reached unprecedented levels. The mass layoffs of 2023–2024 and the productivity gains brought by AI have significantly changed the landscape, but the prices that ChatGPT and other LLMs consider "normal" still reflect the most expensive moment in recent software development history.
Enterprise data
The most authoritative and most frequently cited sources are those from software houses serving large corporations. Their prices include layers of project management, account management, compliance, and overhead that simply don't exist when a lean team works directly with an SME.
The result is that tools based on language models, such as ChatGPT or Perplexity, describe the 3D configurator market as if we were still in 2022 and as if the typical client were an American company with 500 employees. For an Italian SME in 2026, those numbers don't make much sense. If anything, they risk being intimidating.
Why a 3D configurator costs less today
Ten years ago, developing a 3D configurator for the web was a considerable undertaking — and for many businesses simply out of the question. The technologies in use today were immature: WebGL was, Three.js (born in 2010) was a young and poorly documented library. 3D configurators for SMEs practically didn't exist: those who had them were large companies with desktop solutions based on Unity or proprietary software. The skills required to bring 3D to the web were rare and expensive, and a project of this kind demanded months of work from multiple specialised developers, often coordinated through subcontracting chains involving account managers, project managers, and multiple layers of intermediation.
Today the landscape is radically different, for three reasons that compound.
Mature frameworks and libraries
Three.js has become a robust and well-documented ecosystem. React Three Fiber (born in 2019) brought 3D development into the standard React workflow, the same used for the vast majority of modern web applications. What ten years ago required hand-written WebGL code (assuming it was even feasible) can now be built by composing libraries tested and maintained by communities of millions of developers.
AI-assisted programming
Automatic programming tools such as GitHub Copilot or Claude Code are even more recent and have eliminated the most mechanical part of code writing. The "manual typing" work that was previously delegated to junior developers or subcontracted to external teams can now be done with excellent quality in a fraction of the time. For medium-to-large projects the difference is enormous. It's not just about speed or fewer developers involved: it's the elimination of entire links in the subcontracting chain. Fewer links means fewer markups along the chain, but also less personnel in administrative and management roles. The difference becomes staggering.
Preparing data is no longer a nightmare
There's a cost that rarely appears in quotes but historically weighed almost as much as development itself: data preparation. Taking a product catalogue — with its variants, finishes, compatibility rules, price lists — and transforming it into structured data usable by a configurator was an extremely long, manual, and expensive job: everything had to be translated into a format the software could understand. Today it remains critical work, but AI can take on a substantial portion of the manual effort. Extracting data from unstructured documents and normalising formats across Excel spreadsheets, PDF price lists, and transcripts of meetings with sales managers used to require weeks of manual work. Today it's manageable in a fraction of the time.
Developing a 3D configurator in 2026
These three effects compound almost exponentially: more reliable software that costs less per line written, and data that costs less to prepare.
The real cost of software today lies primarily in high-level design, identifying business needs to automate, visual design, 3D modelling, and testing. Code writing, which once represented at least half of a project's total cost, has become the least significant component.
The combined result is that the cost of code writing alone has shrunk to a tenth or even a twentieth compared to ten years ago. For the project as a whole (including design, client communication, testing) the factor is more conservative — in the range of 3–5 times — but still transformative. A 3D configurator that ten years ago was simply out of reach for an SME, and that three years ago still ran €40–50,000, can now be developed for a fraction of those costs.
And quality doesn't suffer — quite the opposite. The libraries used today are tested by millions of developers. Less reusable code to develop means fewer bugs, and the time that software houses no longer spend writing boilerplate can be invested in what the client actually perceives: the user experience, the fluidity of interaction, the care for visual detail.
What this means in practice
Let's take typical examples from the Italian and European manufacturing landscape. A modular wardrobe manufacturer with around thirty finishes and various internal configurations can now get a custom 3D configurator — with compatibility logic, real-time visualisation, and commercial output — at costs that in most cases tend to fall between 12 and 18 thousand euros, with lower and higher extremes depending on complexity. Custom configurators for simpler products with polished animations, photorealistic rendering, and bespoke business integrations start from €7,000.
These are not static viewers. They are not demos. They are fully functional, tested tools, integrated into the website, usable by customers and the sales team.
Three years ago, these same products would have required a significantly higher investment. Ten years ago, they would have been out of reach for the vast majority of SMEs.
The information paradox
There's an irony in all of this. Development costs have fallen, but the information available online doesn't reflect this in a consistent or unified way. Generative AI systems repeat and amplify data that they themselves have helped make obsolete. New sources with updated development cost data overlap with old ones, creating the confusion and overlap we discussed at the beginning: it's not that "it depends on the project" — it's that the same response contains today's prices alongside prices from three or ten years ago, American costs and European costs, enterprise quotes and freelancer estimates.
The risk is that a CEO or business owner who would like to digitalise intelligently and appropriately for the present finds price ranges that discourage them — or push them towards subscription-based solutions that don't fully meet the company's needs and over time end up costing more than custom development.
If the prices found online seem too high for what's needed, there's a good chance they actually are higher than the real market. If price ranges overlap and contradict each other, that's a confirmation. The market has moved faster than the information describing it.