Open source in the enterprise: what AI changes, with the figures

Open source has never been used so widely, and AI is speeding things up: 89% of organisations that use AI also rely on open source AI, and the best open models now trail closed models by only four months on average. This guide brings together sourced figures from 2025 and 2026, what they really say about the relationship between companies and their software vendors, the benefits and limits of open source, and how to make the move.

It is written for executives, CIOs and CTOs wondering whether to replace a vendor’s software or a closed AI service with an open source solution. Integrating AI models into business applications is part of our artificial intelligence offer, and our web development is built on open source components.

Has open source really taken off with AI?

Yes, and GitHub’s figures show it: from September 2024 to August 2025, the platform gained more than 36 million developers, roughly one per second, and public projects received 1.12 billion contributions, 13% more than the year before. AI drives a large share of this growth.

AI works both ways. It produces new open projects: models, agent tools, integration libraries. And it lowers the cost of entering existing code: a coding assistant reads, explains and modifies an open source project much faster than a developer could discover it alone. This second effect is our reading of the field, not a figure: it explains why complex open software, long reserved for specialised teams, is becoming usable by a mid-sized team.

Are open AI models as good as closed ones?

Almost: according to Epoch AI, since January 2026 the best open-weight models have trailed the best closed models by four months on average, measured on its aggregate capability index (Epoch Capabilities Index). Epoch notes the gap has widened slightly since its previous measurement, around three months: catching up is not guaranteed.

Four months is roughly the gap between two generations of closed models. For many business uses (summarising, classifying, extracting, searching internal documents), a good open model is enough. The gap matters mostly for the hardest tasks: long reasoning, complex code, agents chaining many actions.

On the business side, the Linux Foundation reports that two thirds of surveyed organisations consider open source AI cheaper to deploy than proprietary models, and nearly half choose it primarily for that reason. This is a stated perception, not a cost measurement, and the study was commissioned by Meta, which publishes its own open models.

Why are companies leaving their vendors?

In Europe, the main reason is dependency: in Perforce’s 2026 State of Open Source report, published with the Open Source Initiative and the Eclipse Foundation, 63% of organisations in the EU and UK cite fear of vendor lock-in as a key motivation for choosing open source, against 51% in North America. Cost also matters, on both sides of the Atlantic. Three situations keep coming back.

• Imposed price increases. After Broadcom acquired VMware, 86% of the large North American companies surveyed by CloudBolt in January 2026 are reducing their VMware footprint. But 54% stay while reducing, and those who migrate mostly move to public cloud (72%): leaving a vendor does not mean moving to open source.
• Licence changes. When Redis left open source in March 2024, part of its community created Valkey, a copy that stayed open. Redis 8 returned to an open source licence, AGPLv3, on 1 May 2025. Elastic, which left open source in 2021, added AGPL to Elasticsearch and Kibana on 29 August 2024. Open software cannot be closed for everyone: the last open version remains usable and can be taken over.
• Sovereignty. In France, the state rolled out Visio in January 2026, its videoconferencing tool from La Suite numérique, in place of Teams and Zoom, with the switch due by 2027. On 8 April 2026, the Dinum, the state’s digital agency, asked every ministry for a plan to reduce non-European dependencies by autumn 2026 and announced it would move its own workstations from Windows to Linux. On 22 October 2026, the Cour des comptes, France’s public audit court, recommended a gradual switch of the state’s office and collaboration tools to free software, over three to four years from 2027.

The Cour des comptes gives an order of magnitude for costs: around 100 euros per workstation per year for a free suite hosted in a SecNumCloud-qualified cloud, against 250 euros for Microsoft 365. This is an estimate made for the state, not a price that applies as is to a company.

Are vendors really losing their customers?

Not massively, and the figures say so: according to the EY and Numeum Top 250 survey published on 4 November 2026, French software publishers grew their software revenue by 5.2% in 2025, excluding video games, and SaaS accounts for 67% of it.

What changes is the balance of power. Companies leave a vendor on a specific scope, when prices rise or a licence closes, and they no longer commit for ten years without a fallback. Vendors are adapting: according to the same survey, 80% have already built AI into their offer, and usage-based pricing is growing alongside subscriptions. Open source then serves as an exit route, and often as a bargaining argument.

What are the benefits of open source for a company?

The first benefit of open source for a company is reversibility: the code is readable, the licence allows you to run it, modify it and hand it to another provider, and no one can withdraw the right to use the version you have.

• Independence. Changing provider does not mean changing software, and a price increase or the vendor’s acquisition does not bind you.
• Cost. No per-user licence. According to a Harvard Business School study, without open source, companies would spend 3.5 times more on software. Hosting, integration and maintenance still have to be paid for.
• Controlled data. An open AI model can run on your servers or with a European host: the documents it processes stay within your perimeter.
• Verifiable security. The code can be audited by you or a third party, and vulnerabilities are published and tracked, with their fix.
• Adaptability. What is missing can be built, instead of waiting for the vendor’s roadmap.
• Widespread skills. The major open components (PHP, PostgreSQL, Node.js, Linux, Docker) are known by many developers and providers.

What are the limits of open source?

The main limit of open source is that no one maintains your installation for you: without a vendor contract, updates, security and support fall to your team or your provider.

• Updates. Every component has its end-of-support schedule, and an abandoned version no longer receives fixes: our article on the end of support for PHP 8.2 gives an example.
• A few maintainers. According to the same Harvard study, 96% of the value of open source is produced by 5% of its developers. A project backed by a foundation or several companies is sturdier than one run by a single person.
• Licences. AGPL requires publishing modifications to software offered online to users. So-called “source-available” licences (SSPL, BSL) are not open source licences.
• Obligations. The Cyber Resilience Act requires the manufacturer of a digital product to monitor the third-party components it integrates, open source components included.
• Self-hosted AI. Running an open model requires suitable servers and operations. For small volumes, an API is often simpler.

How do you move to open source without taking risks?

Moving to open source is done one component at a time, starting with the one where dependency costs the most and a mature alternative exists. The steps, in order:

• Take stock. Licences paid and how they evolve, contracts and end dates, data, integrations with other tools. An audit of the existing setup answers these questions.
• Choose a first scope. A tool whose price or licence is a problem, and for which a mature open project exists.
• Assess the project. Licence approved by the Open Source Initiative, recent releases, number of contributors, governance (foundation, several companies), support schedule.
• Plan maintenance from the start. Who updates, who fixes vulnerabilities, how often: that is the role of application maintenance.
• Migrate in stages. Run the old and the new side by side, migrate the data, test, keep a rollback ready.
• Use AI to go faster, not to check less. Reading the code, migration scripts, tests: our AI production chain keeps the same checks.

Frequently asked questions

Open source: decide one component at a time

Open source has grown in volume and quality, and AI has accelerated both: it now covers complex subjects, up to AI models. The choice is no longer all or nothing, vendor or open source, but one component at a time: where dependency costs too much, where a mature alternative exists, and with maintenance planned from the start.

The iones team, a nearshore provider based in Tunisia, builds and maintains applications on open source components (PHP, Symfony, Laravel, PostgreSQL, Node.js, Angular, Docker) and integrates AI models into business applications through its artificial intelligence offer. To study a move away from a vendor or an AI project, write to us.