Mozilla has announced Thunderbolt, an open-source enterprise AI client built around a new local-first protocol that keeps corporate data entirely off the cloud, positioning the organization as a serious challenger to proprietary copilot products from Microsoft and Google.
For years, the central friction in enterprise AI adoption has been a straightforward trust problem: companies want intelligent tools, but not at the cost of feeding sensitive data into someone else's cloud. On April 16, Mozilla Corporation made its move on that gap. Thunderbolt, announced by Chairwoman Mitchell Baker and Chief Product Officer Steve Teixeira, is a fully open-source AI client designed to run powerful language models directly on corporate hardware, with no data leaving the machine at any point in the process.
The architecture behind Thunderbolt is what makes this announcement technically interesting rather than just politically symbolic. Mozilla has introduced the Local Neural Environment protocol, a new open standard for peer-to-peer communication between local large language models and corporate infrastructure. The LNE protocol is not a locked format controlled by Mozilla; it's an open specification, which means other vendors and developers can build on it. That's a deliberate strategic move, borrowing a page from Mozilla's own browser history: establish the standard, then grow the ecosystem around it.
Thunderbolt ships immediately with Mozilla-M7, a 13-billion parameter model derived from the Llama 3 architecture. In internal benchmarking it scores 92% on the GPQA benchmark, a graduate-level reasoning test that has become one of the more meaningful proxies for real-world model capability. That puts it within competitive range of several proprietary cloud models, which is a credible result for something designed to run entirely offline. The hardware floor is a minimum of 16GB RAM and 40GB of storage, targeting the new generation of NPU-equipped AI PCs that Lenovo and Qualcomm, both named as strategic partners in the launch, are actively pushing into the enterprise market.
Version 1.0 is available now on Windows, Linux, and macOS at $15 per user per month for enterprise teams. That price point is a pointed statement. Microsoft Copilot for Microsoft 365 runs at $30 per user per month. Google's Workspace AI add-ons sit in similar territory. Mozilla is offering comparable task-level functionality at half the cost, with the additional selling point that no subscription buys access to your data in the process.
What this means for the enterprise AI market
The timing is not accidental. Enterprise procurement teams across regulated industries like finance, healthcare, and legal have spent the past two years watching AI capability accelerate while struggling to justify cloud-based deployments to compliance and legal departments. Thunderbolt is essentially a product built for that specific stalemate. If the LNE protocol achieves any meaningful adoption, it could shift the competitive axis in enterprise AI away from who has the best model API and toward who can deliver reliable local inference at scale.
Mozilla is also making a broader argument about where value lives in the AI stack. Cloud AI services have been built on the premise that compute and model access are the scarce resources. Thunderbolt's local-first model suggests the scarce resource is actually trust, and that organizations will pay a premium, or accept a performance trade-off, for sovereignty over their own data. Two decades of open-source advocacy gives Mozilla an institutional credibility here that a startup making the same pitch simply wouldn't have.
The practical test will come in the months ahead. Enterprise sales cycles are long, IT procurement is conservative, and the incumbents Mozilla is targeting have deep integration advantages across their existing productivity suites. What to watch is whether the LNE protocol attracts third-party hardware and software partners beyond Lenovo and Qualcomm, and whether regulated-industry buyers, the ones with the clearest compliance rationale for local AI, begin piloting Thunderbolt at scale. If they do, Mozilla will have accomplished something genuinely rare: a second act that changes an industry twice.
Also read: OpenAI is losing ground to Claude and Gemini as GenAI users start shopping around • Anthropic's new ID and facial scan requirement for Claude is pushing privacy-conscious users toward local AI models • Anthropic releases Claude Opus 4.7 and makes its strongest case yet for owning the enterprise AI market
