Mozilla’s chief technology officer, Raffi Krikorian, argues that artificial intelligence should be built the way the internet was—open, interoperable, and under the control of users rather than a handful of proprietary vendors.
Open‑source AI closing the performance gap
In a report released last month, the nonprofit Mozilla community documented that the difference in performance between leading open‑source models and commercial systems such as Claude and ChatGPT has narrowed to roughly three percent. The paper noted a surge in adoption of open models, citing February 2026 data that Alibaba’s open‑source model Qwen was downloaded more times than the next eight models combined.
That trend gained fresh momentum after Meta announced a new open‑weight model on August 10, promising additional releases. He says the shift reflects businesses’ desire for greater control over data and infrastructure, especially in regulated sectors that must keep information inside their firewalls.
Why enterprises are turning to open models
According to him, many IT and HR teams are moving away from consumer‑grade services like ChatGPT because open models can be self‑hosted and fine‑tuned for specific use cases. “Price‑performance reasons” and the need for full data sovereignty drive the migration.
Related: AI poses threat to democratic elections
The CTO highlighted that open‑source AI is no longer a hobbyist niche; it now supports a multi‑hundred‑billion‑dollar commercial layer. “It’s potentially one of the best well‑kept secrets right now,” he added, noting that businesses are building economies, infrastructure, and products on these models.
He compared the evolution to Linux’s rise from a simple kernel to the foundation of Android and virtually every modern computer. “Every single new vending machine probably is running on Linux,” Krikorian remarked, noting how open ecosystems can become ubiquitous.
Consumer awareness remains limited, and the gap is partly due to the complexity of open‑weight models. Users must download and run large files, which can seem daunting compared with the simplicity of a hosted service. He cautioned that misunderstandings about security and data privacy can deter adoption, especially when models originate from regions with heightened geopolitical concerns.
Governments are beginning to notice the shift but are still wrestling with how to categorize AI. “Products are by definition things that you can rent or turn off, but it should be filed as infrastructure,” he said, suggesting that policy frameworks need to catch up to the technology’s reality.
Related: Google Play Services adds Express Pay refreshes Shorts
One way to gauge the broader impact is to look at past open‑source successes. When Linux became the default for Android, it forced major firms to adapt or lose market share. Similarly, the growth of open‑weight AI could compel dominant AI providers to open their models or risk marginalization. This parallel offers a hint of how the field might evolve, though the timeline remains uncertain.
True openness requires transparency about training data and evaluation methods. While many current models allow download, execution, and fine‑tuning, they often lack full disclosure of the datasets used.
Large companies will continue to prioritize regions where they can generate the most revenue, which may leave parts of the world underserved. Open‑source ecosystems provide a pathway for local developers to tailor AI to language, cultural values, and specific community needs.
He envisions a future where the biggest AI service provider offers an open version of intelligence that any company can customize, akin to current cloud storage or office‑suite subscriptions. “It will be a services provider that can walk into a company and help them actually get over that hump and use AI,” he concluded.