For more than two decades, Mozilla fought to keep the web open. Now its chief technology officer, Raffi Krikorian, wants to do the same thing for artificial intelligence.
In a conversation with Rest of World, Krikorian made a case that is rapidly becoming mainstream outside Silicon Valley: AI should be treated as infrastructure, not as a rented product. He believes the open-source and open-weight ecosystem has already become a multi-hundred-billion-dollar commercial layer, that the performance gap with proprietary systems has nearly vanished, and that Western governments are dangerously wrong to view AI through the lens of their favorite frontier labs.
The 3% Gap
In a report published last month, Mozilla found that the performance gap between top open-source models and proprietary systems like Claude and ChatGPT has narrowed to just 3%. That is close enough that price, control, and customizability increasingly matter more than raw benchmark supremacy.
The popularity of open models is not theoretical. In February 2026, Alibaba’s open-source Qwen model was downloaded more times than the next eight models combined, according to Mozilla’s research. Alibaba’s open-weight push is not a charity project. It is a distribution strategy that is working at industrial scale.
The momentum gained another jolt on August 10, when Meta released a new open-weight model and promised more would follow. Open-weight releases from Chinese labs, American giants, and European challengers are now arriving on a near-weekly cadence. The question is no longer whether open models can compete. The question is who gets to define what “open” actually means.
Infrastructure, Not a Product
Krikorian’s central argument is that the United States has miscategorized AI. Washington, he says, sees intelligence as a product — something you rent from OpenAI or Anthropic. Products can be turned off, price-hiked, or geo-blocked. Infrastructure cannot.
“Americans have generally filed intelligence as a product,” Krikorian told Rest of World. “Products are by definition things that you can rent or turn off, but it should be filed as infrastructure, which is what the rest of the world seems to want to buy. That categorization error is actually the problem.”
The consequences are practical. Businesses in regulated industries want to self-host. Governments want data to stay inside national firewalls. Hospitals, banks, and defense agencies cannot ship sensitive information to an API endpoint in California. For them, open-weight models that run locally are not a hobbyist preference. They are a requirement.
Who Is Actually Using Open Models?
Consumers may still reach for ChatGPT and Claude, but Krikorian says enterprise workflows are quietly migrating toward open models. The reasons are price-performance and control. An IT or HR team building an internal tool can fine-tune an open model on proprietary data without sending it anywhere. A company in a regulated industry can keep everything inside its own firewall.
“It is because of the marketing blitz of the big frontier labs that we all view AI just as something that Anthropic, OpenAI, and a few other companies are doing,” Krikorian said. “There’s actually a burgeoning ecosystem on the open-source, open-weight side that a lot of people actually do know about. It’s potentially one of the best well-kept secrets right now when it comes to technology.”
The China Factor
Krikorian does not pretend the politics are simple. Many of the most popular open-weight models are Chinese, including Qwen and models from Tencent, DeepSeek, and others. That makes some policymakers nervous. He understands why.
“Open-source AI models are complicated to understand, and I think that is also part of the issue,” he said. “The simplest explanation is the one that usually ends up winning.” He also believes the West should ask a harder question: if global demand for open models is this large, why are American, Indian, and European companies not racing to fill it?
“The confounding thing should be why aren’t more countries and organizations trying to fill that demand?” he asked. Instead, Washington is caught between backing its current standards-bearers — OpenAI and Anthropic, both heading toward IPOs — and taking the risk of nurturing a more distributed ecosystem.
Open, But Not Open Enough
Krikorian is careful to distinguish between “open source” and “open weights.” Most of what is marketed as open-source AI today is actually open-weight: you can download and run the model, but you do not necessarily know what data was used to train it or how it was fine-tuned. True open source would require transparency around pre-training, post-training, evaluation, and datasets.
Still, open weights unlock something important. They allow communities, companies, and countries to take a large foundation model and fine-tune it for local languages, legal systems, and cultural values. That pathway is essential for the “rest of the world” that proprietary labs will never prioritize.
“I’m not faulting what the big companies are doing, but they are going to optimize for the places where they can make the most money,” Krikorian said. “So they’re going to optimize for the Western world and maybe parts of Asia, but they’re not going to optimize for the rest of the world. The rest of the world only becomes part of this AI transformative conversation through the open-source and open-weight ecosystem.”
Can Mozilla Pull It Off?
The skepticism is obvious. Mozilla championed the open web, yet Google owns search, Meta owns social, and Amazon owns cloud. Why would open AI end any differently?
Krikorian’s answer is that Firefox itself shifted the industry even with minority market share. Its existence forced web standards, kept protocols open, and prevented any single company from fully enclosing the web. He wants the same dynamic in AI: a vocal, technically viable open alternative that forces proprietary players to compete on interoperability, user agency, and privacy.
“You could make an argument that we get a small amount of traffic on the AI internet through the open models, but that would be enough to force a conversation around model choice, agentic harness choice, user agency, and making sure that memory stays on my side, not the user side,” he said.
🔥 Hot Takes
1. The “open source vs closed source” debate is over. Open source already won the enterprise; the only fight left is geopolitical. The 3% performance gap is inside the noise for most real-world applications. What matters is sovereignty, cost, and control. Enterprises and governments outside the US are voting with their procurement budgets, and they are voting open. The only people still pretending this is a technical debate are the ones selling closed APIs.
2. Washington is about to regulate itself out of the global AI market. If the US keeps treating AI as a product to be exported by two or three San Francisco companies, it will cede the infrastructure layer to Qwen, DeepSeek, and Meta’s open models. The rest of the world will not pay a premium to rent intelligence from a country that could sanction, surveil, or simply hike prices overnight.
3. Mozilla’s real challenge is credibility, not technology. Firefox kept the web open, but Mozilla also missed mobile, lost search revenue to Google, and spent years looking irrelevant. If it wants to lead the open AI movement, it needs to ship tools that developers and enterprises actually use — not just publish reports and host panels. The thesis is right. The execution is everything.
Bottom Line
Raffi Krikorian is not saying open-source AI is perfect. He is saying it is inevitable, and that the West is making a strategic mistake by pretending otherwise. The global demand for controllable, affordable, local AI is too large for a handful of proprietary labs to serve. Open-weight models — many of them Chinese, many of them imperfectly transparent — are filling that gap because no one else is moving fast enough. Mozilla wants to be the organization that turns that groundswell into durable infrastructure. Whether it succeeds depends less on ideology than on whether it can ship something the world actually wants to use.