Why China's free AI models are rattling Silicon Valley

The News

Silicon Valley spent much of the past week on alert after Moonshot AI unveiled Kimi K3 at the World Artificial Intelligence Conference in Shanghai on 20 July. The Chinese model reportedly matches or beats several leading American systems while costing a fraction to run, and it ships as open weights that anyone can download and modify.

Kimi K3 is only the latest freely released Chinese model. Alibaba's Qwen family has now passed 700 million downloads, making it one of the most widely used open-weight lineups anywhere. The trend has pushed US labs to respond in kind: OpenAI released GPT-OSS and Google keeps its Gemma range, both departures from the closed, paid-API playbook that defined the sector two years ago.

The question in boardrooms is no longer whether Chinese models can compete on quality, but what it means that they are being handed out for nothing.

Why It Matters

Give-it-away economics unsettle a business built on scarcity. As Chinmayi Sharma, a law professor at Fordham Law School, put it, "A free set of weights is not a free AI service." A firm can release the model and still earn from hosting, tooling and support around it.

For China, open weights double as industrial policy. Curbs on advanced chips make a brute-force compute race hard to win, so distribution becomes the lever: seed the world's developers with Chinese models and the standards and dependencies follow. Kyle Miller, a senior research analyst at Georgetown's Center for Security and Emerging Technology, has warned that widely adopted open models can become a "de facto standard" for everyone who builds on them.

The last comparable jolt was DeepSeek in early 2025, which erased hundreds of billions in AI-linked market value and reset expectations for what a frontier model should cost. Kimi K3 extends that pressure, and a letter signed by 25 tech companies opposing new restrictions shows how split the US industry has become.

Indian Angle

For India, cheap open weights look more like an opportunity than a threat. The IndiaAI Mission has funded home-grown foundation-model work, and startups such as Sarvam and Krutrim are building Indian-language systems on tight budgets. Every capable model released for free lowers the entry cost for a Bengaluru or Pune team that cannot match a Californian lab's compute bills, and it keeps rupee spend from leaking out through metered foreign APIs.

There is a governance flip side. If Indian public services and startups quietly standardise on Chinese-origin weights, MeitY and data-sovereignty advocates will press on provenance, auditing and long-term dependence - the same worry Washington now voices, seen from the other end of the supply chain. India's response so far has been to back its own open releases rather than pick a side.

FAQ

What is Kimi K3?

An open-weight large language model from Chinese startup Moonshot AI, shown at the World Artificial Intelligence Conference in Shanghai on 20 July 2026. Open weights mean the files can be downloaded, run and fine-tuned by anyone, not just reached through a paid interface.

Why would a company give its model away?

Because the weights are not the whole product. As Fordham's Chinmayi Sharma notes, firms can still earn from hosting, support and tooling built around a free model, while the release itself drives adoption and quietly sets ecosystem standards.

How does this affect Indian developers?

It lowers costs. Free, capable models cut reliance on metered foreign APIs and help Sarvam, Krutrim and smaller teams ship Indian-language products cheaply, keeping more spend inside India rather than paying it out in foreign currency.

This story was reported by The Verge. Read the full original coverage at The Verge.