NVIDIA launches Open Secure AI Alliance as Silicon Valley splits over open-source AI

NVIDIA's Open Secure AI Alliance: Silicon Valley's Fault Line

Jensen Huang is joining Twitter for the first time to settle a fight about the future of AI. The NVIDIA CEO, who until now avoided the platform entirely, has launched the Open Secure AI Alliance—a coalition of 35+ companies with $18 trillion in combined market cap advocating for open-source AI as a cybersecurity asset rather than a national security threat.

The alliance includes SpaceX, Microsoft, Palantir, CrowdStrike, Databricks, Cloudflare, OpenAI, and a who's who of tech incumbents. Notably, it does not include Anthropic, which has spent the past year warning that open-source models—particularly those from Chinese labs like Moonshot and DeepSeek—pose serious risks to national security and could be used for offensive cyberattacks.

The coalition's core argument is direct: open models democratize defensive capabilities. By distributing AI tools widely, everyone from large enterprises to small businesses can build security systems. The alternative—concentrating AI behind closed API walls—leaves the long tail of smaller companies vulnerable. The alliance cites the recent Hugging Face hack as Exhibit A: when OpenAI and Anthropic's proprietary models refused to analyze attack logs due to safety guardrails, Hugging Face turned to an open-weight Chinese model, which completed the analysis and allowed the company to expel the attacker.

The distillation accusation

Anthropic and OpenAI have leveled a different charge: Chinese developers are using distillation—a process of extracting knowledge from proprietary models to create knockoffs—to copy their frontier models illegally. The White House administration appeared to signal support for action against this practice, with officials suggesting distillation violates terms of service. Anthropic CEO Dario Amodei has not published a direct rebuttal or policy proposal, but his silence in the face of an $18 trillion coalition speaks louder than a blog post.

The tension reflects a split within the administration itself. David Sacks, White House AI advisor, has pushed back hard against the idea of banning open-source models, calling it "a tragic mistake" that would "hurt America's position in this AI race." Emil Michael shares that view. But other officials appear more skeptical of open-source safety claims.

The economic divergence

What makes this moment different from prior tech fights is the economic scale and symmetry. When Apple and Google fought over mobile in 2008, Android eventually became the ad-supported model while iPhone remained premium. The winner was obvious. Now, the Magnificent Seven are roughly balanced in market cap, and all of them are building AI in parallel—whether closed models (Anthropic, OpenAI), open partnerships (Microsoft with OpenAI), or open-source initiatives (Meta, Google). No company is sitting out the AI boom, and no clear winner has emerged.

The revelation comes from Flo Crivello, founder of Lindy AI, who broke ranks with his own financial incentives. His company depends on Chinese open-source models, which are dramatically cheaper than Anthropic's Claude. Yet Crivello announced support for a ban on Chinese models, arguing that:

• China is a geopolitical adversary fundamentally misaligned with American values.
• Moonshot and DeepSeek achieved their performance through distillation of stolen American frontier models.
• Allowing artificially cheap Chinese models to flood the U.S. market undermines American open-source development, which would otherwise catch up.
• "The race for AGI is the defining struggle of our generation."

Crivello framed this not as founder interest but as patriotic duty, accepting short-term business damage for long-term strategic advantage.

What likely happens next

The $18 trillion coalition likely prevents a sweeping ban. The economic weight is too large to ignore. Instead, the likely path is litigation. Anthropic could sue Moonshot for distillation as a terms-of-service violation and intellectual property theft, much as Apple is currently suing OpenAI. The case would move through U.S. courts with discovery and testimony, but litigation against foreign companies on foreign IP allegations is inherently slow and jurisdictionally messy.

Meanwhile, the security argument cuts both ways. Open models do democratize defensive tools, but they also democratize offensive ones. Chinese researchers have shown competitive capability with models like K3, suggesting they've either independently matched American training efficiency or successfully distilled frontier models. The cost gap—open models running at a fraction of proprietary pricing—has invited downstream players to adopt them.

The wildcard: whether a significant cybersecurity incident attributed to open-source AI (say, North Korean hackers using a Chinese model) reframes the debate. Anthropic would likely benefit from such an event, validating concerns that sparked little mainstream attention beforehand.

For now, Silicon Valley has drawn a line, and Anthropic stands visibly alone on one side of it. That isolation, rather than any technical argument, may prove decisive in shaping U.S. policy.