By Doug Kelly:
Today, Meta and Apple, two leading American innovators, came together on something that matters to every American business and to our competition with China – artificial intelligence (AI).
Meta released Muse Spark Open, which it calls the most capable American open-weight AI model to date. It is also partnering with Apple so developers can run the model on local desktop hardware, rather than the cloud. That puts frontier-scale, American-made AI directly in the hands of people across the globe who build products, start companies, and create jobs.
Why open models matter to American leadership
Open-weight models let developers download, customize, and build on AI without starting from scratch. This empowers individuals and prevents power from concentrating in the hands of a few gatekeepers. That means a small manufacturer in Ohio on a tight budget could use AI built on an open-weight model to cut waste on the production line. A startup could use it to compete with companies a hundred times larger.
That kind of access to innovation is a big deal for America’s more than 36 million small businesses, which employ nearly half of our private-sector workforce. Most can’t afford to build their own AI. Now they don’t have to.
China’s strategy is built around open models
American Edge has warned that China is using its open-source ecosystem as a Trojan horse to embed Chinese Communist Party (CCP) values into the globe’s digital infrastructure. Its strategy is paying off. Alibaba’s Qwen models have passed 3 billion downloads worldwide. On OpenRouter, a popular platform developers use to access AI models, China’s open-weight models jumped from under two percent of all usage on the platform in late 2024 to more than 45 percent by June this year.
But there’s a worrisome catch. These models carry Beijing’s censorship with them. CBS News reported that Qwen won’t acknowledge the 1989 Tiananmen Square crackdown. On sensitive political questions, the Qwen model gave censored or propaganda-aligned answers nearly 90 percent of the time, compared to less than three percent of the time with Western models.
The risks go beyond censorship and propaganda. Booz Allen tested China’s leading models and found those models wrote less secure code when they believed they were working for the U.S. government. In those cases, Qwen’s code had 130 percent more vulnerabilities. That’s a potential open door for hackers to steal trade and government secrets and a hidden risk in the systems that protect our national security.
America must lead in open and closed AI
This isn’t a choice between open- and closed-source models. America needs both to lead or China will fill the gap. That’s a gap we can’t afford because the AI race is closer than many Americans and policymakers realize.
China is already ahead in three of the four areas that decide AI leadership: energy, talent, and adoption. Our sole edge lies in compute power, including advanced chips, top models, and data centers. But short-sighted legislation and moratoriums on data center construction and other critical AI infrastructure threaten that edge at home, while China’s AI models erode it abroad.
The bottom line is that AI can be America’s modern-day moonshot, delivering decades of economic growth and national security advantages. But we won’t get there if China builds the infrastructure faster and gets its AI into more hands around the world.
That’s what makes the Meta-Apple partnership so encouraging. Two competitors decided that strengthening American AI is paramount. On this, they’re on the same team.
This is one of America’s greatest strengths. Our innovators compete hard, collaborate when it counts, and the whole country wins. In the race with China for AI leadership, we need a lot more of that collaboration.