The AI arms race is heating up, and it's not just the U.S. leading the charge. China's AI models are quietly gaining ground, even on American soil, challenging the notion that the U.S. holds an unassailable lead in artificial intelligence. Take Zhipu AI's GLM 4.7, for instance. While not entirely new, its growing popularity in the U.S. is turning heads. I first stumbled upon Zhipu's WeChat announcement, revealing overwhelming demand for their coding tool, leading to access restrictions. This echoes the ripple effect caused by DeepSeek's R1 model a year ago, which sent shockwaves through the American AI landscape. Intrigued, I reached out to Zhipu, expecting radio silence like my previous attempts with DeepSeek. Surprisingly, Zhipu's investor relations team, fresh off their Hong Kong IPO, responded promptly, confirming that their user base is predominantly in the U.S. and China.
This is significant because American developers have historically been wary of Chinese AI models. Zhipu's traction in the U.S. suggests a potential paradigm shift, reminiscent of DeepSeek's breakthrough. Just last week, we were amazed by the capabilities of Replit and Claude Code, showcasing the cutting edge of American AI innovation. These tools are driving a tangible surge in app development, with a 60% increase in new releases. But here's where it gets controversial: If Chinese models like Zhipu's GLM 4.7 are equally powerful, user-friendly, and potentially cheaper or open-source, are American AI agents truly six months ahead, as Google DeepMind's Demis Hassabis claims? And if so, what competitive advantage do U.S. models truly hold?
CNBC's Deirdre Bosa and I decided to put Zhipu to the test, comparing it to our experiences with Replit and Claude. We tasked it with building a tracker for China's largest public companies, a notoriously complex dataset. Zhipu delivered faster than its American counterparts but with less polished results. Skeptical of self-reported usage claims, we consulted AI builders who would encounter Zhipu if it were genuinely gaining traction. The verdict? Zhipu's GLM 4.7 is indeed making waves in the U.S., though opinions are divided.
One notable tester is Tuhin Srivastava of Baseten, a platform integral to AI applications across enterprises. Baseten's recent funding round, with Nvidia's participation, underscores its role in large-scale AI workloads. Srivastava's insights, shared in a livestream with Bosa, offer a deeper dive into Zhipu's real-world performance.
And this is the part most people miss: The AI landscape is no longer a U.S.-dominated field. China's advancements are not just catching up—they're carving out their own space. This raises a thought-provoking question: Are we witnessing a new era of AI competition, or is collaboration the key to unlocking the full potential of this technology? What do you think? Let us know in the comments below.
