The State of America’s AI Landscape: Rise of Open Models
Earlier this year, Misha Laskin, a theoretical physicist turned machine learning engineer, voiced significant concerns about the landscape of artificial intelligence in America. Having played a pivotal role in creating powerful AI models at Google, Laskin observed a remarkable trend: a growing preference among American AI companies for free, customizable, and increasingly sophisticated “open” AI models, many of which originate from China.
The Chinese AI Surge
As Laskin noted, these open models from China weren’t just lagging behind; they were getting dangerously close to competing with, and in some instances overtaking, their American counterparts. “These models were not that far behind the frontier. In fact, they were surprisingly close to the frontier,” he remarked. Such advancements are prompting Laskin to take action: he founded a startup called Reflection AI, recently valued at an impressive $8 billion, to provide open-source alternatives specifically designed to battle the growing capabilities of Chinese systems.
Open Models vs. Closed Models
The competition between American closed models from companies like OpenAI and Anthropic and Chinese open-source models like DeepSeek’s R1 and Alibaba’s Qwen is becoming increasingly pronounced. American models, typically proprietary and accessible through data centers, have historically outperformed their open competitors. However, there’s been a shift. Over the past year, a growing percentage of startups in the U.S. have begun to lean on Chinese AI models, which are often cheaper, faster, and more customizable. This trend has raised eyebrows among industry experts.
Startup Perspectives
To gain deeper insights, NBC News interviewed more than 15 AI startup founders, engineers, and investors. A common theme emerged: while American models continue to set the pace in terms of advancements, many Chinese alternatives are now robust enough to replace these established systems in various applications. Investors have poured billions into American AI companies, banking on their dominance in the global market. But the increasing reliance on free Chinese models raises questions about the unique advantages those American models claim to possess.
Cost and Efficiency
Michael Fine, the head of machine learning at Exa, a well-regarded AI search company valued at $700 million, provided a practical viewpoint. He noted that using Chinese models hosted on their infrastructure has proven to be significantly faster and less expensive than employing larger models like OpenAI’s GPT-5 or Google’s Gemini. “What often happens is we’ll get a feature working with a closed model and realize it’s too expensive or too slow,” Fine explained.
This logical pivot toward open models reveals a significant advantage: adaptability. When operational challenges arise, companies often explore open alternatives that can be modified and optimized to meet their specific needs.
Examining Model Performance
The nature of AI models from China, such as DeepSeek’s R1 and Alibaba’s Qwen, is noteworthy. These models are classified as “open-source,” meaning they can be freely downloaded, modified, and run by anyone. This open-access framework stands in contrast to American models that require users to rely on proprietary data centers.
Historically, U.S. closed models eclipsed both American and Chinese open alternatives in performance. Even enterprises with ample resources struggled to compete. For instance, Bloomberg attempted to create its own tool, BloombergGPT, using open-source models, only to find that it lagged behind OpenAI’s offerings in the intricacies of financial knowledge.
Rapid Advancements in Chinese Technology
However, this landscape is rapidly changing. In the past year, Chinese companies have made remarkable strides: their open-source products are starting to match, and in some cases even exceed, the performance of American closed models in various applications, as analyzed by metrics tracked by Artificial Analysis, a benchmarking company specializing in AI.
Lin Qiao, CEO of Fireworks AI and co-creator of PyTorch, highlighted the narrowing gap between these competing models. He emphasized, “The gap is really shrinking,” pointing to a future where American closed-source models may no longer dominate the discussion.
Implications for the Future
The bespeaking trends in the AI sector signal a shifting paradigm. The increasing adoption of Chinese open models could pose a challenge to American companies historically insulated by their proprietary systems. The evolving landscape emphasizes a need for American innovation to pivot in response to emerging global competitors, encouraging adaptability, customization, and open-source collaboration.
With technology moving at lightning speed, the dynamic between closed and open models is one to watch closely as the ramifications could redefine the framework for AI development and utilization across the globe.
