Business
Open source AI is ‘China’s game right now’ — and that’s a problem for the U.S. and its allies, Andreessen Horowitz partner says
- China is currently dominating the open-source AI space, according to Andreessen Horowitz’s Anjney Midha. Speaking at the Fortune Global Forum, Midha said this raises geopolitical and competitive concerns for the U.S. and its allies. However, he thinks Western companies can make a comeback, citing U.S. policy support and the potential for American labs to release new open-source models in the coming months.
China’s progress with open-source models could pose a problem for the U.S. in the global AI arms race, according to Anjney Midha, a general partner at Andreessen Horowitz.
“What has not been great is that the speed at which the open-source ecosystem has accelerated the race geopolitically,” Midha told the audience at the Fortune Global Forum in Riyadh. “It’s not looking great for the United States and its allies.”
“The reality is that if you look at the most powerful models that are open source today—outside of Mistral from France and a couple of models that are small, specialized models from the U.S.—it’s really China’s game right now. And I think that’s not a particularly encouraging picture,” he added.
China’s progress in the open-source arena is largely down to DeepSeek, the Chinese startup behind the R1 model. It sparked a sharp sell-off in American tech stocks at the start of the year after investors realized that the model was built at a fraction of the cost of frontier U.S. models, but outperformed or matched several of them in key benchmarks.
R1’s popularity and other recent advancements from the company have demonstrated China’s prowess in AI innovation and intensified concerns in Washington over how open-source development could shift the global balance of power.
DeepSeek has continued to be among the most innovative AI companies in the world when it comes to finding new ways to optimize AI models and extract high performance from smaller, less expensive models. Their latest breakthrough found that having models process information as visual tokens rather than language tokens can make a model 10 times more efficient.
Leading AI companies, including OpenAI, have changed their stance on open-source AI following the launch of and market reaction to DeepSeek’s R1. In August, the leading AI lab released two open-weight language models called gpt-oss-120b and gpt-oss-20b, designed to serve as lower-cost and accessible alternatives to its frontier models. CEO Sam Altman teased the models back in March, two months after admitting, in the wake of the success of DeepSeek’s open-source R1, that the company had been “on the wrong side of history” when it came to opening up its models to developers and builders.
Midha said he believed the West was going to make a comeback on open-source AI innovation, crediting the recent “AI Action Plan” from the Trump administration as clearing the way for American entrepreneurs and researchers. Midha previously campaigned against state-level AI regulation, notably the NY RAISE Act, arguing that a patchwork of regulation across the country will hurt the U.S. at a time when rival nations are racing ahead.
“Researchers who have the skill set to push the frontier should be spending their time on pushing the frontier of capabilities, not on navigating 50 different pieces of legislation,” he said. “I think what you’re going to see as a result of all that freed mind space of these entrepreneurs and scientists is that three, four, or five months from now, you’re going to start to see a wave of open-weight models from American labs.”
