Chinese companies are leading the AI arms race. Chinese politician and computer scientist Lou Qinjian said as much, recently commending DeepSeek for their accomplishments: “DeepSeek adheres to an open-source approach and promotes the widespread application of AI technology globally, which contributes Chinese wisdom to the world,” he said. 

“Through the rise of companies like DeepSeek, we can see the innovation and inclusiveness of China’s technological development.”

In February, at the Artificial Intelligence Action Summit in Paris, US Vice President JD Vance made clear where the Trump Administration stands on artificial intelligence. He said that, first and foremost, the Trump administration will ensure that American AI technology remains “the gold standard” worldwide and that US companies remain the partner of choice for international companies and foreign countries. 

The Vice President argued that excessive regulation in the AI sector would kill the nascent industry, and that the administration would encourage pro-AI growth policies. “And I’d like to see that deregulatory flavor, making its way into a lot of the conversations at this conference,” he said. Vance also made it clear that AI should be free of ideological bias and that “American AI will not be co-opted into a tool for authoritarian censorship.” 

Finally, the Trump administration will safeguard a pro-worker growth path for AI so it can create jobs in the United States. Vance also brought up the notion of foreign adversaries weaponizing AI software to rewrite history, surveil users, and censorship. As Vance stated:

“This is hardly new, of course, as they do with other tech. Some authoritarian regimes have stolen and used AI to strengthen their military intelligence and surveillance capabilities, capture personal data, and create propaganda to undermine other nations’ national security.”

He warned conference attendees against partnering with such regimes. “From CCTV to 5G equipment, we’re all familiar with cheap tech in the marketplace that’s been heavily subsidized and exported by authoritarian regimes,” he said. “But as I know, and I think some of us in this room have learned from experience, partnering with them means chaining your nation to an authoritarian master that seeks to infiltrate, dig in, and seize your information infrastructure.”

Under the hood of DeepSeek

DeepSeek shocked global markets in January with low-cost models that made it seem like US companies were now behind in the AI arms race. The AI lowered the costs of developing reliable AIs, proving itself to be a powerful and cost-efficient open-source language model. 

It changed the way we view how much capital and computational resources are needed to develop AI.  Researchers across the Western world are now left playing catch-up, studying DeepSeek’s technical advances and social implications. 

There are clear benefits to DeepSeek. For instance, startups without the deep pockets of Google and OpenAI can now compete in the AI sector. AI models can do more with less in the post-DeekSeep world. The company claims it took a mere $6 million using 2,000 Nvidia H800 graphics processing units (GPUs) versus the $80 million to $100 million cost of GPT-4 and the 16,000 H100 GPUs needed for Meta’s LLaMA 3. 

The Hangzhou-based startup’s AI model employs reasoning capabilities that allow smaller models, whereas other AIs have had to employ larger models. It also uses reinforcement learning, eliminating the need for supervised fine-tuning. Moreover, DeepSeek’s multi-head latent attention (MHLA) mechanism decreases memory usage to 5%, down from 13%, in earlier AI methods. 

DeepSeek raises privacy concerns and questions regarding data-sourcing and copyright. DeepSeek is open-weighted, not open source. Open source models share the full source code and data, and open weight models share trained weights but not the code. Therefore, the exact source code used to train the models is not available. 

Due to DeepSeek’s open weight model, it is unknown what its sources are. This seems to be the way most AI companies operate. DeepSeek made public its R1 training and open weight models, which will allow other AI developers to copy and build on the model, but not its sources.

DeepSeek and geopolitics 

A race for AI dominance between China and the US has come into focus, while Russian capabilities on the matter remain a secret. Sberbank—Russia’s largest state-owned bank—has revealed its intentions to collaborate with Chinese researchers on AI projects. Russia and China, which share what they call a “no limits” strategic partnership, have long talked about AI cooperation—including in military applications—but little is publicly known about its depth or scope.

Sberbank, under CEO German Gref, once a Soviet-style former state savings bank burdened by onerous bureaucracy, is today one of Russia’s leading players in artificial intelligence. It released its GigaChat model in 2023. “Sberbank has many scientists. Through them, we plan to conduct joint research projects with researchers from China,” Sberbank First Deputy CEO Alexander Vedyakhin told Reuters. 

As the AI arms race heats up, the benefits of open source innovation come to the forefront. Little flowers bursting through the concrete all around the world, coming up with cool tech that is open-sourced and decentralized.