The USA vs China AI race may define Trump's second term

DeepSeek shocks the tech world, but some remain skeptical.

DeepSeek, a Chinese artificial intelligence company, shocked the technology world over the weekend when it released its latest machine learning model.

Renowned tech entrepreneur Marc Andreessen, who became a major Trump donor in the runup to the 2024 election, likened the development to “AI's Sputnik moment. " He described DeepSeek as "one of the most amazing and impressive breakthroughs" he's witnessed.

---

Check out The Dossier’s fantastic sponsors, curated exclusively for our readers

Squarespace has the best tools to grow your future online presence.

Design a custom website with Squarespace's professionally curated layout and styling options designed to sell anything. Start with a flexible designer template or build your own, then customize to fit your style using our drag-and-drop website tool.

---

Satya Nadella, the CEO of Microsoft, called it "super impressive" and noted its claimed efficiency and the breakthrough it represents despite U.S. restrictions on semiconductor exports to China.

DeepSeek presents itself as a groundbreaking initiative in developing large language models (LLMs) that claim to offer unparalleled cost efficiency alongside advanced capabilities. DeepSeek executives claim that its AI program can compete with Chat GPT, Claude, Gemini, and other costly, renowned AI models at a fraction of the cost.

DeepSeek has deep ties to the Chinese Communist Party

The cost-efficiency claim caused an enormous reaction in U.S. markets, as the NASDAQ lost over 600 points on Monday due to the fallout from the powerful narrative.

While many in Silicon Valley remained awestruck, not everyone is drinking the Kool-Aid.

Elon Musk poured cold water on the hype and suggested that Deepseek (through its patron, theent) is more likely smuggling upwards of billions of dollars worth of advanced Nvidia computing Chinese governmpower into the country despite export controls that the Biden Administration attempted (and failed?) to enact.

Musk found a skeptical ally in Scale AI CEO Alexandr Wang.

“My understanding is DeepSeek has 50,000 [Nvidia] H100s,” Wang told CNBC. “They can't talk about [them] because it is against the export controls that the U.S. has put in place.”

Training AI models involves enormous datasets, which means high computational costs, energy consumption, and significant human effort for data curation and model tuning. DeepSeek claims to overcome these costs through innovative training methods and presumably more efficient algorithms. Still, the specifics of how this is achieved have been glossed over or presented with broad, optimistic statements rather than concrete metrics. Additionally, continuous learning and updating an AI model to keep up with knowledge evolution can be resource-intensive. If DeepSeek's model requires frequent retraining or significant human oversight to maintain performance, the long-term cost efficiency might be less than advertised.

While DeepSeek has open-sourced its software under an MIT license, there is little transparency around its methodology, data sources, and actual cost metrics. Without detailed breakdowns or independent audits, their claims about cost efficiency boil down to buzzword marketing hype over proven facts.

But for now, DeepSeek claims to be revolutionizing AI with cost-effective solutions. Skepticism should remain until more concrete evidence and transparent methodologies are provided, but it’s clear that Chinese AI companies are investing considerable time and resources into competing in the space.

DeepSeek is far from the lone player in China’s AI sector.

Alibaba’s Qwen, which released an extremely ambitious AI update on Monday, is considered the top LLM in China. ByteDance, the company that owns TikTok, runs a powerful AI called Doubao. Baidu, another major player in China, is continually advancing its Ernie series, which, like DeepSeek, claims to offer competitive performance at lower costs compared to leading U.S. models.

The release of Deepseek’s latest AI model came just days after the Trump Administration, in collaboration with some of America’s AI industry heavyweights, announced the formation of the Stargate Project, which intends to deploy $500 billion into the AI space in the United States.

Trump announces The Stargate Project

There are several other non-Stargate Project members with ambitions to accelerate the American AI industry. There’s xAI and its Grok model. Google has Gemini. Meta developed models like LLaMA. And there’s Anthropic, with their model Claude. These companies have invested untold billions to advance their models.

Tae Kim, the author of The Nvidia Way, pointed out that DeepSeek rolled out its first major AI software on the morning of Inauguration Day, raising questions about the possibility that there is a significant information operation running in concert with the AI developments we are seeing out of China.

Nonetheless, Beijing has launched its opening salvo in the AI competition between the United States and China. Such a fierce competition between the world’s two greatest powers, and the considerable stakes attached to the outcome of the AI race, may very well define President Trump’s second term.