DeepSeek’s Latest AI Model Disappoints Amid Rising Competition
You hit a paywall. Here’s the context on this topic based on publicly available information. We did not access any paywalled content. View original article.
DeepSeek's Latest AI Model Disappoints Amid Rising Competition
- DeepSeek's new models, released over a year after its breakthrough, underperformed compared to top Western rivals despite efficiency claims.
- The Hangzhou-based startup, founded in 2023 by High-Flyer hedge fund, shocked markets in early 2024 with low-cost models rivaling GPT-4.
Full Summary — powered by AI
DeepSeek, a Chinese AI startup launched in July 2023 by the hedge fund High-Flyer Quant, grabbed global attention in January 2024 with its V2 models. These achieved near parity with leading Western large language models like those from OpenAI, but at a reported training cost of just $5.6 million—far below the hundreds of millions typically required. The reveal triggered market jitters, causing Nvidia’s shares to drop 18% in a single day as investors feared a slowdown in demand for high-end AI chips.
That initial hype faded with DeepSeek’s recent “sequel” models, announced around April 2025, which failed to match the leap forward. Benchmarks showed them lagging behind frontrunners like OpenAI’s o1 and Anthropic’s Claude 3.5, prompting criticism that the firm had hit a performance plateau. Analysts point to intensifying global competition, with U.S. labs scaling up via massive compute resources—Nvidia’s data center revenue hit $30.8 billion in Q4 2024 alone—outpacing Chinese efficiency gains.
Compounding challenges is China’s regulatory environment. DeepSeek operates under strict state oversight, including export controls on advanced chips and mandates for data localization, which meddle with innovation. Beijing’s push for AI self-reliance has spurred over 100 domestic labs since 2023, fragmenting talent and resources. High-Flyer, DeepSeek’s backer with $4 billion in assets, has poured hundreds of millions into the venture, yet faces hurdles like U.S. sanctions limiting access to Nvidia’s H100 GPUs.
The episode underscores shifting AI dynamics: while China excels in cost-effective training—leveraging optimized algorithms and weaker domestic chips—sustained leadership demands breakthroughs in reasoning and multimodal capabilities. DeepSeek’s stumble highlights how state intervention and chip shortages could hinder China’s chase of U.S. dominance, even as its models remain competitive for cost-sensitive applications in Asia.
(Word count: 278)