The DeepSeek-R1 Shock — A Low-Cost Chinese Reasoning Model Rocks the Market

On 20 January 2025, the AI company DeepSeek, based in Hangzhou, China, released DeepSeek-R1—an open-weights reasoning model—under the MIT license.

A week later, on 27 January, NVIDIA's share price fell about 17% in a single day, erasing roughly US$589 billion in market value. It was the largest single-day loss for any company in the history of the US stock market.

What DeepSeek Is

DeepSeek (深度求索) was founded in July 2023. Its founder Liang Wenfeng also runs the Chinese hedge fund High-Flyer Capital—DeepSeek began as the fund's side project.

In about eighteen months it released a reasoning model rivalling OpenAI's o1. Technically and economically, that shook the industry's assumptions.

Why It Shocked

1. Performance. R1 matched or exceeded OpenAI o1-1217 on reasoning benchmarks such as MATH, AIME, and Codeforces. The message: reasoning models are not the exclusive property of closed labs.

2. Open weights. The entire model file was released under MIT. Local execution, derivative models, commercial use—all permitted. OpenAI's o1, by contrast, lives behind a closed API.

3. Reported training cost. DeepSeek's paper reported about US$5.6 million for the V3 base model (≈2,048 H800 GPUs for roughly two months). Compared with the reputed hundreds of millions for OpenAI- or Anthropic-class models, this is two orders of magnitude lower. Even if the figure understates real cost, the gap is at least one order.

4. Hardware under US export controls. Training used NVIDIA H800, the down-spec variant the US created for export to China. The implication—"the regulated, weakened chip caught up with the regulator's own frontier"—became a question about the regulatory policy itself.

5. Reinforcement-learning innovation. R1 acquired reasoning capability from pure reinforcement learning (the GRPO algorithm) without supervised fine-tuning. Academically novel methodology in its own right.

27 January — The Market

Through the weekend, US markets absorbed the DeepSeek news. On Monday 27 January:

• NVIDIA: −16.97% (−US$589 billion in market cap, largest single-day loss in history)
• Broadcom: −17.4%
• Oracle: −13.8%
• The wider AI-exposed basket: roughly −US$1 trillion

What the market read: if AI training is far cheaper than assumed, future NVIDIA GPU demand contracts; if frontier-class models ship open-weights, OpenAI/Anthropic API margins compress; the US chip export-control regime may have strengthened rather than slowed China.

Aftershocks

The shock outlived a single trading day.

Versus Stargate. Days earlier, on 21 January, the Trump White House had announced the Stargate Project (OpenAI + Oracle + SoftBank, up to US$500 billion). "Four days after the US pledges $500 billion, China releases a comparable model for $5.6 million" became a meme—and a serious doubt about scale-first strategy.

Export-control re-debate. The strengthened end-of-Biden controls on chip exports to China came under renewed congressional scrutiny.

US firms' strategy shift. OpenAI accelerated release of its reasoning model o3 (31 January). Anthropic and Google increased investment in reasoning models.

Open-source revival. Llama (Meta), Mistral, Qwen (Alibaba), and the broader open-weights camp gained momentum from R1's gravitational pull.

What It Demonstrated

DeepSeek-R1 demonstrated a form of democratisation of AI development. The frontier was no longer the exclusive province of the US tech giants. A mid-sized Asian startup, working under regulated hardware, on a hundredth of the budget, could produce a model at the same level.

If ChatGPT's release (November 2022) was a vector of "AI to every household", DeepSeek-R1 was a vector of "AI development to mid-sized companies anywhere". The industry's terrain flattened by one more degree.

How that flattening will combine with regulation, national strategy, and market logic is still an unfolding story.