AIster · The agentic age · Launches
DeepSeek R1 arrives and NVIDIA posts a record one-day loss
A low-cost, openly licensed Chinese model made investors question AI spending
Key facts
- Date
- 20–27 Jan 2025
- Era
- The agentic age
- Interest
- Launches
- Threads
- Model wars, Open vs closed, Compute
- Who
- DeepSeek, a Chinese AI lab
- Released
- 20 Jan 2025, MIT licence
- Market reaction
- NVIDIA down nearly 17% on 27 Jan 2025
- Value lost
- About $589–593 billion in one day
What was released
DeepSeek, a Chinese AI lab, published DeepSeek-R1 on 20 January 2025. Its release note claimed “performance on par with OpenAI-o1” on maths, coding and reasoning tasks, and credited large-scale reinforcement learning for the gains. The weights were released under the MIT licence, which the company summed up as letting people “distill & commercialize freely”. It also published six smaller distilled models (compact versions trained on R1’s output), including 32B and 70B versions it said matched OpenAI's o1-mini.
The API, which developers use to build the model into their own software, was priced at $0.55 per million input tokens (on a cache miss) and $2.19 per million output tokens.
The app and the price tag
Over the following week, DeepSeek's free assistant app climbed the charts. By Monday 27 January Reuters reported that it had surpassed ChatGPT in downloads from Apple's App Store in the US. Attention also fell on DeepSeek's reported training cost of about $5.6 million for its underlying model, set against the tens of billions US companies were committing to data centres.
The sell-off
That Monday, investors questioned whether the AI build-out needed as many chips as assumed. NVIDIA's shares fell nearly 17%, erasing about $589 billion in market value by the count cited by Yahoo Finance and $593 billion by Reuters' count, a record one-day loss in market value for a Wall Street stock. The Nasdaq fell 3.1% and the Philadelphia semiconductor index dropped 9.2%, its steepest fall since March 2020.
President Trump called DeepSeek a “wake-up call” for US industry. The investor Marc Andreessen described R1 as “one of the most amazing and impressive breakthroughs I've ever seen”, while analysts at Bernstein said the reaction was “not really worthy of the hysteria”.
Why it matters
R1 showed that a lab outside the largest US companies could release a competitive reasoning model openly and cheaply. It sharpened debates about US export controls, open weights and the economics of AI infrastructure, and it arrived the day before the Stargate announcement in Washington.
Questions people ask
What is DeepSeek R1?
DeepSeek-R1 is a reasoning model released by the Chinese lab DeepSeek on 20 January 2025. The company said it performed on par with OpenAI's o1 and released it under the MIT licence, so anyone can use, modify and commercialise it.
Why did NVIDIA stock drop in January 2025?
On 27 January 2025 investors worried that DeepSeek's cheaply built, high-performing model meant AI would need fewer expensive chips. NVIDIA's shares fell nearly 17%, wiping out roughly $589–593 billion of market value, a record one-day loss.
Sources
- DeepSeek, DeepSeek-R1 Release, 2025-01-20
- Reuters (via Yahoo Finance), DeepSeek sparks AI stock selloff; Nvidia posts record market-cap loss, 2025-01-28
- Yahoo Finance, Nvidia stock plummets, loses record $589 billion as DeepSeek prompts questions over AI spending, 2025-01-28