Within a week of being launched, DeepSeek is now leading Apple download charts for AI search tools, Science Correspondent Martin Stew on what it is and what it means.
Financial analysts are warning investors could lose more than $1 trillion (£798 billion) today because of a Chinese startup, DeepSeek, threatening to shake up the world of artificial intelligence.
The app has jumped to the top of the download charts in the UK, US and China, sparking a rethink of the landscape of the AI sector.
Until now, the US – home to OpenAI’s ChatGPT, Google’s Gemini, as well as Meta and Microsoft – has been seen as the immovable global leader in AI.
But DeepSeek’s sudden emergence, and claims of performance on par with ChatGPT despite having been developed for a fraction of the cost, has spooked the markets.
Shares in AI-related firms based in the US were down on Monday ahead of Wall Street opening.
So what is DeepSeek and how does it work?
What is DeepSeek R1?
It’s a Chinese-made. AI-powered search tool similar to ChatGPT.
It appears to match or even surpass ChatGPT in ability but does it using a fraction of the computing power.
Within a week of being launched, DeepSeek is now leading Apple download charts for AI search tools.
Whilst ChatGPT cost hundreds of millions to compute, DeepSeek’s creators claim they have spent less than £5 million.
How does it work?
Compared to other Large Language Model (LLM) programmes like ChatGPT (OpenAI), Bard (Google), LaMDA (Google) and Llama (Meta), DeepSeek appears to do things differently.
According to Morgan Brown, the author of Hacking Growth – a book about how big corporations managed their breakthroughs – DeepSeek “built an ‘expert system'”.
“Instead of one massive AI trying to know everything (like having one person be a doctor, lawyer, and engineer) they have specialised experts that only wake up when needed.”
“Traditional models? All 1.8 trillion parameters active all the time.”
“DeepSeek? 671 billion total but only 37 billion active at once. It’s like having a huge team but only calling in the experts you actually need for each task.”
Brown goes on to explain that the results are “mind-blowing”.
Here’s how DeepSeek stacks up compared to ChatGPT:
- Training costs: $100M → $5M
- GPUs (graphics processing units) needed: 100,000 → 2,000
- API (application processing interface) costs: 95% cheaper
- Can run on gaming GPUs instead of data center hardware
Why does this matter?
Computing and computer chips are big business.
AI requires huge amounts of computing power. The more data it absorbs and the more people who use it, the more data centres and supercomputers are required to run it.
DeepSeek is a huge challenge to the dominance of western big tech companies. It has made its coding open source meaning others can copy it, opening up huge opportunities for other startups.
It also appears that DeepSeek doesn’t require the cutting edge semiconductor microchips which the West have monopolised – and prevented from being sold to China.
Shares in Nvidia, which makes the leading microchips, fell 4.3% at one stage on Monday.
Another huge potential is in reducing energy use.
By 2030 it’s estimated that computers powering AI – shown in orange on this graph – will use as much electricity as the USA does in total.
If you cut the computing power required, you reduce the impact on the environment.
Follow STV News on WhatsApp
Scan the QR code on your mobile device for all the latest news from around the country
