It would be an understatement to say the AI landscape is changing fast.
Chinese company, DeepSeek, is playing a role in this evolution through the release of its latest artificial intelligence model, which delivers similar capabilities to other AI models like ChatGPT but at a fraction of the cost.
What we know
DeepSeek has provided a blueprint for how to do things better and more efficiently.
Essentially, DeepSeek looks to utilise only that part of the AI model required to perform a particular action, rather than activating the whole model.
This is not a new concept. Meta and other companies have been exploring this for more than a year, but DeepSeek has found a way to make it work.
Its product is in high demand. DeepSeek was the most downloaded free application in both the App Store and Google Play in the two weeks immediately following its launch.
Microsoft, Amazon and Tencent were also quick to add DeepSeek to their platforms.
Where we are uncertain
There are aspects of DeepSeek’s claims that should be scrutinised.
Questions have been asked about the source of training data for its AI model. OpenAI claims that DeepSeek used ChatGPT to train the model, which is ironic given questions raised about the data sources used to train ChatGPT.
Even so, open-source AI models can only operate with access to training data and owners of such data could become more circumspect about how it gets used.
Once fully trained, open-source AI models can be downloaded and applied without access to data. However, they currently need to be retrained on new data for any improvements to be made.
There are also questions about the overall cost of training DeepSeek’s latest AI model and the computing power required.
DeepSeek claimed the cost was US$5.6 million, but many experts believe this only represents a portion of the total expense and doesn’t consider prior research or the cost of previous AI models that laid the foundation for the latest version.
What DeepSeek actually did
AI development is based on the availability of data, the computing power to process data and the algorithms used to optimise the process.
Most advancements to date have been achieved by adding more data and computing power.
Instead, DeepSeek has successfully improved the algorithms that dictate how its AI model learns, analyses data and makes decisions. This innovation reduces the amount of computing power needed.
Where we are today
DeepSeek has significantly lowered the cost curve, which will have positive implications for AI demand and innovation going forward.
Until now, the cloud service providers like Microsoft and Amazon have controlled the pace of innovation. DeepSeek has disrupted this to some extent and accelerated the rate of development.
What industry leaders think
Leaders in the technology sector have given the DeepSeek model a thumbs up, noting its improved reasoning and efficiency.
Andrew Jassy, Amazon CEO, said, “Like many others, we were impressed.”
He further stated that everyone developing AI models is learning from each other, which will lead to further innovation.
Alphabet CEO, Sundar Pichai, was equally impressed and said, “I think they have done very, very good work.”
These sentiments are echoed by many other experts, although there is a wide range of views on the implications.
Large US companies have not been perturbed by the DeepSeek news and plan to invest even more towards the buildout of AI infrastructure.
According to recent announcements, Microsoft, Amazon, Alphabet and Meta expect to spend a combined US$320 billion this year. This is 40 percent higher than their AI spending in 2024.
Chinese technology gets a boost
The DeepSeek breakthrough has proved to be a hit for Chinese technology companies.
It highlights that AI innovation is happening in China and at a pace that could rival progress in the United States. Furthermore, the technology is a positive addition to the cloud services of some Chinese businesses like Tencent and Alibaba.
Lower costs mean more spending
According to the Jevons paradox, lower computing costs will likely result in more overall spending on technology rather than less.
This played out with the adoption of cloud computing. The migration of data to the cloud increased when costs fell, as users saw the benefits of the services offered.
This article was also shared on Newsroom.
Market Insights enewsletter
Keep up to date with our fortnightly Market Insights enewsletter. Our research team provide timely and regular commentary and analysis on market developments, understanding investment jargon, and the impact of current events.
Roger Garrett
Senior Research Analyst (International Equities)
Market Insights enewsletter
Keep up to date with our fortnightly Market Insights enewsletter. Our research team provide timely and regular commentary and analysis on market developments, understanding investment jargon, and the impact of current events.