All Research

Home > Research > DeepSeek – The Unexpected Player in AI

DeepSeek – The Unexpected Player in AI

On January 27, 2025, the launch of R1 – Chinese start-up DeepSeek’s AI model – sent shockwaves through the technology markets. DeepSeek claims its open-source model can achieve equivalent or even better performance against some LLMs benchmarks with almost 90% less cost compared to Western companies’ models (explicitly rivaling OpenAI’s o1) – all built within two months. In many benchmarking tests, DeepSeek outperforms the competition in math performance and code assistance/generation, and it demonstrates on par performance when it comes to English.

In addition to the tech markets, the energy sector was also hit. The anticipated energy required to power the AI boom has led to major provider companies looking to invest in their own nuclear energy (such as Microsoft and AWS). However, DeepSeek – in being limited to NVIDIA’s reduced-capability chips (H800s) owing to US sanctions – built workarounds to produce a more efficient solution. Undoubtfully, necessity is the mother of invention; DeepSeek derived innovative methods to leverage the most out of available hardware while drastically reducing their costs. DeepSeek has shown that there is still room for algorithmic efficiency before advancements in hardware are required.

DeepSeek continued to make the headlines with their release of Janus Pro, an open source multimodal model. DeepSeek claims that their largest model (Janus-Pro-7B) can produce images that outperform OpenAI’s DALL-E 3 and StabilityAI’s open-source Stable Diffusion XL.

Our Take

Currently, there is little knowledge of DeepSeek’s company fundamentals and what their next move will be. For instance, DeepSeek may follow OpenAI's strategy and make the next iteration of their model closed source. While organizations may have some hesitation about utilizing DeepSeek, owing to its association with China, it is possible to locally host or integrate DeepSeek (via API) into workflows. While there is certainly some hype, it is perhaps too early to jump on the boat completely. Since DeepSeek’s models are presently open source, it is likely we are going to see other vendors modify their upcoming releases against R1’s algorithmic operations.

Moreover, organizations may witness a slight shift toward cost efficiency. Almost all AI models have been competing based on benchmark scores, with investments pouring in to keep moving the needle. DeepSeek’s low-cost operational efficiency, and (now) public attention will force other companies to respond quickly.

Any AI initiative can be a significant investment for organizations and require detailed assessments before any undertaking. If you are struggling with figuring out which LLM solution aligns with your organization’s needs or evaluating various platforms’ performance, Info-Tech's Leverage Metrics and Benchmarks to Evaluate LLMs blueprint can guide you through the complexities of LLM provider selection.

Want to Know More?


All Research