The report critiques Apple's model, which employs a mixture of experts (MoE) approach, stating that its 3 billion data parameters are insufficient for effective consumer utility.

Jefferies has expressed skepticism about the future of AI smartphones, highlighting the necessity for hardware upgrades to support larger models, which would ultimately increase costs.

In light of DeepSeek's achievements, Jefferies outlines two potential strategies for the AI industry: one focusing on enhancing computing power for better model performance, and the other shifting towards efficiency and return on investment (ROI) with reduced computing demands by 2026.

The report emphasizes ROI concerns in AI, noting that DeepSeek's efficient model could influence future AI development strategies.

Jefferies raised alarms regarding the demand for computing power, pointing out that significant investments in GPUs, such as NVIDIA's projected revenue exceeding USD 200 billion in 2024, have not yielded substantial returns.

Despite the success of DeepSeek, Jefferies believes it does not significantly improve the near-term outlook for AI smartphones, as the need for hardware upgrades remains a major concern.

DeepSeek's large language model (LLM) is recognized for its efficiency, having been trained at a cost of only USD 5.6 million, making it a competitive alternative to industry leaders like GPT-4.

As investors increasingly scrutinize financial performance, US AI companies may face mounting pressure to justify their rising capital expenditures.

Concerns about ROI in AI are prevalent, with the report indicating that substantial investments in AI technologies have not yet produced the expected returns.

Jefferies highlights that the AI market has not achieved anticipated monetization, leading to skepticism regarding the sustainability of high investments in AI technologies.

Jefferies suggests that smaller variants of DeepSeek's open-source LLM may be more suitable for smartphones, given the low consumer adoption of AI.

The report posits that DeepSeek's advancements could signal a potential to overcome the limitations faced by smaller models like Apple's, indicating a positive outlook for future AI developments.