On-Device AI Accelerating Beyond Expectations, Temporary Uncertainty in Cloud AI Compute Growth
TSMC and Nvidia expected that on-device AI would grow significantly in 2026. TSMC's earlier earnings
call indicated that on-device AI trends would become prominent in 2026, while Nvidia planned to begin mass production of their AI PC chips N1X/N1 in 4Q25/1H26.
After DeepSeek went viral, the on-device AI trend has been speeding up.
DeepSeek's surge in popularity has directly boosted the demand for Nvidia H100 training, demonstrating that optimized training approaches (which effectively reduce costs) can drive training demand. It also
reaffirms the advantages of the CUDA ecosystem (explaining why users opt for H100).
More significantly, DeepSeek's popularity has sparked a surge in local LLM deployment. DeepSeek R1's optimized training approaches enhance the performance of small to medium-scale LLMs on local
devices, while data security concerns about cloud-based DeepSeek services further fuel this trend.
More open-source models similar to DeepSeek are expected to emerge, sustaining the momentum in
local LLM deployment.
Common approaches and specifications for local DeepSeek deployment include: LM Studio for easier
deployment, Ollama for running models, 4-bit/8-bit quantization to reduce VRAM requirements while
maintaining performance, models ranging from 1.5B to 70B parameters, and hardware spanning from
low-end laptops to high-end PCs equipped with Nvidia discrete GPUs.
Currently, local DeepSeek deployment remains limited to a niche user base, having no immediate impact on Nvidia's cloud AI chip demand. In the long term, while on-device computing will partially replace cloud services, it may also generate new cloud demands (as seen with the H100 example). Both segments are expected to grow simultaneously, evolving into an integrated AI ecosystem.
While I maintain a positive outlook on long-term cloud growth, it's important to note that the on-device
trend is picking up faster than expected, which may cause the cloud's growth rate to fall below previous market expectations for a time and potentially affect investment sentiment. Looking ahead, factors that
could lower cloud growth uncertainties include accelerated scaling laws due to successful GB200 NVL72 mass production and increased visibility in AI commercialization across robotics, autonomous driving,
and multimodal applications.
Regarding TSMC and Nvidia mentioned earlier, TSMC remains a primary beneficiary of on-device AI
trends (due to processor upgrades), while Nvidia faces significantly more competition in the on-device
market compared to cloud computing, which could impact near-term investment sentiment.
$Taiwan Semiconductor(TSM.US) $NVIDIA(NVDA.US)
The copyright of this article belongs to the original author/organization.
The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.