| | | DeepSeek model highlights potential pivot to rise of on-device AI chips
Feb. 15, 2025 8:00 AM ET
By: Brandon Evans, SA News Editor
The release of DeepSeek-R1 last month prompted temporary volatility among tech stocks, as its creators boasted the cutting-edge reasoning model was made at a fraction of the price of similar models and required less computing power to train and run.
A white paper published this week by Qualcomm (NASDAQ: QCOM) finds the rise of DeepSeek signals a possible turning point in AI development.
"This pivotal moment is part of a broader trend that underscores the innovation in creating high-quality small language and multimodal reasoning models, and how they're preparing AI for commercial applications and on-device inference," the research said. "The fact that these new models can run on devices accelerates scale and creates demand for powerful chips at the edge."
The research finds that smaller AI models have superior performance, models are decreasing in size, the increase in the number of models is assisting developers in creating better applications and personalized multimodal AI agents are becoming a new type of user interface. These advancements are expected to decrease the cost of training new models and make AI more accessible.
"Many popular model families including DeepSeek R1, Meta (NASDAQ: META) Llama, IBM (NYSE: IBM) Granite, Mistral Mistral feature small variants which overdeliver in terms of performance and benchmarks for specific tasks, regardless of their size," the paper said. "The reduction of large, foundational models into smaller, efficient versions enables faster inference, smaller memory footprint and lowers power consumption – all while maintaining a high bar on performance, allowing deployment of such models within devices like smartphones, PCs, and automobiles."
The rise of smaller models could lead to a proliferation of on-device AI across a wide range of industries. It is already occurring to some degree on smartphones, such as Apple Intelligence (NASDAQ: AAPL), and in AI PC's featuring Qualcomm's Snapdragon X Series.
Qualcomm also eyes on-device AI providing technological enhancements in automotive, industrial IoT and networking.
"The distillation of large foundation models has unleashed a surge of smarter, smaller, and more efficient models, empowering industries to integrate AI faster and at scale – increasingly within devices themselves," Qualcomm noted.
Other industry observers and insiders have concluded that creating AI models at lower costs will lead to the proliferation of AI.
"DeepSeek's work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant," according to Nvidia (NASDAQ: NVDA).
"Wisdom is learning the lessons we thought we already knew," former Intel (NASDAQ: INTC) CEO Pat Gelsinger said. "DeepSeek reminds us of three important learnings from computing history: Computing obeys the gas law. Making it dramatically cheaper will expand the market for it ... Engineering is about constraints ... Open Wins. DeepSeek will help reset the increasingly closed world of foundational AI model work." |
|
