In This Article:
Although the emergence of China's DeepSeek R1 model contributed to a 17% stock plunge in Nvidia's (NVDA, Financial) shares, the company recognized China's DeepSeek's recent AI advancements and considered the model an excellent AI advancement. The open source R1 model, reportedly outperforming top U.S. models like OpenAI's, has kids shaking up the AI: DeepSelf claims training costs as little as $6 million vs billions by U.S. tech giants.
Nvidia nevertheless sees the success of DeepSeek as an opportunity to grow itself. However, the company noted that the breakthrough would probably create additional demand for Nvidia's graphics processing units, or GPUs, which are necessary for AI inference tasks. However, Nvidia pushed back against any claim that DeepSeek's R1 model - which used fully compliant Nvidia GPUs, made use of banned Nvidia hardware and said that Nvidia had put Nvidia GPUs in fully compliant units for export controls.
DeepSeek's R1 model release has highlighted how little the current AI spending can produce. Now, analysts are wondering if Microsoft's (MSFT, Financial) and Meta's (META, Financial) major investments in Nvidia-based infrastructure could fall on hard times if cheaper, better options emerge. Last week, Nvidia's CEO Jensen Huang highlighted the idea of 'test time scaling', a central concept in DeepSeek's model and other leading AI systems, from pretraining to more efficient posttraining.
This article first appeared on GuruFocus.