NVIDIA has rebounded significantly; has the "DeepSeek" impact ended on Monday?

After Deepseek triggered significant market volatility, Goldman Sachs stated that the key now is to understand the extent to which Deepseek poses a threat to the ability of hyperscale enterprises to monetize their AI capital expenditures, as well as how advanced the R1 model actually is.

Goldman Sachs analyst Edoardo Lorenzo Greco mentioned in a recent report that Deepseek is unlikely to disrupt the announcement schedule for AI capital expenditures in the upcoming earnings season.

Hyperscale data center operators are likely to still view AI as an arms race, where those who make progress first and have the most advanced technology reap the greatest rewards. However, the focus of discussion will shift more from infrastructure to application areas.

While the results that Deepseek can achieve depend on previous training models (Meta Llama and Alibaba's Tongyi Qianwen), it is clear that the debate over resource efficiency has now become the focal point.

Goldman Sachs believes that if GPUs can be more efficient in the later stages of training, the demand for the most advanced GPUs will slightly decline. This means that the investments made by hyperscale enterprises so far can have a longer payback period. Those who originally thought they could use top GPUs for six years may find their actual usage duration to be longer. Under other unchanged conditions, this is a positive for almost all participants except NVIDIA.

Regarding the impact of inflation, there is currently no clear consensus. The market debate centers on whether AI will overall lead to inflation (for example, cloud computing, where the cost per unit of storage may decrease, but as the required storage increases, the total price paid will continue to rise) or, like most technologies, will lead to deflation. However, Deepseek marks the beginning of the debate on "good enough" Gen AI technology deflation.

Goldman Sachs advises investors that for most AI intelligent agents, perfect AI is unlikely to be necessary; suddenly "good enough" means that pricing power shifts from AI suppliers to AI providers, and software companies like Salesforce and ServiceNow may benefit.

Analysts warn that NVIDIA and the entire semiconductor supply chain will face more uncertainty.

In the long run, the demand for more computing power may not change, but the speed of achieving this goal may be questioned. The demand for demonstrating AI monetization use cases may shift from monetizing AI agents to proving the ability to monetize AI at the highest end through robotics and automation (or at least before facing commoditization pricing pressure in these areas).

For NVIDIA, their CUDA language remains the preferred programming language, and their products are still core. Uncertainty will remain high.

Meanwhile, short-term electricity demand may decrease.

In addition to cooling still being one of the biggest drivers of expected growth in electricity demand, the market has re-priced and lowered expectations in the short term. The declines of GE Vernova and Siemens Energy have exceeded those of any hyperscale enterprise or electrical company