Goldman Sachs believes that the key issue currently is how much of a threat Deepseek poses to the monetization of AI capital expenditures, and how advanced the R1 model actually is. If the use of GPUs can be more efficient, then the demand for the most advanced GPUs will slightly decline. The investments made by hyperscale companies so far can have a longer return period. What was originally thought to be the lifespan of six years for top-tier GPUs may actually last longer in use. Under other unchanged conditions, this is bullish for almost all participants except NVIDIA.
After Deepseek triggered severe market fluctuations, 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, and how advanced the R1 model actually is.
Goldman Sachs Analyst Edoardo Lorenzo Greco stated in a recent report that Deepseek is unlikely to disrupt the announcement schedule for AI capital expenditures in the upcoming Earnings Reports season.
Hyperscale Datacenter operators are likely to still view AI as an arms race, with the largest rewards going to those who make the fastest advancements and possess the most advanced technologies. 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 ( $Meta Platforms (META.US)$Llama and $Alibaba (BABA.US)$The discussion about resource efficiency has evidently become the focus.
Goldman Sachs believes that if GPUs can be more efficient in the later stages of training, the demand for cutting-edge GPUs will slightly decrease. This means that the investments made by hyperscale companies so far can have a longer return period. Those who originally thought they could use top-tier GPUs for six years may actually use them longer. Under unchanged conditions, this is Bullish for almost all participants other than. $NVIDIA (NVDA.US)$
Currently, there is no clear consensus regarding the impact of inflation. The market's focus of debate is whether AI will overall lead to inflation (for example, in Cloud Computing, the cost per unit of storage may decrease, but as the required storage volume increases, the total price paid continues to rise) or, like most technologies, will lead to deflation. However, Deepseek marks the beginning of the "good enough" Gen AI technology deflation debate.
Goldman Sachs warns investors that for most AI intelligent agents, perfect AI is unlikely to be needed; suddenly, "good enough" means the pricing power shifts from AI suppliers to AI providers, similar to.$Salesforce (CRM.US)$、$ServiceNow (NOW.US)$Software companies might benefit.
Analysts warn that NVIDIA and the entire Semiconductors Industry Chain will face more uncertainties.
In the long term, the demand for more computing power may not change, but the speed of achieving this goal may be called into question. The demand for demonstrating AI monetization use cases may shift from monetizing AI agents to proving the ability to monetize AI at the highest level through Siasun Robot&Automation (or at least facing pricing pressure from Commodities in these areas before that).
For NVIDIA, their CUDA language remains the preferred programming language, and their products remain core. Uncertainty will still be high.
Meanwhile, the demand for Electrical Utilities may decrease in the short term.
Aside from cooling, which remains one of the biggest drivers for expected growth in Electrical Utilities demand, the market has repriced and lowered expectations in the short term.$GE Vernova (GEV.US)$ and $SIEMENS AG (SIEGY.US)$ The decline exceeds that of any mega-scale enterprise or Electrical Utilities.
Editor/danial
Comment(2)
Reason For Report