According to the financial news app Zhitong Finance, investment firm New Street Research's data shows that there is no sign of a slowdown in the demand for artificial intelligence infrastructure in the short term, as higher investment in computing power will result in better performing models.

Analysts led by Pierre Ferragu stated in an industry report: "OpenAI has shown how the size of models has become a heavy burden on inference costs. Although the cost of inference is decreasing rapidly as models mature, the operating costs of the next generation models are much higher than the previous versions (this situation will only get worse with the introduction of inference capabilities)."

Despite the high demand and high sales of companies that provide hardware support for artificial intelligence during this cycle, the increasing costs raise the question of whether it is worth introducing the next level model. Pierre Ferragu said: "Generative artificial intelligence is a business with very expensive incremental costs (for the latest models). This is completely opposite to software, which has zero incremental costs. We are not sure how this will play out, but it will have a profound disruptive impact on the value chain structure."

$Meta Platforms (META.US)$ And startups like xAI, founded by CEO Elon Musk, continue to ramp up their efforts in the ai hardware race. Last week, Musk announced an ai training cluster powered by 0.1 million NVIDIA H100 GPUs. He plans to double that in the coming months. Similarly, meta platforms CEO Zuckerberg plans to have 0.35 million H100 GPUs by the end of this year to support the company's ai projects. $Tesla (TSLA.US)$

Pierre Ferragu said: "As our understanding of the economics of artificial intelligence continues to improve, both on the supply side and the demand side, we believe that the current level of spending is sustainable and there is further significant room for growth. This does not mean that there will not be pauses or adjustments, but rather that we have not yet seen the peak of an unsustainable bubble."

$NVIDIA (NVDA.US)$ Currently, it is winning in the competition to provide necessary processors for the ongoing development of artificial intelligence models. One main reason is NVIDIA's latest GPU's full-stack and system-level integration capabilities, which can achieve greater chip-to-chip bandwidth connectivity in larger domains. New Street Research points out that this is NVIDIA's advantage compared to its competitors. $Advanced Micro Devices (AMD.US)$ This is where NVIDIA's advantage lies compared to its competitors.

However, Pierre Ferragu added, "This does not negate AMD's advantage, because its architecture has poor universality and scalability, but it is still competitive, just with more limited use cases and requires dedicated optimization efforts." Perhaps AMD also understands this, which is also one of the reasons why the company recently acquired ZT Systems.