Doubts and Prospects of AI Investment Goldman Sachs' Perspective

TapTechNews July 14th news, the world-renowned investment bank Goldman Sachs (Goldman Sachs) has recently raised doubts about the return on investment in artificial intelligence (AI). Although major enterprises and investors are pouring billions of dollars into AI research and development, Goldman Sachs is worried whether such a huge investment can really bring rich returns.

Currently, the training cost of the LLM large language model we use (such as GPT-4o) is as high as hundreds of millions of US dollars, and the training cost of the next-generation model is expected to soar to 1 billion US dollars. Sequoia Capital, a venture capital giant, after calculation, said that the entire AI industry needs to generate 600 billion US dollars (TapTechNews note: currently about 4.36 trillion yuan) in revenue every year to break even, highlighting the high pressure of R & D costs.

In the fierce AI race, tech giants such as NVIDIA, Microsoft, and Amazon have all increased their investment to strive for the lead. However, Goldman Sachs interviewed many experts and got inconsistent opinions.

Some experts hold a cautious attitude towards the prospects of AI, believing that its contribution to the US economy will be very limited and cannot solve complex problems more economically than existing technologies. Professor Daron Acemoglu of the Massachusetts Institute of Technology estimates that generative AI will only increase economic productivity by about 0.5% and GDP by about 1%. This is in sharp contrast to the prediction of Goldman Sachs economists, who expect that generative AI will increase productivity by 9% and GDP by 6.1%.

Acemoglu also pointed out that even if AI technology continues to develop and costs decline, but just by putting more data and computing power into the model, it may not be able to realize our vision of general artificial intelligence faster. "Human cognition involves a variety of cognitive processes, sensory inputs, and reasoning abilities. Although large language models have made impressive progress, it still requires a great deal of imagination to believe that by predicting the next word, the intelligence like HAL9000 in the science fiction movie "2001: A Space Odyssey" can be achieved. To be sure, the current AI models will not be able to approach this level in the next decade."

However, there are also different voices within Goldman Sachs. Senior equity research analyst Kash Rangan and Eric Sheridan believe that although the return on investment in AI may be longer than expected, it will eventually bear fruit. Rangan said: "Each computing cycle follows a development sequence called IPA, that is, infrastructure first, then the platform, and finally the application. At present, AI is still in the stage of infrastructure construction, and it will take more time to find the killer application, but I believe we will make a breakthrough."

Sheridan added: "Compared with previous investment cycles, the prospects of this AI investment cycle seem to be brighter because the trendsetters are no longer start-ups but industry giants, which reduces the risk of technology not being popularized. Giants like Microsoft and Google have huge financial reserves, extremely low financing costs, and a huge distribution network and customer base, which enables them to make more attempts and find ways to make capital pay off."

Despite different views, Goldman Sachs still admits that AI faces two major challenges: chip availability and power consumption. Thanks to NVIDIA being able to deliver chips 2-3 months in advance (previously it took 11 months), the shortage of GPUs in the AI field seems to have eased.

However, power consumption in data centers is becoming a major limiting factor. New AIGPUs consume astonishing amounts of power, and a single GPU can consume up to 3.7 megawatt-hours per year. The total power consumption of all GPUs sold last year alone is enough to support the electricity consumption of more than 1.3 million average American households. In order to meet the demand of huge AI data centers, some large enterprises have even begun to consider using modular nuclear power plants.

Whether AI will thrive like the Internet and e-commerce, or burst like a 3D TV bubble, only time can tell. But it is certain that the development of AI will still be unstoppable. As Goldman Sachs said: "We still look favorably on the theme of AI development, one of the reasons is that AI may fulfill its promise, and the other is that the bubble may take a long time to burst."