Huang Renxun: It takes the US three years to build a data center, but China can build a hospital in just one weekend

In recent years, with the continuous expansion of the artificial intelligence (AI) industry, the demand for data centers has reached an unprecedented level. American tech giants such as NVIDIA have launched a surge in building data centers. However, due to the aging infrastructure and slow construction process in the US, the construction of data centers is not going smoothly.

According to a report by the US magazine Fortune on December 6, Huang Renxun, CEO of NVIDIA, said during a conversation with John Hamre, head of the US think tank Center for Strategic and International Studies (CSIS), that although the US still leads in the AI field, its infrastructure construction capabilities are far behind China, which may allow China to overtake it in the "AI competition."

Huang Renxun simplified the AI industry into a "five-layer cake," consisting of energy, chips, infrastructure, models, and applications. He pointed out that in the most fundamental layer of energy, "China has twice the energy of the US." He stated that the US government is pushing for manufacturing to return, "but without energy, how can we build chip factories, supercomputer factories, and AI data centers?"

Huang Renxun also took the opportunity to praise US President Trump, claiming that Trump realized the importance of energy growth and was "pushing through pressure to restore America's energy." But he also admitted that the cost of energy in the US is still much higher than in China.

Regarding the chip issue, Huang Renxun said that the US currently holds an advantage in this area, with American tech companies like NVIDIA having leading AI chip technology, but China has shown great potential. He said, "We have leading technology, but we cannot be complacent. Remember, semiconductors are a manufacturing process, and anyone who thinks China will not be able to manufacture them is completely wrong."

In terms of infrastructure, the US construction speed is completely unable to match China's, Huang Renxun said: "If you want to build a data center in the US, it takes about three years from the start of construction to the completion of an AI supercomputer. But China can build a hospital in just one weekend. We face real challenges in the infrastructure field. China is a builder, and they build things at such a fast pace."

Huang Renxun said that the US still maintains approximately a "six-month" lead in AI frontier models, but China leads far ahead in open-source models. "Without open source, startups cannot thrive, university researchers cannot conduct research, and scientists cannot use AI. Without open source, almost all industries of the economy cannot achieve fundamental progress. Different technologies that enable the flourishing of AI are all open source, and China leads the US in open source."

He reminded Hamre that about 50% of AI researchers globally are Chinese, and last year about 70% of AI patents were released by China. "The AI ecosystem in China is vibrant and extremely innovative; this is a powerful country."

"All these scientists are located in the model and application layers. Now, China is putting all these capabilities into the construction of its own system because the US has left the Chinese market," Huang Renxun said. "They have many top computer scientists and strong software development capabilities. They will work on building their own technology chain. Once China builds this system, they will promote it at a speed you can't imagine."

Huang Renxun believes that the competition in the application layer will determine the outcome of the "AI competition." "AI is about automation. We cannot lag behind in AI applications and promotion. Whoever can apply this technology first and most widely will win this industrial revolution."

Driven by companies like NVIDIA, the AI boom has quickly boosted the demand for data centers. However, the problem of aging infrastructure and slow construction in the US is hindering the country's data center construction. Bloomberg published an article on November 10 stating that two data centers in Santa Clara, California, where NVIDIA is headquartered, have been vacant for years due to local power shortages.

A spokesperson for Digital Realty, the developer of one of the data centers, Jordan Sadler, said that waiting for three years to obtain energy is "a normal pre-requisite time in half of the US." In high-demand areas such as Silicon Valley and Northern Virginia, the waiting time is even longer.

"In Santa Clara, if you find a piece of land today and start applying for new power, you might have to wait for several years," Sadler said. "We don't want to fall behind."

Bloomberg estimates that by 2035, the AI computing power demand in the US alone could more than double. Power supply is becoming the biggest bottleneck for US data centers, mainly due to aging grid infrastructure, slow transmission line construction, and various regulatory and licensing barriers. "This highlights a major challenge in the US tech industry, and even more broadly in the economy."

Original: toutiao.com/article/1850862875086858/

Statement: This article represents the views of the author.