by 
Suggests that the global community requires China's AI experts to work on these chips for everyone else to gain from their advancements.
Computex The CEO of Nvidia, Jensen Huang, has stated that the U.S. ban on exporting his company’s top-tier accelerators to China is "exactly the incorrect approach" and represents "a failure."
During a Q&A session at the Computex conference in Taiwan, Huang expressed his dissatisfaction with the Trump administration's choice to prohibit the sale of Nvidia’s H20 accelerators in China. He was concerned that this move would negatively impact Nvidia’s earnings and profits, thereby reducing their capacity to pay taxes and create jobs within the U.S.
He believes the policy could hurt humanity since approximately half of global AI researchers are based in China, and he considers their contributions to be top-tier worldwide. His aim is for these researchers to utilize Nvidia equipment so that their advancements can benefit people globally.
“He mentioned that DeepSeek was developed using Nvidia technology,” and then referred to it as “a blessing” because of the numerous optimization methods the firm employed for their R1 version.
When R1 made its debut, investors became anxious, causing the stock prices of AI hardware companies to plummet.
Huang believes that the reaction misunderstood the scenario, since AI has progressed beyond what he refers to as "one-off" uses like employing generative AI to address a prompt—a task that demands only a brief surge of computational power. DeepSeek provides reasoning models that necessitate greater computation—Huang contends this means R1 has thus driven innovations that cumulatively boosted the demand for computing resources by 100 to 1000 times over.
He contended that stopping Nvidia from selling to China would result in researchers in the Middle Kingdom ceasing collaboration with the company’s technology, and users worldwide wouldn’t benefit from their advancements. This scenario would lead to decreased sales for Nvidia and consequently lower tax payments.
Huang highlighted that Chinese companies are developing their own AI hardware, indicating that US policies have actually accelerated progress instead of hindering it. He stated consequently that such policies are thus "ineffective."
Haung stated that Nvidia aims to develop accelerators that adhere to U.S. export regulations. However, he added, "We are unsure how to further downgrade the GPU architecture known as Hopper." He went on to say, "Our current focus is on finding ways to optimally cater to the market needs. The extent of degradation applied to this product makes it challenging to meet customer demands effectively."
The CEO similarly expressed approval for certain policies implemented during the Trump administration, particularly the shift in approach towards AI dissemination. This new direction allows the U.S. to export AI-related technology and infrastructure to an increased number of nations. According to Huang, this change will facilitate growth for Nvidia and enable American-developed AI technologies to lead globally.
However, he believes that governments should take further steps to speed up the approvals for energy production initiatives—regardless of the energy source—to ensure that data centers have access to the necessary power for running AI tasks.
He mildly endorsed the Trump administration’s strategy of expanding domestic manufacturing facilities, noting that the surge in artificial intelligence could generate significant demand and open up opportunities for producing more goods within the U.S. Nonetheless, he acknowledged that it would be unfeasible to conduct all manufacturing domestically. During multiple presentations at Computex, he highlighted that Nvidia systems consist of over a million parts sourced globally from various suppliers.
During the Q&A session, Huang promoted Nvidia’s NVlink Fusion, enabling various accelerators from competing manufacturers to utilize Nvidia’s NVLink connection technology.
The CEO mentioned that Nvidia customers have requested expanded usage of NVLink in various areas, as he believes "UALink is not progressing satisfactorily."
UALink is an interconnection specification created by a group of NVIDIA competitors. In April, they released the initial version, although they acknowledged that devices utilizing this technology won’t be available for at least another year.
Huang stated that NVLink Fusion will become popular as customers await UALink, but he downplayed inquiries regarding the latter’s development and future potential.
He similarly refuted claims about Nvidia intending to establish an R&D facility in China, stating that their current location doesn’t have enough room for such expansion.
The reason for needing new offices is that we lack sufficient chairs," he explained. "When one individual is typing, another must cease their work.
Larger office spaces will consequently enhance Nvidia's efficiency.
"There will be additional restrooms, and the queues will be shorter," Huang stated, possibly indicating that certain fundamentally human issues persist beyond AI's capabilities.
No comments:
Post a Comment