NVIDIA's Big Bets: More Seats in Taiwan, AI Chip Fights, and a New "Law"
NVIDIA chief Jensen Huang has been dominating the headlines lately. It's not exclusively for groundbreaking new GPUs. He recently offered some insight into the company's expansion to Taiwan, the ongoing game of tug-of-war over AI chip strategies with China, and his lofty declaration regarding the end of an old tech tenet.
Why A New "Constellation" HQ in Taiwan. Just A Matter of More Chairs.
You might think a new global headquarters is all about grandiose strategic statements. But in Jensen Huang's view, the main driver behind NVIDIA's new "Constellation" headquarters in Taiwan is infuriatingly simple: they just need more space. He quipped their office was so cramped, "If one employee sits down, another has to stand."
This new center, slated for the Beitou Shilin Science Park, highlights Taiwan's important position in NVIDIA's operations and the wider tech manufacturing universe. Huang stressed that Taiwan is essentially a second home to NVIDIA and that international stability is inextricably linked with Taiwan's own. In addition to office space, NVIDIA is also partnering with Foxconn and TSMC to construct a strong Blackwell-based AI supercomputer in the area, indicative of long-term investment in the region.
The China AI Chip Dilemma: Navigating Treacherous Waters
The conversation also came around to the complicated trilateral relationship between NVIDIA, the US government, and China's quickly expanding AI aspirations. Huang has previously warned that US restrictions on AI chip exports to China threaten to drive China into developing an autonomous AI ecosystem of its own. He considers this potentially harmful to US dominance in AI and, of course, to NVIDIA's trade.
Yet it seems the US government, despite players like President Trump, isn't flexing much on these restrictions. Senior policy advisors have said there are continuing concerns regarding what could be done with these super GPUs when they are physically in China. In spite of NVIDIA's heavy investments and Jensen Huang's persuasion, the White House remains resolute, even prohibiting the exportation of specific accelerators such as the H20. This puts NVIDIA to walk a thin line, likely developing some AI products specific to the Chinese market within these constraints, a Blackwell-based version perhaps not far behind.
Is Moore's Law Dead. Enter "Huang's Law".
One of Jensen Huang's most talked-about comments is his version of Moore's Law. For decades, Moore's Law pointed out that the number of transistors on a microchip doubles about every couple of years, leading to corresponding performance increases. Huang, however, believes this observation no longer holds in today's era of AI.
He argues that physical limitations mean that simply reducing nodes is not resulting in the kind of performance gains NVIDIA is experiencing. Instead, he points to such technologies as with advanced packaging technology (e.g., TSMC's CoWoS) and NVIDIA's NVLink interconnect technology as being the real drivers of change. These technologies enable chip performance scaling via interconnecting numerous platforms, outside the capability of simply reducing nodes.
While NVIDIA is developing on a hyper-product cadence, new architectures emerging as quickly as every six months, Huang foresees the "sky is the limit." He even suggested that NVIDIA might advance its road map to a quarterly cadence – a pace virtually unknown in the semiconductor industry. It seems, according to Jensen, we're beyond Moore's Law and perhaps into a period governed by a new set of principles for AI innovation.
