Yazılar

Nvidia CEO Jensen Huang Reports Surging Demand for Blackwell Chips

/in /tarafından

Nvidia CEO Jensen Huang said on Saturday that demand for the company’s cutting-edge Blackwell AI chips is “very strong,” as the semiconductor giant deepens its partnership with Taiwan Semiconductor Manufacturing Co (TSMC) to meet soaring global demand.

Speaking at an event hosted by TSMC in Hsinchu, Taiwan, Huang said Nvidia’s Blackwell platform — which integrates GPUs, CPUs, networking, and switching systems — requires an extensive supply of wafers and components. “We build the GPU, but we also build the CPU, the networking, the switches… there are a lot of chips associated with Blackwell,” he explained.

TSMC CEO C.C. Wei confirmed that Huang had “asked for wafers,” but declined to disclose quantities. “TSMC is doing a very good job supporting us,” Huang said, emphasizing that Nvidia’s record-breaking success “would not be possible without TSMC.”

In October, Nvidia became the first company to surpass a $5 trillion market value, prompting Wei to call Huang a “five-trillion-dollar man.”

When asked about supply challenges, Huang acknowledged there would be “shortages of different things,” though memory makers SK Hynix, Samsung, and Micron have expanded capacity to meet demand. Nvidia has already received next-generation memory samples from all three suppliers.

SK Hynix recently said it had sold out all production for 2026, forecasting a long-lasting chip “super cycle” driven by AI growth. Samsung is also in “close discussion” to supply HBM4 memory to Nvidia.

Huang reiterated that Nvidia has no active discussions to sell Blackwell chips to China, as U.S. restrictions remain in place.

China’s AI Strategy Leans on Huawei Chip Clusters and Cheap Energy to Counter the U.S.

/in /tarafından

China has found a powerful workaround to the U.S. chokehold on advanced semiconductors — combining Huawei’s massive chip clusters with abundant cheap energy to accelerate its artificial intelligence (AI) ambitions.

While Nvidia remains the global gold standard for AI chips, U.S. export restrictions have cut China off from the American company’s most powerful processors. Yet, Chinese tech giants like Huawei, Alibaba, and DeepSeek continue to build large-scale AI models using domestically produced hardware.

At the core of this effort is Huawei’s Ascend series — less advanced than Nvidia’s GPUs individually, but competitive when linked together in vast, high-speed “clusters.” One example is the Huawei CloudMatrix 384, which connects 384 Ascend 910C chips to deliver performance rivaling Nvidia’s GB200 NVL72, despite relying on five times as many chips.

“This approach leverages high-speed interconnects to compensate for weaker chips,” said Brady Wang, associate director at Counterpoint Research. “It suits China’s strengths — large-scale engineering and manufacturing.”

The tradeoff is power consumption. Huawei’s architecture demands far more energy than Nvidia’s — but China’s cheap and plentiful electricity turns that disadvantage into an asset. Supported by investments in solar, wind, and nuclear energy, as well as local government subsidies, Beijing has created a favorable environment for energy-intensive AI infrastructure.

“Less efficient chips are sustainable in China because energy is inexpensive and government-backed,” said Wendy Chang of the Mercator Institute for China Studies.

Still, a structural weakness remains. Huawei’s chips are made by SMIC, China’s top semiconductor foundry, using older 7-nanometer tools that lag far behind TSMC’s cutting-edge technology. Export restrictions, especially on ASML’s extreme ultraviolet lithography machines, limit China’s ability to close that gap.

“China’s main challenge isn’t scaling power or hardware clusters,” said Hanna Dohmen from Georgetown University’s CSET. “It’s whether they can keep up technologically as Nvidia and TSMC push performance forward.”

For now, though, Beijing’s combination of Huawei’s hardware muscle and low-cost power is proving enough to keep China in the global AI race.