Photo credit: www.cnbc.com
Nvidia’s GTC Conference Highlights: Jensen Huang Unveils High-Performance Chips
Nvidia’s CEO Jensen Huang delivered an impactful keynote address at the GTC AI Conference held in San Jose, California, on March 18, 2025. His remarks emphasized the importance of obtaining the most advanced chips the company offers, signaling a strong push towards accelerating performance in artificial intelligence (AI) applications.
During the unscripted two-hour presentation, Huang addressed the critical concerns of clients regarding the cost-effectiveness and return on investment associated with Nvidia’s graphics processing units (GPUs). He posited that these issues would be alleviated through the introduction of faster chips capable of supporting simultaneous AI services for millions of users.
“Over the next decade, we anticipate significant performance improvements, making speed the most effective cost-reduction strategy,” Huang stated in a subsequent discussion with journalists. This highlights Nvidia’s commitment to staying at the forefront of AI technology.
The conference featured a segment where Huang shared insights into the economic advantages of enhanced chips tailored for cloud service providers. He engaged in an “envelope math” demonstration to illustrate each chip’s cost relative to token production, a key metric for assessing AI output expenses.
Huang explained that this financial breakdown is crucial for hyperscale cloud and AI companies currently grappling with investment strategies. Nvidia’s upcoming Blackwell Ultra systems are projected to offer data centers a revenue potential 50 times greater than the earlier Hopper systems, significantly boosting efficiency in AI services.
Investor concerns linger regarding the ability of major cloud providers—Microsoft, Google, Amazon, and Oracle—to sustain their rapid capital investment in costly AI hardware. While Nvidia does not disclose specific pricing for its AI chips, industry analysts estimate that the Blackwell GPU could reach around $40,000 each.
To date, the four leading cloud providers have acquired 3.6 million Blackwell GPUs, adopting a new counting method that equates each Blackwell to two GPUs. This marks an increase from the 1.3 million Hopper GPUs previously sold, showcasing increasing demand for Nvidia’s technology.
Looking further ahead, Huang revealed plans for the development of 2027’s Rubin Next and 2028’s Feynman AI chips, aiming to provide clarity for cloud clients already investing in expansive data center projects. “We are aware that as we speak, several hundred billion dollars are slated for AI infrastructure development,” he remarked, noting that clients are ready with budgets, approvals for power, and land acquisitions.
Huang commented on the competitive landscape, downplaying the impact of custom chips developed by cloud service providers, which he believes lack the adaptability required for evolving AI algorithms. He expressed skepticism about the commercial viability of many recently introduced custom AI chips, often referred to as ASICs, citing the prevalent trend of cancellations in that sector.
“A lot of ASICs get canceled. The ASIC still has to be better than the best,” Huang stated, reaffirming his commitment to ensuring that major projects leverage the most advanced Nvidia systems available.
“So the question is, what do you want for several $100 billion?” he posed, underscoring the significance of selecting cutting-edge technology for substantial investments in the AI field.
WATCH: CNBC’s full interview with Nvidia CEO Jensen Huang
Source
www.cnbc.com