GTC Notes: Stay Sensitive, Think About the Next Transformation | 5Y View

Last week's NVIDIA GTC 2024 once again set the tech world abuzz. NVIDIA unveiled its new Blackwell architecture chip platform, showcased its latest advances in artificial intelligence, deep learning, robotics, and graphics processing, and the debut of its first humanoid robot model had the industry electrified. As one of NVIDIA's most important annual tech launch platforms, GTC is widely regarded by the industry as the "AI weather vane."

Today's post comes from field notes taken by a 5Y Capital colleague at GTC. We've compiled some excerpts — hope they spark something for you :)

Author

Yaopeng Xing, Vice President, 5Y Capital

On Scaling Law

The ceiling for Scaling Law in compute and infrastructure remains high (1000x). Over the past decade, rather than AI model scale driving demand for compute, it's been the scaling up of compute that's pushed AI forward. Looking back at the birth of the Transformer architecture, it was designed to let models compress larger-scale training data, adapt to GPU parallel processing, and solve translation problems.

We see this trend continuing. While Moore's Law effects are diminishing, Jensen is thinking about how to boost computing power from a total-system perspective, achieving greater multiplicative effects (process * architecture * memory * networking * software * models). We should spend more time thinking about what major transformations the world will see if available compute increases another 1000x over the next decade.

A Visionary, Resilient, and Agile CEO

Observing Jensen Huang up close at GTC, you see a CEO who "combines rare vision with focused execution." Jensen has never viewed NVIDIA as a chip-selling hardware company. Breaking free from linear thinking, he's always considering how NVIDIA can drive the next technological revolution and create entirely new market spaces. We too should spend more time standing in Jensen's shoes, thinking about truly big opportunities, seeing the systemic revolution in computing architecture from CPU to accelerated computing.

He's equally resilient and agile. Every year at GTC, Jensen flexes technically to showcase experiments in long-term business directions. This GTC, he also laid out his vision for AI-driven drug discovery, general-purpose robotics, autonomous driving, and Omniverse. Like CUDA, many of Jensen's bets show no short-term results. To persist, he's endured enormous pressure and pain — yet remains acutely attuned to new opportunities, sticking to the long-term right direction through continuous practice and trial-and-error.

Most Inspiring Sessions and Exhibits

World Model: Listening to and exchanging ideas with some of the world's brightest minds at the conference, I came away feeling that foundation models are beginning to enter the physical world. Over the next decade, language models, video, 3D gaming, robotics, and autonomous driving will become deeply intertwined. The convergence of these technologies will continuously expand into unprecedented scenarios and products.

Edge Model: Microsoft VP Sebastien Bubeck's talk on small models was, in my view, one of the most entertaining talks of the entire event. The theme "Textbook is all you need" was fascinating — with sufficiently high-quality datasets, small models of 1-2 billion parameters can still excel at many complex tasks. Perhaps in the future, larger-parameter models will serve as tutors to smaller ones, with a new iteration flywheel emerging: large model → synthetic data → curation and training algorithms → more efficient on-device models.

Waiting for the Next Transformation

Today NVIDIA has become the rock star of the global tech industry, but few pay attention to how this company historically survived like an unkillable cockroach in the semiconductor industry, crowded with giants. Many massive transformative trends had their seeds planted 10-20 years ago (CUDA, AlexNet, and so on). Jensen kept those seeds alive, let them grow and sprout, until the transformation broke through its tipping point. I believe that in some inconspicuous corner of this year's GTC, another great seed has been planted. Investing works the same way — stay sensitive, stay resilient, shed inertial thinking, and think about what transformation comes next.

Sunrise in Los Altos, deer, new life.

Postscript & Reflections

1. The expanding event horizon of the black hole: Transformer + GPU compute acts like a black hole, pulling most AI talent into scaling up. What important work is being neglected today?

2. Cracks in the empire: The war over compute and models has evolved into a power game among giants. Which giants' potential is being underestimated, and which opportunities are being severely overestimated?

3. Emergence of intelligence: Just as people in the 1970s couldn't grasp what PCs were for, when intelligence and compute become accessible everywhere, what entirely new business forms will emerge?

Would love to hear your thoughts.

Engagement Giveaway

What are your reflections and perspectives on the three questions above, and what insights do you have on the full article? Share in the comments — we'll select two standout responses to receive a 5Y Capital portfolio company gift package :)

5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing support from the spiritual to all operational aspects. We believe that if the "crazy you" in others' eyes begins to be believed in, the world will become a different place.

BEIJING · SHANGHAI · SHENZHEN · HONG KONG

WWW.5YCAP.COM