Katarzyna Petru
NVIDIA is not slowing down in the race for dominance in the artificial intelligence segment. The company, which has become a pillar of AI infrastructure and the highest-valued tech company in the world in recent years, is set to present products that aim to set a new direction for the entire industry. During the NVIDIA GTC 2026 conference in San Jose, Jensen Huang is expected to showcase chips like we have never seen before. The details remain a mystery, but the context clearly points to the data center and AI accelerator segment.
Rubin, Feynman, or a completely new architecture?
In the industry, two main hypotheses are circulating. The first assumes the development of the Rubin architecture, which is meant to be the natural successor to the Hopper and Blackwell lines. It is possible to showcase a more advanced version of the systems designed for operation in hyperscale data centers.
The second option is much more exciting. It concerns the Feynman generation, which has been speculated about for months. According to unofficial information, the new architecture may utilize greater integration of SRAM memory directly within the chip structure, 3D stacking technology to connect layers of chips, and potential new computational units optimized for inference. If these assumptions are confirmed, NVIDIA could radically increase throughput and reduce latency, which today is one of the biggest challenges in handling generative models.
The AI Market is Changing Its Needs
Back in the Hopper and Blackwell era, training large language models was key. Currently, the focus is shifting towards inference, which is the rapid generation of responses in real-time.
This means new bottlenecks. It's not just about raw computational power anymore, but about:
memory bandwidth,
minimizing latency,
energy efficiency,
scalability in data centers.
NVIDIA needs to design its chips for completely different scenarios compared to just two years ago. The flexibility of architecture and component integration is becoming essential.
If we indeed see a new generation of chips at GTC 2026, it could reshuffle the AI market once again. NVIDIA already dominates the computing infrastructure for the largest technology companies. Each subsequent technological advantage strengthens its position against competitors developing their own accelerators. March 15 could therefore turn out to be one of the most important days for the semiconductor and data center sector in 2026.
Source: Techpolis.pl
