Katarzyna Petru
NVIDIA isn't slowing down in the race for dominance in the artificial intelligence segment. The company, which has become a cornerstone of AI infrastructure and the highest valued tech company in the world in recent years, is set to showcase 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 present chips like we've never seen before. The details remain under wraps, but the context clearly points to the data centre and AI accelerator segment.
Rubin, Feynman or completely new architecture?
In the industry, two main hypotheses are circulating. The first assumes the development of the Rubin architecture, which is expected to be the natural successor to the Hopper and Blackwell lines. It is possible to demonstrate a more advanced version of the systems designed to work in hyperscale data centres.
The second option is significantly more exciting. It concerns the Feynman generation, which has been speculated about for months. According to unofficial information, the new architecture may utilise greater integration of SRAM memory directly within the chip structure, 3D stacking technology to connect layers of circuits, and potentially new computing units optimised for inference. If these assumptions are confirmed, NVIDIA could radically increase throughput and reduce latency, which today represents one of the biggest challenges in handling generative models.
The AI market is changing its needs
Back in the Hopper and Blackwell era, training massive 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 no longer just about raw computing power, but about:
memory bandwidth,
minimising latency,
energy efficiency,
scalability in data centres.
NVIDIA must design its chips for entirely different scenarios than just two years ago. Flexibility of architecture and integration of components are becoming crucial.
If we see a truly new generation of chips during GTC 2026, it could once again reshuffle the AI market. NVIDIA is already dominating the computing infrastructure for the largest tech 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 centre sector in 2026.
Source: Techpolis.pl
