In an exclusive interview with the Korea Economic Daily on February 18, 2026, Nvidia CEO Jensen Huang confirmed that the company is preparing several “entirely new chips never seen before” for unveiling at GTC 2026 next month.
The announcement comes as the industry prepares for the Vera Rubin platform to enter full production, but Huang’s comments suggest Nvidia is already looking past current physical limits.
GTC 2026: The “World-Surprising” Reveal
The conference, scheduled for March 16–19, 2026 in San Jose, will center on a keynote where Huang is expected to detail these new architectures.
- Feynman Architecture: While the Rubin platform is the focus for 2026, industry insiders speculate that Huang might provide a first look at the “Feynman” architecture. Set for 2028, Feynman is expected to use 1.6nm (A16) process technology and silicon photonics, using light instead of electricity to move data.
- The “AI SSD”: Reports suggest Nvidia has partnered with SK Hynix to develop a revolutionary “AI SSD” capable of 100 million IOPS. This would solve the “data bottleneck” where AI GPUs are often faster than the storage feeding them.
- HBM4 Integration: Huang described Nvidia and SK Hynix as “one giant team,” working to integrate HBM4 (6th-gen high-bandwidth memory) into the Vera Rubin chips to maximize inference performance.
Pushing Past “Physical Limits”
Huang acknowledged that the pace of innovation is becoming harder as the industry hits the physical boundaries of silicon.
| Concept | Status (Feb 2026) |
| Current Lead | Blackwell (full deployment) |
| Incoming | Vera Rubin (Successor to Blackwell; shipping H2 2026) |
| The “Surprise” | Speculated Rubin Ultra or Feynman prototype |
| Primary Goal | 10x reduction in the cost of generating AI tokens |
Strategic Shift: The AI Infrastructure Project
Huang dismissed “AI bubble” concerns, stating that we are only at the beginning of a “tens of trillions of dollars” infrastructure project.
- Acquisition Strength: To bolster its “non-chip” AI stack, Nvidia recently completed a $20 billion acquisition of Groq, integrating their LPU (Language Processing Unit) talent to dominate the AI inference market.
- Energy Focus: In a surprising move, Nvidia also invested $860 million in the fusion startup Commonwealth Fusion Systems, signaling that the company plans to control the energy source powering its future data centers.
“Nothing is easy because all technologies are reaching their limits. But with a team like this [Nvidia and SK Hynix], nothing is impossible. At next month’s GTC, we’ll unveil chips that will astonish the world.” — Jensen Huang, Feb 18, 2026.
