Daily AI-Investing Landscape Update
Nemotron 3 Ultra vs. Chinese AI Models: Benchmarks Reveal Growing Performance Gap as Nvidia Pushes Physical AI
Thursday, February 26, 2026 · 32 items · 4 min read · Updated 3:07 PM
By the Numbers
>7%
U.S. oil price increase
Financial€12 billion
EU investment commitment for South Africa
Critical Minerals$146M
Timmins expansion capex
Critical MineralsThe Day's Thesis
▶Signal of the Day: Nvidia's Nemotron 3 Ultra becomes the most capable open US AI model yet still trails Chinese competitors on benchmarks, while the company simultaneously launched its Cosmos 3 world model and RTX Spark chip targeting physical AI and Windows devices.
The 30-Second Read:
- Nvidia's Nemotron 3 Ultra leads US open models but China maintains benchmark superiority
- RTX Spark combines Blackwell GPU with Arm Grace CPU for 1,000 TOPS local AI processing
- EU commits €12 billion to South African critical mineral projects targeting supply chain diversification
- Oil prices surge 7% on Iran's threats to block Strait of Hormuz amid escalating tensions
Today's developments expose a three-front competitive tension: US-China AI model performance gaps, the race for local AI processing dominance, and supply chain resilience amid geopolitical instability.
AI & Research Frontier
Nvidia's Nemotron 3 Ultra claims the top US open model position but China's lead persists across key benchmarks. According to Artificial Analysis, Nemotron 3 Ultra represents the most capable open-weight model from a US company to date, yet Chinese models continue outperforming American counterparts in standardized evaluations. Meanwhile, Chinese startup MiniMax released its M3 model featuring a million-token context window and native multimodality, directly challenging proprietary model leaders. The performance gap reflects deeper infrastructure and talent allocation differences that will influence enterprise AI adoption patterns and geopolitical technology dependencies through 2027.
Technology & Infrastructure
Nvidia's RTX Spark chip targets Windows devices with 1,000 TOPS of FP4 performance through unified Blackwell-Grace architecture. The chip combines Blackwell GPU cores with Arm-based Grace CPU and up to 128GB of shared memory, targeting fall 2026 deployment across ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI devices. This represents Nvidia's direct challenge to Apple Silicon and Qualcomm in the Windows AI laptop market. Simultaneously, Nvidia launched Cosmos 3 world model and Alpamayo 2 Super driving model at GTC Taipei, positioning itself as the primary infrastructure provider for physical robotics and autonomous systems across manufacturing and logistics sectors.
Sources & Items
Alternative search engine DuckDuckGo launches 'no AI' web extensions for Chrome and Firefox users.
Read original →