Nvidia unveils Nemotron Nano as Google adds new games

Key Takeaways

• NVIDIA released Nemotron-Nano-3-30B-A3B-NVFP4, a 30B parameter AI model utilizing a 4-bit NVFP4 format for up to four times faster performance on Blackwell B200.
• NVIDIA developed Hybrid-EP to enhance the efficiency of training large Mixture-of-Experts (MoE) models, such as DeepSeek-V3, by optimizing communication.
• Google DeepMind expanded its AI testing by adding Poker and Werewolf to Kaggle Game Arena, with Gemini 3 Pro and Gemini 3 Flash leading the leaderboards.
• Taiwan surpassed China in the MSCI Emerging Markets Index, holding 21.06%, driven by strong AI-related investments and its role as an Asian chipmaking hub.
• Apple CEO Tim Cook reported chip supply limits and rising memory costs, with TrendForce projecting DRAM prices could jump over 90% and NAND over 30% in early 2026.
• Many machine learning projects fail to reach actual use due to issues like selecting the wrong problem, data quality, and a gap between the model and the final product.
• Gartner research indicates that only one in 50 AI investments brings major changes, and only one in five provides a measurable return on investment.
• A Fox News poll found that six out of ten registered voters believe AI is developing too quickly, and 63% lack confidence in federal government regulation.
• For business AI to be truly effective, it needs to be connected across the entire company and integrated into daily tasks, as stated by Jan Gilg from SAP SE.
• The AI boom is increasing component costs for companies like Apple, reducing their power over suppliers who are prioritizing high-profit AI infrastructure contracts.

NVIDIA AI boosts Nemotron-3-Nano with new NVFP4 technology

NVIDIA released Nemotron-Nano-3-30B-A3B-NVFP4, a new AI model. This model runs a 30B parameter reasoning model using a 4 bit NVFP4 format, keeping its accuracy close to the BF16 version. It uses a special hybrid Mamba2 Transformer Mixture of Experts design and a Quantization Aware Distillation (QAD) method. This allows it to deliver up to four times faster performance on Blackwell B200. The NVFP4 format is a 4 bit floating point format that helps reduce memory use and increase speed for NVIDIA GPUs.

NVIDIA improves AI training with Hybrid Expert Parallel

Training large AI models like DeepSeek-V3 using Mixture-of-Experts (MoE) faces challenges, especially with communication. NVIDIA developed Hybrid-EP, a new solution to make this process more efficient. MoE models require frequent communication between experts, which can take over 50% of training time without optimization. Hybrid-EP helps solve this by using advanced NVIDIA hardware and software, like TMA commands for NVLink and IBGDA for RDMA networks. This new communication library improves how tokens are sent to and from experts, making AI model training faster and more balanced.

AI Layoff Myth Page Not Found

This article about the "AI Layoff Myth" was not found. The page displayed a "404 Page Not Found" error. It suggests that the content may no longer be available.

Taiwan surpasses China in emerging markets stock index due to AI

Taiwan has surpassed China in the MSCI Emerging Markets Index for the first time since July 2007. As of January end, Taiwan holds 21.06% of the index, slightly more than China's 20.93%. This shift highlights the strong interest in AI-related investments and Taiwan's key role as an Asian chipmaking hub. Experts like Joshua Crabb from Robeco note the dominance of the AI theme. Taiwan's MSCI index expects 37% earnings growth in the next 12 months, compared to China's 15%. South Korea also moved past India to become the third-largest market, driven by rising memory chip prices and companies like Samsung Electronics Co. and SK Hynix Inc.

Why many machine learning projects do not succeed

Many machine learning projects fail to reach actual use, with some studies showing high failure rates. This happens due to five main problems: picking the wrong problem, issues with data quality, a gap between the model and the final product, differences between offline and online performance, and non-technical obstacles. To succeed, teams must define clear business goals, treat data carefully, and build evaluation tools early. It is also important to encourage teamwork and ship small, impactful wins while exploring bigger ideas. The ML project lifecycle is long and complex, requiring careful management to avoid wasted effort.

Nine trends shaping the future of work after 2026

CEOs have high hopes for AI to drive growth in 2026 and beyond. However, employees are facing a more realistic view of how AI is performing. Gartner research shows that only one out of 50 AI investments brings major changes. Furthermore, only one in five AI investments actually provides a measurable return on investment. This suggests a gap between expectations and current AI results in the workplace.

Fox News poll reveals voter concerns about fast AI growth

A recent Fox News poll shows that six out of ten registered voters believe artificial intelligence is developing too quickly in the United States. Only three out of ten think it is moving at the right speed. Most voters, 63%, also lack confidence in the federal government's ability to properly regulate AI, a view consistent since 2023. While 26% feel AI has helped them personally, 53% say it has not made much difference in their lives. The poll was conducted from January 23-26, 2026, surveying 1,005 registered voters.

Connected intelligence is key for future business AI success

Business AI has moved beyond testing, but many companies struggle to get real results from their investments. Jan Gilg from SAP SE states that AI's power is often stuck in separate systems. For AI to be truly effective, it needs to be connected across the entire company and built into daily tasks. When AI, data, and cloud platforms work together smoothly, organizations can unlock AI's full potential. It is important to improve existing workflows before adding AI, as AI only speeds up current processes. Companies that connect their data foundations will see the biggest benefits from AI, allowing human creativity to thrive.

Google DeepMind expands AI testing with new games

Google DeepMind is improving how it tests AI models by adding new games to Kaggle Game Arena. Oran Kelly announced that Poker and Werewolf are joining Chess to challenge AI in more complex, real-world situations. Chess tests strategic reasoning, while Werewolf, a social deduction game, evaluates an AI's ability to communicate, negotiate, and handle unclear information. Poker focuses on decision-making under uncertainty, calculated risks, and bluffing. Gemini 3 Pro and Gemini 3 Flash currently lead the leaderboards in all three games, showing how fast AI models are improving.

AI boom raises Apple's chip costs and supplier power

The growing demand for AI infrastructure is increasing component costs for Apple, reducing its power over suppliers. Apple CEO Tim Cook noted chip supply limits and rising memory costs on January 29. TrendForce reports that DRAM prices could jump over 90% and NAND prices over 30% in early 2026. This is because suppliers are focusing on high-profit AI infrastructure contracts. Analysts believe the memory costs for the upcoming iPhone 18 could be almost $60 higher than the iPhone 17. Apple, a huge buyer of NAND, now faces challenges as its traditional negotiation tactics are less effective against the strong AI market demand.

Sources

• NVIDIA AI Brings Nemotron-3-Nano-30B to NVFP4 with Quantization Aware Distillation (QAD) for Efficient Reasoning Inference
• Optimizing Communication for Mixture-of-Experts Training with Hybrid Expert Parallel
• The “AI Layoff” Myth
• AI Fever Propels Taiwan’s Rise Over China in EM Stocks Benchmark
• Why Most Machine Learning Projects Fail to Reach Production
• 9 Trends Shaping Work in 2026 and Beyond
• Fox News Poll: Too Fast, Too Unchecked? Voters sound off on rapid AI use & government regulation
• Why connected intelligence is the future of enterprise AI
• Advancing AI benchmarking with Game Arena
• AI infrastructure surge begins squeezing Apple’s component costs — company considering supplier other than TSMC for lower-end chips, report claims