Nvidia Unveils AI Supercomputer While AMD Launches Helios Rack

The tech world saw significant advancements at CES 2026, particularly in AI hardware. Nvidia unveiled its Vera Rubin NVL72 AI supercomputer, slated for late 2026, promising five times faster inference performance and ten times lower cost per token compared to Blackwell. This powerful system will deliver 3.6 exaFLOPS of inference and 2.5 exaFLOPS of training performance, integrating Vera CPUs, Rubin GPUs, and NVLink 6. Concurrently, AMD introduced its "Helios" AI rack, featuring Instinct MI455X GPUs and EPYC 'Veince' CPUs, offering 2.9 exaflops of AI compute and 31 TB of HBM4 memory to meet the growing demand for generative AI. Nvidia further detailed its AI infrastructure with the BlueField 4 data processor, announced on January 5, 2026. This technology powers the Inference Context Memory Storage Platform, an AI-native storage system designed to enhance AI agents' long-term memory and facilitate fast data sharing, boosting performance and power efficiency by up to five times. Additionally, Nvidia's Spectrum-X Ethernet Photonics aims to build power-efficient AI factories, reducing power use by five times per port and offering ten times greater network reliability. Nvidia also boosted open-source AI tools on RTX AI PCs, making ComfyUI up to three times faster and llama.cpp and Ollama generate tokens up to 35% faster. AMD CEO Lisa Su emphasized the escalating need for compute power, predicting AI will require over 10 yottaflops in the next five years, a 10,000-fold increase from 2022 levels. AMD also launched its Ryzen AI 400 series chips for PCs, designed to enhance content creation and multitasking in upcoming laptops. Beyond hardware, AI is making strides in critical applications; researchers from MIT and Microsoft are leveraging AI, specifically the CleaveNet model, to develop nanoparticle sensors for early cancer detection through simple urine tests. The broader impact of AI continues to expand, with a 2025 Wharton-GBK study revealing that 82 percent of businesses use generative AI weekly, and 46 percent daily, with three out of four leaders reporting positive returns. DeepSeek announced a new method to train large language models more affordably by improving training stability, challenging the notion that better AI always requires higher costs. Meanwhile, Himachal Pradesh University is introducing new undergraduate courses in data science and artificial intelligence for the 2026-27 academic session, reflecting the growing demand for AI skills. The fitness company Equinox even used "strange" AI-generated art in a New Year's ad campaign, sparking conversations about technology's role in marketing. Amidst these advancements, a study suggests "cognitive grit," or thinking endurance, is becoming increasingly vital as AI provides easy answers, potentially reducing the need for prolonged human mental effort. This raises questions about how human cognition will evolve when endurance is no longer a primary requirement. Concurrently, Steven Jang of Kindred Ventures highlighted the symbiotic relationship between artificial intelligence and robotics, noting that progress in one field directly fuels growth and innovation in the other, creating a mutually beneficial cycle.

Key Takeaways

• Nvidia launched the Vera Rubin NVL72 AI supercomputer at CES 2026, expected late 2026, offering 3.6 exaFLOPS inference and 2.5 exaFLOPS training, with 5x faster inference and 10x lower cost per token than Blackwell.
• Nvidia's BlueField 4 data processor powers the Inference Context Memory Storage Platform, an AI-native storage system enhancing AI agent memory and boosting performance/power efficiency by up to five times.
• Nvidia introduced Spectrum-X Ethernet Photonics for power-efficient AI factories, reducing power use by five times per port and increasing network reliability tenfold on the Rubin platform.
• AMD unveiled its "Helios" AI rack at CES 2026, featuring Instinct MI455X GPUs and EPYC 'Veince' CPUs, providing 2.9 exaflops of AI compute and 31 TB of HBM4 memory.
• AMD CEO Lisa Su predicted AI will require over 10 yottaflops of computing power within the next five years, a 10,000-fold increase from 2022 levels.
• Nvidia significantly updated open-source AI tools on RTX AI PCs, making ComfyUI up to three times faster and llama.cpp and Ollama generate tokens up to 35% faster through NVFP4 and FP8 quantization.
• DeepSeek announced a new method for training large language models that improves performance without increasing costs, focusing on training process stability.
• Researchers from MIT and Microsoft are using AI (CleaveNet model) to develop nanoparticle sensors for early cancer detection, which could signal cancer-linked proteases in a urine test.
• A 2025 Wharton-GBK study found 82% of businesses use generative AI weekly, with 46% daily, and three out of four leaders reporting positive returns on AI investments.
• Himachal Pradesh University will introduce new undergraduate courses in data science and artificial intelligence starting the 2026-27 academic session.

Nvidia Unveils Vera Rubin AI Supercomputer at CES

Nvidia launched its new Vera Rubin NVL72 AI supercomputer at CES, expected in late 2026. This powerful system promises five times faster inference performance and ten times lower cost per token than Blackwell. It uses six types of chips, including the Vera CPU and Rubin GPU, along with NVLink 6 for high-speed connections. The Vera Rubin NVL72 rack offers 3.6 exaFLOPS of inference and 2.5 exaFLOPS of training performance. It also introduces a new Inference Context Memory Storage Platform using BlueField 4 DPUs to improve AI model responsiveness.

NVIDIA BlueField 4 Boosts AI Storage and Performance

NVIDIA announced its BlueField 4 data processor at CES on January 5, 2026. This new technology powers the NVIDIA Inference Context Memory Storage Platform, an AI-native storage system for large-scale AI. It helps AI agents process long information and improves their memory. The platform extends AI agents' long-term memory and allows fast sharing of data across AI systems. This boosts AI performance and power efficiency by up to five times.

NVIDIA Spectrum X Ethernet Powers Efficient AI Factories

NVIDIA introduces Spectrum-X Ethernet Photonics to build power-efficient AI factories. This technology uses co-packaged optics for optimized Ethernet networking on the NVIDIA Rubin platform. It ensures stable data flow and improves AI model performance, especially for Mixture of Experts architectures. The Spectrum-X Ethernet Photonics switch reduces power use by five times per port and offers five times longer uptime. It also provides ten times greater network reliability for critical AI applications.

NVIDIA BlueField 4 Boosts AI Context Memory Storage

NVIDIA introduced its BlueField 4-powered Inference Context Memory Storage ICMS Platform on January 6, 2026. This new AI-native storage system helps agentic AI models handle huge amounts of information, known as Key-Value KV cache. It addresses challenges with existing memory systems by providing a new context tier in Rubin AI factories. The platform uses NVIDIA Spectrum-X Ethernet for fast data access. This leads to higher AI throughput, better power efficiency, and improved reuse of KV cache.

AMD Unveils Helios AI Rack and Ryzen AI 400 Series at CES

At CES 2026, AMD CEO Dr. Lisa Su revealed "Helios," a powerful new AI rack. Helios is designed to support the growing demand for generative AI content. It features AMD Instinct MI455X GPUs, AMD EPYC 'Veince' CPUs, and other advanced hardware, offering 2.9 exaflops of AI compute and 31 TB of HBM4 memory. AMD also announced the Ryzen AI 400 series chips for PCs, which are faster for content creation and multitasking. These new chips will soon be available in laptops from major brands.

AMD CEO Lisa Su Predicts Massive AI Compute Needs

At CES 2026, AMD CEO Lisa Su stated that AI will need over 10 yottaflops of computing power in the next five years. A yottaflop is a one followed by 24 zeros, meaning this is 10,000 times more compute than in 2022. She noted that global AI compute grew from one zettaflop in 2022 to over 100 zettaflops by 2025. Su also used the event to introduce AMD's new MI455 GPU, designed for data centers.

Equinox Uses Strange AI Art in New Year Campaign

The fitness company Equinox launched a New Year's ad campaign called "Question Everything But Yourself." This campaign uses unusual, AI-generated images that some call "slop." The ads feature surreal visuals, like a melting face, mixed with real people exercising. Equinox chose AI art to be thought-provoking and to challenge traditional advertising. The company aims to encourage self-reflection and inner strength, sparking conversations about technology's role in marketing.

DeepSeek Finds Cheaper Way to Train AI Models

DeepSeek announced a new method for training large language models that can improve performance without increasing training costs. This challenges the common idea that better AI always needs more expensive training. DeepSeek's research focuses on making the training process more stable, which reduces wasted computing power. If proven reliable, this approach could make AI development more affordable for businesses. It would also allow for easier creation of specialized AI models.

NVIDIA Boosts Open Source AI Tools on RTX PCs

At CES 2026, NVIDIA announced major updates to speed up open-source AI tools on its RTX AI PCs and DGX Spark systems. These upgrades improve performance for small language models and diffusion models like FLUX.2 and Nemotron 3 Nano. Tools such as ComfyUI now see up to three times faster performance, while llama.cpp and Ollama generate tokens up to 35% faster. NVIDIA achieved this through new features like NVFP4 and FP8 quantization, along with better memory management. They also released the LTX-2 audio-video model and upgraded SDKs for advanced media effects.

AI Creates New Sensors for Early Cancer Detection

Researchers from MIT and Microsoft are using AI to create new sensors for early cancer detection. These sensors are nanoparticles coated with special peptides designed by an AI model called CleaveNet. The peptides react to proteases, enzymes that are very active in cancer cells. If cancer-linked proteases are present, the sensors give off a signal that can be detected in a simple urine test, potentially even at home. This method could help doctors diagnose many types of cancer much earlier, improving treatment outcomes.

Businesses Embrace AI for Daily Work and Growth

A 2025 study from Wharton-GBK shows that businesses are widely adopting generative AI for daily tasks. The report surveyed 800 decision-makers, finding that 82 percent use AI weekly and 46 percent use it daily. Three out of four leaders reported positive returns on their AI investments. AI is now seen as a crucial tool, not just a novelty, with companies tying its use to business goals. Studies show AI improves efficiency in areas like analysis, summarization, and coding, leading to measurable gains for organizations.

HPU Adds Data Science and AI Courses for Students

Himachal Pradesh University HPU's Executive Council approved new undergraduate courses in data science and artificial intelligence. These courses will be available in colleges affiliated with HPU starting in the 2026-27 academic session. The university also approved a new academic counsellor position for its Centre for Distance and Online Education. Additionally, HPU will introduce part-time PhD programs and made appointments for assistant professor roles in physics and electronics and communication engineering.

Cognitive Grit Becomes Crucial Amidst Easy AI Answers

A recent study suggests that "cognitive grit," or thinking endurance, is becoming more important as AI provides easy answers. Historically, deep thinking developed through sustained engagement with difficult problems. However, large language models now quickly generate and organize information, reducing the need for prolonged mental effort. This shift raises questions about how human cognition will develop when endurance is no longer a requirement. The article argues that cognitive grit, which builds mental durability, may need to be actively practiced rather than assumed in the age of AI.

AI and Robotics Industries Grow Together Says Expert

Steven Jang, founder and managing partner of Kindred Ventures, stated that the artificial intelligence and robotics industries are mutually beneficial. He explained on 'The Claman Countdown' that advancements in AI help improve robotics. In turn, robotics provides new platforms and data for AI development. This strong connection means both fields are growing and pushing each other forward.

Sources

• Nvidia launches Vera Rubin NVL72 AI supercomputer at CES — promises up to 5x greater inference performance and 10x lower cost per token than Blackwell, coming 2H 2026
• NVIDIA BlueField-4 Powers New Class of AI-Native Storage Infrastructure for the Next Frontier of AI
• Scaling Power-Efficient AI Factories with NVIDIA Spectrum-X Ethernet Photonics
• Introducing NVIDIA BlueField-4-Powered Inference Context Memory Storage Platform for the Next Frontier of AI
• CES 2026: AMD Just Showed Off 'Helios,' the Hardware That Will Power the AI Content in Your Feeds
• AMD CEO Lisa Su says AI will need 10 'yottaflops' of compute
• Why Equinox Leaned on AI Slop in Its New Year’s Ad Campaign
• DeepSeek Says AI Performance Gains Don’t Require Higher Training Costs
• Open-Source AI Tool Upgrades Speed Up LLM and Diffusion Models on NVIDIA RTX PCs
• AI-generated sensors open new paths for early cancer detection
• Experts agree: AI is here to stay
• HPU approves data science, AI UG courses in affiliated colleges
• Cognitive Grit in the Age of AI's Easy Answers
• The AI and robotics industries fuel each other: Kindred Ventures founder