Moves by related industries to go beyond large language models (LLMs) and develop physical AI that fuses AI with the real world are gathering pace. From Nvidia, which dominates the AI chip market, to AI-focused firms, companies are increasing the cash they pour into physical AI.

At CES 2026, the global IT exhibition being held in Las Vegas, a range of companies also rolled out technologies aimed at physical AI.

Nvidia unveiled its Nvidia Alpamayo product lineup to accelerate development of next-generation inference-based autonomous vehicles (AV) in the United States. This year, Mercedes-Benz plans to begin shipping vehicles equipped with Nvidia autonomous driving technology to compete with Tesla Autopilot.

RLWRLD is advancing training so its robot Allex and a five-finger robotic hand can handle objects stably, based on the open humanoid robot AI model “Nvidia Isaac GR00T N1.5” unveiled by Nvidia Research.

• Nvidia unveils ‘Alpamayo’ open AI model lineup, accelerating inference-based autonomous vehicle development • Nvidia ‘Rubin’ can train AI models using only a quarter of Blackwell • RLWRLD advances robot hand manipulation training with open robot AI ‘GR00T N1.5’

Cooperation on physical AI between major manufacturers that wield large influence in the real world, such as robotics and automotive companies, and AI firms also appears to be gaining traction. The Korea Association of AI & Software Industry (KOSA) signed a memorandum of understanding (MOU) with the Korea Iron & Steel Association on building an AI autonomous manufacturing ecosystem for the steel industry and fostering convergence talent. The two sides plan to set up a provisional “Steel AI Autonomous Manufacturing Diffusion Committee” and 추진 specific tasks in stages, including policy proposals, talent development and technology verification. U.S. robotics company Boston Dynamics has formed a strategic partnership with Google DeepMind to accelerate development of its next-generation humanoid robot Atlas. Construction equipment brand Caterpillar is working with Nvidia to introduce AI and automation technologies into construction equipment.

• KOSA-Korea Iron & Steel Association to cooperate on AI, forming Steel AI Autonomous Manufacturing Diffusion Committee • Boston Dynamics to work with Google DeepMind on next-generation humanoids • Caterpillar to work with Nvidia on AI-based construction equipment development

Automotive chipmaker Mobileye will acquire AI humanoid robot company Mentee Robotics in a $900 million deal. Through the acquisition, Mobileye will expand its business into physical AI and plans to develop AI systems with context understanding, intent inference, natural interaction with humans and economic scalability.

• Mobileye to acquire AI humanoid robot firm Mentee Robotics for $900 million

As the industry increasingly moves into an aggressive mode, signs are also emerging that applying AI to manufacturing is not as easy as expected. Nvidia’s Omniverse project, introduced to address issues at robot design and manufacturing sites, also appears not to be going as the company intends. According to foreign media reports, Nvidia has worked for 4 years to monetise the Omniverse platform but has yet to make much progress.

• “Nvidia halts manufacturing-focused digital twin platform Omniverse Cloud”... why

Moves by domestic and global tech companies seeking to develop or use AI directly also continued this week.

Samsung Electronics declared at CES 2026 that it will usher in an era of everyday companionship by embedding controlled AI in 400 million devices a year. Samsung Electronics will first apply AI across all products and services, including mobile devices, TVs and home appliances.

LG Electronics also emphasised “Affectionate Intelligence” at CES. At its exhibition booth, it introduced its home robot “LG CLOiD” and its AI home appliance-based “Zero Labor Home” vision. LG CLOiD takes milk out of a refrigerator and puts croissants into an oven based on a meal plan. It takes laundry out of a basket, puts it into a washing machine, and folds and organises towels after washing is complete.

• Samsung to apply AI across mobile, TV and home appliances, embedding AI in 400 million devices a year • Samsung, LG Electronics declare at CES an ‘invitation to AI in everyday life’

OpenAI has integrated its engineering, product and research teams in recent months to revamp its audio model and plans to launch an audio-centric personal device within 1 year. It has also been reported that OpenAI is moving to strengthen healthcare features as a key strategy to grow ChatGPT into a “personal super assistant.”

• OpenAI to make large-scale investments in audio AI, launch audio-centric AI device within 1 year • OpenAI to introduce ‘health hub’ feature in ChatGPT, linking with wearables for personalised health advice

Elon Musk-led AI startup xAI raised $20 billion in funding. That exceeds its initial target of $15 billion. Anthropic is seeking to raise $10 billion in new funding at a valuation of $350 billion.

• Elon Musk’s xAI raises $20 billion, with Nvidia and Cisco among participants • Anthropic nears additional $10 billion funding at $350 billion valuation

AI data cloud company Snowflake will offer Google Cloud’s large language model (LLM) “Gemini 3” as a default in “Snowflake Cortex AI.” Snowflake customers can develop, deploy and scale generative AI applications, including Gemini data agents, within Snowflake’s governance-based data environment without moving or replicating data across platforms. Amazon is expanding its AI-based digital assistant Alexa+ to the web. With the update, Alexa+ provides an interface that allows users to converse directly with AI like ChatGPT and Google Gemini. Google unveiled Gemini AI features for TVs.

• Snowflake to support Gemini 3 by default on Cortex AI app development platform • Amazon to expand AI assistant Alexa+ to the web, competing with ChatGPT and Gemini • Google Gemini moves into TVs, enabling voice-based content search

Upstage open-sourced its in-house large language model (LLM) “Solar Open 100B” (Solar Open). According to the company, Solar Open is a 102 billion-parameter model that is only 15 percent the size of “DeepSeek R1 (DeepSeek R1-0528-671B),” one of China’s leading global AI models, but outperformed it in key benchmark evaluations across 3 languages: Korean (110 percent), English (103 percent) and Japanese (106 percent). SK AX received government certification for its in-house “AI capability certification platform,” the first government-recognised competency certification system in South Korea’s generative AI field. Naver Cloud will begin rolling out administrative services using generative AI with Busan Metropolitan City in earnest from this year. The plan includes building a Busan-specific AI using HyperCLOVA X and adopting the Naver Works collaboration tool.

• Upstage unveils proprietary LLM ‘Solar Open’, says it surpasses DeepSeek R1 • SK AX ‘generative AI utilisation certificate’ becomes South Korea’s first government-recognised credential • Naver Cloud-Busan to apply generative AI administrative services across the board

In the security market, dealmaking led by major companies has heated up from the start of the year. Reports say Palo Alto Networks and Cisco are moving to acquire security startups. Palo Alto Networks is pursuing the acquisition of Israeli cybersecurity startup Koi Security. Cisco is reportedly reviewing the acquisition of cybersecurity asset management startup Axonius. The cybersecurity mergers and acquisitions (M&A) market recorded its largest-ever 규모 in 2025, and there is also a 전망 that 2026 will be even more aggressive.

• Major players drive early-year M&A fervour in global security market • Cisco to acquire security asset management startup Axonius for $2 billion? • “Cybersecurity M&A hit record high last year, to be even more active this year”