[DigitalToday reporter Chi-gyu Hwang] Meta has unveiled Muse Spark, an AI model it developed with ambition. It is the first AI model released since it hired Alexandr Wang (알렉산드르 왕), formerly chief executive of Scale AI, nine months ago. Attention is focused on whether it can become a variable that reshapes the AI model landscape led by OpenAI, Google and Anthropic, as well as Chinese companies. For now, the start does not look bad.

ㆍMeta unveils new AI model 'Muse Spark'...steps up external API business ㆍMeta, Muse Spark whirlwind...key is proving revenue beyond interest ㆍMeta AI 'Muse Spark' effect...jumps from 57th to 5th on U.S. App Store ㆍTrying Meta AI 'Muse Spark'...feels like embedding ChatGPT in Facebook and Instagram

A push for AI-optimized IT infrastructure, so-called AI-native infrastructure, is also sweeping through the networking industry. Cisco is stressing shifts in computing and networking paradigms driven by the rise of AI and quantum computing, and is betting on an "Internet of Cognition." AI networking startup Aria Networks raised $125 million in a Series A round. Aria Networks is focused on helping AI data centers operate more efficiently and at lower cost.

ㆍ[DT People] "The era of probabilistic computing driven by AI is coming, and a big bang in internet infrastructure is ahead" ㆍAria Networks raises $125 million..."Providing the most efficient network for AI"

Industry moves around foundation models aimed at robots are also accelerating. Robot intelligence startup Generalist AI Inc. unveiled GEN-1, a robot-learning foundation model. ShengShu, the developer of AI video generation tool Vidu, closed a $290 million Series B funding round. ShengShu will use the funds to develop a 'general-purpose world model' linking a digital world based on games and AI video with a physical world based on autonomous driving and robots. Robot AI software company Skild AI also raised $1.4 billion. It plans to upgrade its AI model training infrastructure and speed up commercial scaling through the funding.

ㆍGeneralistAI unveils robot foundation model GEN-1 ㆍChina's ShengShu raises $290 million from Alibaba and others...steps up world model development ㆍRobot AI startup SkildAI raises $1.4 billion...Nvidia and Samsung Electronics also join

The article also rounds up moves by companies in South Korea and abroad around AI.

OpenAI launched a $100-per-month Pro plan that significantly increases usage of its AI coding assistant Codex. The move takes into account competition with Anthropic's Claude Code. OpenAI also unveiled two small models designed to handle delegated tasks for AI agent systems. They are GPT-5.4 mini and nano. The two models focus on codebase search, file review and parallel subtasks that need to be handled quickly and cheaply.

ㆍOpenAI adds $100-a-month Pro plan for heavy Codex users...targets Claude Code ㆍOpenAI launches GPT-5.4 mini and nano...opens an era of sub-agents

Anthropic officially launched Claude Cowork, an AI agent tool for non-developers. Claude Cowork is available across all paid Claude plans, including Pro, Team and Enterprise.

Anthropic also launched an open beta for Claude Managed Agents, which lets companies quickly build and deploy cloud-based agents, expanding its territory into agent infrastructure. Anthropic is also considering the possibility of designing its own AI chips. Signs have also emerged that Anthropic is trying to expand into HR. Anthropic hired Peter Bailis (피터 바일리스), formerly chief technology officer at cloud-based HR software company Workday.

ㆍAnthropic officially launches Claude Cowork...adds role-based access control and MCP permission management ㆍAnthropic expands into agent infrastructure..."Cuts time to one-tenth" ㆍAnthropic considers developing its own chips as Claude grows rapidly ㆍIs Anthropic expanding into HR software?...hires former Workday CTO

LG AI Research unveiled EXAONE 4.5, a multimodal AI model that understands and reasons over text and images at the same time. It is a preparation stage to expand modalities for K-EXAONE, which is being developed under its proprietary AI foundation model project. Douzone Bizon officially launched WEHAGO T AI Edition, which includes Proactive AI that prepares and suggests needed tasks in advance.

ㆍLG unveils multimodal AI EXAONE 4.5...focuses on strengthening reasoning capabilities ㆍDouzone Bizon launches WEHAGO T AI Edition with AI agents

Alibaba Group developed HappyHorse 1.0, an AI video generation model. Since its release, HappyHorse 1.0 has ranked first on the text-to-video conversion leaderboard of third-party site Artificial Analysis.

ㆍAlibaba unveils AI video generation model 'HappyHorse 1.0'...No. 1 on global leaderboard Amazon is considering selling its in-house AI chips to external companies as well. Cisco is acquiring AI observability startup Galileo Technologies. Galileo develops tools to observe and evaluate AI models. It provides a platform that monitors multi-agent systems in real time and applies guardrails to ensure accuracy.

ㆍAmazon considers selling its in-house AI chips beyond AWS to outside companies ㆍCisco to acquire AI observability company Galileo...expands security for AI agents

ServiceNow is overhauling its entire product lineup around AI agents. It aims to provide an 'AI-native architecture' that embeds agent automation across services, platforms and products. Visa unveiled Intelligent Commerce Connect, a platform that lets AI agents make payments on behalf of consumers. Design platform Canva acquired both AI collaboration and agent management platform Simtheory and customer data and marketing automation company Ortto. Atlassian launched 'Remix with Rovo,' a feature that instantly converts Confluence page content into visual materials such as charts and infographics.

ㆍServiceNow shifts to AI-native architecture...embeds agent automation in all products ㆍVisa unveils platform where AI agents pay on behalf of consumers ㆍCanva in another M&A...buys Simtheory and Ortto to strengthen AI agent and marketing automation capabilities ㆍAtlassian adds AI visualisation to document collaboration tool...also unveils MCP-based partner agent

Companies developing software with AI coding tools such as Cursor or Claude Code are facing a new dilemma. Review processes cannot keep up with the speed at which AI writes code, creating so-called 'code overload' or a 'deluge of code.' Beyond the sharp rise in the volume of AI-written code, another subtle management issue is that AI coding tools work better on laptops than in web-based environments on security servers run by companies such as Amazon and Microsoft.

ㆍA growing 'deluge of code' dilemma as AI coding spreads

OpenAI and Anthropic, which represent global AI startups, are preparing for blockbuster initial public offerings with this year as the target, but criticism persists that their internal structures are fragile. Surging costs required to train new AI models are also cited as an Achilles heel.

ㆍ[Tech Inside] Still a cash-guzzler...AI economics ahead of OpenAI and Anthropic IPOs