More companies are limiting AI use after aggressively rolling out AI tools to employees, citing cost burdens.

Amazon, Walmart, Cisco, Uber and Meta are moving to control AI spending by setting usage limits, curbing wasteful use and encouraging cheaper models.

As employees begin using AI agents that autonomously carry out complex tasks beyond chatbots, much more computing power is needed. Companies are being forced to assess whether the work assigned to AI agents is worth the cost.

AI cost burdens are becoming reality, and Amazon, Walmart and Meta are imposing usage limits.

As cost pressures grow, interest is rising among companies in using cheaper AI models. One survey found Chinese AI models far outpacing U.S. models in token usage. Anthropic abruptly put on hold a planned change to pricing for the Claude Agent SDK just before it was to take effect. As a result, SDK users can continue using the same limits that applied to their existing Claude subscriptions. Databricks launched Unity AI Gateway, a tool designed to help companies control AI costs.

Chinese AI models have overtaken U.S. rivals in token usage, with DeepSeek V4 ranking No. 1. Anthropic abruptly halted changes to Claude Agent SDK billing, saying, "There will be no changes for the time being." Databricks unveiled Unity AI Gateway, a tool to control AI costs.

OpenAI and Anthropic are accelerating their push into the South Korean market. Anthropic opened an office in Seoul. It aims to compete in enterprise sales by promoting trustworthy AI.

Anthropic's Seoul office faces an early test, saying export controls will be resolved within days.

OpenAI and Anthropic are also moving to expand their foothold in the startup ecosystem. The two companies said South Korean startups are creating many innovative business cases using their AI models, and feedback from those startups plays a meaningful role in improving the models. OpenAI is also expanding its ChatGPT advertising pilot to South Korea.

OpenAI and Anthropic are focusing on innovation among K-startups, saying they will expand support. OpenAI is expanding its ChatGPT advertising pilot to South Korea.

OpenAI signed a memorandum of understanding with the Ministry of Science and ICT's AI Safety Institute to secure the safety of advanced AI models in high-risk fields.

The AI Safety Institute and OpenAI signed an MOU to secure the safety of high-risk AI.

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

Amazon Web Services announced three new services billed as a context intelligence stack for AI agents. The core is AWS Context, which automatically builds a knowledge graph from existing data and learns agent usage patterns to improve itself. AWS also added release management features to its DevOps agent. It reflects a situation in which review, testing and deployment have emerged as new bottlenecks as AI-powered coding spreads.

AWS launched a knowledge graph service for AI agents, joining the context intelligence race. AWS upgraded its DevOps agent with automated AI code testing and review features.

Data management company Cohesity launched Cohesity Maestro, which lets users manage data protection solutions directly from Claude, Gemini and ChatGPT.

Cohesity launched Maestro to manage data protection from Claude, Gemini and ChatGPT.

Salesforce agreed to acquire AI customer agent startup Fin for $3.6 billion. Salesforce plans to accelerate enterprise adoption of AI and strengthen its agentic AI platform, Agentforce, through the deal. DeepL is acquiring the team and technology of Mixhalo, a San Francisco-based real-time, ultra-low-latency audio platform.

Salesforce to acquire AI customer service startup Fin for $3.6 billion. DeepL acquires Mixhalo to expand real-time translation to large-scale events.

Adobe strengthened its Firefly AI assistant and added it to Premiere, Illustrator, InDesign and Frame.io. Disney plans to launch a beta version of an AI-based TV ad production tool in July.

Adobe enhances its Firefly AI assistant and expands it to Premiere and Illustrator. Disney to launch a beta AI ad production tool in July, targeting small and mid-sized advertisers.

Upstage unveiled a full-stack AI blueprint linking models, agents and a portal. The core is vertical integration by bringing AXZ, the operator of the portal Daum, and the general-purpose agent platform Timely under the Upstage umbrella.

Upstage's Sung-hoon Kim (김성훈) said blocking foreigners in the Fabled 5 increases the need for sovereign AI in South Korea.

AI and data company UCLIX opened the Google Gemini Enterprise Experience Center to support enterprise adoption of generative AI. Douzone Bizon, following its tax law assistant, launched ONE AI Labor Assistant, an AI agent that provides customised answers to labour issues based on laws, precedents and company-specific internal rules. LG AI Research signed a main contract with D&D Pharmatech for a joint project to develop next-generation peptide drugs. LG AI Research will develop an AI model that analyses the structure of disease-causing substances to design suitable peptide sequences that are difficult to discover with existing methods and to identify new drug candidates. IT infrastructure and AI solutions company Daol TS is teaming up with global real-time AI orchestration platform company Vantiq to target the real-time event-based AI market. Hancom signed an MOU with Poland's nationally certified R&D centre 7bulls.com to jointly conduct R&D to localise a next-generation sovereign agentic OS in Europe.

UCLIX opened the Google Gemini Enterprise Experience Center. Douzone Bizon launched ONE AI Labor Assistant after its tax law helper. LG AI Research and D&D Pharmatech move to full-scale AI-based peptide drug development. Daol TS partners with Vantiq, a real-time AI orchestration platform company. Hancom steps up its push into Europe's agentic OS market through cooperation with Polish R&D firm 7bulls.

Attention is focused on how much Google and Amazon can shake Nvidia's dominance as they accelerate sales of AI chips to external companies beyond use in their own clouds. Google is offering funding support to buyers to expand the ecosystem for its TPU (Tensor Processing Unit) AI chips. Amazon, the world's largest cloud provider, is also in talks to sell its Trainium AI chips to external data centre operators.

Tech Inside: Selling like Nvidia, with big cloud providers entering a new phase of the AI chip war. AWS is in talks to sell its in-house Trainium AI chips externally.

Competition is also intensifying among leading enterprise tech companies vying for the title of the first AI agent that corporate users use at work.

Which AI knows the company best? The race for AI agent platforms is heating up.

As AI model developers rapidly expand beyond models into AI applications, scenes are increasingly common in which relationships with former partner companies become delicate. Anthropic's recent moves have become a sensitive issue.

Anthropic and OpenAI expand at breakneck speed, unsettling the AI app ecosystem.