[DigitalToday reporter Chi-gyu Hwang (황치규)] Global AI companies are speeding up efforts to target the South Korean market for sovereign AI, or AI infrastructure that government agencies and companies can control.

Palantir, Cohere and Mistral are cited as leading global AI companies expanding investment in South Korea’s sovereign AI market.

Palantir recently held its second pop-up store event in South Korea and stressed its commitment to targeting the country’s sovereign AI market. Palantir cites three reasons the Korean market has important value: its focus on core industries such as telecommunications, semiconductors and shipbuilding; its experience over 50 to 60 years of the value of the South Korea-U.S. alliance and national defense; and a deep pool of technical talent. French AI startup Mistral is recruiting full-time applied scientists and research engineers to work in Seoul. Cohere also recently held a press briefing in South Korea. It also unveiled plans to set up local entities in South Korea and Japan and to double its Asia-Pacific technology and sales workforce by the end of this year.

• "Sovereign AI is about controlling the technology stack... in reality, many FDEs are consultants" • Mistral gains momentum in targeting enterprise AI in South Korea... begins full-scale hiring of technical staff • "Enterprise-optimized sovereign AI will decide the game... to release a model with around 1 trillion parameters within the year"

The article also sums up moves by companies in South Korea and overseas around AI.

OpenAI provides its AI models to developers on Google Cloud. This means OpenAI models can now be used across the world’s three biggest cloud services.

• OpenAI sells AI models on Google Cloud after Microsoft and AWS

Anthropic is reported to be preparing to release a new AI model that better understands user intent. The model is said to have stronger reasoning and intuition capabilities than Anthropic’s current models.

• Anthropic prepares to release new AI model... "It picks up what you mean even if you don’t spell it out"

Meta acquired an AI-based data analytics platform, Sundial. Sundial helps companies analyse data quickly. Meta launched its personal AI agent, Muse, in Japan. Meta first introduced Muse in Japan among regions outside North America.

• Meta acquires AI data analytics platform Sundial • Meta steps up global expansion beyond North America with Muse... lands in Japan first

Salesforce will acquire AI-based marketing platform company Braze for about $5.66 billion. Salesforce also introduced its specialist training programme for AI agents, "Partner FDE (forward-deployed engineer)", in South Korea for the first time.

• Salesforce to acquire AI marketing company Braze for $5.66 billion • Salesforce introduces "Partner FDE" programme to train AI agent specialists in South Korea for the first time

AI startup Recursive unveiled a free coding model, Flint, that runs on laptops. Recursive said it carried out most of the development of Flint using AI. AI startup Flapping Airplanes raised $1 billion in a Series A funding round. Flapping Airplanes is focusing on developing technology to train large models with less data.

• Recursive unveils free coding model Flint made with AI... runs on laptops • Flapping Airplanes: "Boost AI training efficiency with a megakernel that doesn’t leave GPUs idle"

Chinese AI company Moonshot AI unveiled a new flagship large language model, Kimi K3.1 Pro. The model outperformed Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra in some tasks. Chinese AI startup DeepSeek raised $15 billion in its second private funding round. DeepSeek is also bringing forward its listing schedule and is aiming to enter the Shanghai stock market early next year.

• Moonshot AI releases new LLM Kimi K3.1 Pro… highlights 3D modelling and financial analysis performance • DeepSeek raises $15 billion... targets Shanghai stock market listing early next year

Nvidia unveiled the "OpenShell Reference Architecture" to control enterprise AI agents. Nvidia made the entire Run:ai platform for managing AI workloads and GPU resources open source. AWS and Google unveiled tools that let AI apps built with the PyTorch development framework run on their own AI chips without changing the code.

• Nvidia unveils OpenShell Reference Architecture to control AI agents • Nvidia to fully open source Run:ai GPU management platform • Amazon open sources TorchNeuron... "Run PyTorch code as-is on Trainium" • Google open sources TorchTPU… supports migrating PyTorch apps to TPUs

Industry moves to provide cloud infrastructure optimised for AI agents are also accelerating. Google Cloud unveiled Agent Compute Engine, an infrastructure service optimised for running AI agents. Dell Technologies unveiled new desk-side agentic AI solutions and purpose-built AI infrastructure solutions as part of strengthening the Nvidia-based Dell AI Factory.

• Google Cloud unveils agent-only infrastructure, Agent Compute Engine • Dell expands desktop lineup targeting agentic AI... also offers modular data centres

Virtual twin company Dassault Systemes will cooperate with digital infrastructure company Vertiv to accelerate the design and construction of AI factories. Through the partnership, the companies will support configuring, simulating and validating AI factory infrastructure in a virtual environment before building it in the real world.

• Dassault Systemes-Vertiv alliance... verify AI factory infrastructure in a virtual setting before construction

Duzon Bizon signed a memorandum of understanding with law firm Shin & Kim on cooperation in AI governance and compliance. Through the cooperation, the two sides will set up a framework to jointly respond to demand for legal advice related to AI governance and compliance, building security systems and managing legal risks required as companies pursue AX, or AI transformation. All-in-one AI marketing agent VCAT.AI participated as an AI production company for TV Asahi’s historical drama series Sanada Juyushi. MegazoneCloud obtained Amazon Web Services status as an Authorized Government Reseller and will supply AWS GovCloud (US) to U.S. public sector and regulated industry customers. HR data-based AI platform flex signed an MOU with Microsoft Korea to support AI transformation at domestic companies. Channel Corporation introduced an AI phone, Voice ALF, that supports voice AI consultations, and provides ALF marketing, which calls customers first.

• Duzon Bizon joins forces with law firm Shin & Kim to target AI governance and compliance market • VCAT participates in AI production for TV Asahi, Netflix-distributed historical series in Japan • MegazoneCloud steps up entry into U.S. public cloud market • flex cooperates with Microsoft Korea on corporate AX business... also to push Copilot expansion • ChannelTalk launches Voice ALF for AI phone consultations... also supports outbound calls that contact customers first

Moves in the industry appear to be accelerating as AI models that operate differently from large language models gain weight in the tech arena. More AI companies are choosing to compete with non-LLM models, including tabular foundation models, focused on predicting structured data; models that do not generate text but only make judgments and decisions; and code execution models that operate on their own inside running programs without human intervention and find ways to run code more efficiently.

• From TFM to CEM... AI models that differ from LLMs are coming