The race to build large language models (LLMs) is heating up again. Prominent AI companies are rolling out new LLMs in quick succession in both the United States and China. Competition is also shortening the cycle for launching new models. Some observers say profitability at AI companies, already weak, could deteriorate further because model development requires substantial spending.

OpenAI unveiled its new AI model GPT-5.5 and began distributing it to paying ChatGPT users. The company says the model can reason faster and more precisely with fewer tokens than GPT-5.4, helping businesses and consumers use more frontier AI. Anthropic earlier released the latest model in its Opus series, Claude Opus 4.7.

ㆍOpenAI unveils GPT-5.5 ... a step closer to an AI 'super app' ㆍAnthropic releases latest model, Claude Opus 4.7

DeepSeek, a Chinese AI developer, also released preview versions of its new AI model V4. Some assessments say V4 could have a bigger impact than its earlier reasoning model R1, released last year, because it offers frontier-level performance as open source while being priced far below Opus 4.7 and GPT-5.5. Chinese AI startup Moonshot AI also unveiled the latest version of its open-source large language model Kimi series, Kimi-K2.6.

ㆍDeepSeek finally releases new V4 model ... competing on overwhelming value for money ㆍ"DeepSeek offers overwhelming value for money ... could have a bigger impact than R1" ㆍMoonshot AI releases open-source LLM Kimi-K2.6 ... "surpasses GPT-5.4 and Claude Opus 4.6"

The coding AI market is being reshaped increasingly quickly around companies backed by large amounts of capital. SpaceX, led by Elon Musk, has secured the right to acquire AI coding startup Cursor this year for $60 billion.

ㆍSpaceX secures option to acquire AI coding startup Cursor for $60 billion

Google DeepMind has formed a dedicated team to improve its AI coding models, according to reports. Anthropic is said to have had a major influence. Google DeepMind researchers are reported to have assessed that Anthropic's coding tools outperform Google's Gemini model.

ㆍGoogle founder: "Boost competitiveness of coding models" ... Google DeepMind forms dedicated coding AI team ㆍWhy some Google employees use Claude Code despite having Gemini

The article also summarises moves by companies in and outside South Korea around AI.

Google Cloud rolled out a wave of AI and cloud-related updates at its annual Cloud Next conference. Google also unveiled TPU 8t for AI training and TPU 8i for inference. It added a large number of AI features to its subscription productivity platform Workspace. The centerpiece is Workspace Intelligence. Workspace Intelligence supports automation across work using data from Gmail, Google Calendar, Google Chat and Google Drive.

ㆍGoogle unveils 2 next-generation TPUs for AI training and inference ㆍGoogle Workspace AI upgrade ... 9 times faster spreadsheet input and automatic document drafting

Thomas Kurian, CEO of Google Cloud, expressed confidence that a "full-stack" AI strategy — making chips, data centers, foundation models and products in-house — can narrow the gap with Amazon Web Services (AWS) in the cloud market.

ㆍGoogle Cloud CEO: "AI landscape will be reshaped in 1 to 2 years ... economics will determine survival"

Google Cloud is also strengthening cooperation with major South Korean IT service companies. At the 'Google Cloud Next 2026' conference, major South Korean IT service companies also made announcements about strengthening cooperation with Google Cloud.

ㆍGoogle Cloud expands alliance with major South Korean IT service companies ... major push on AI and security

Anthropic has attracted investment in succession from big tech companies. This month alone, it secured investments worth tens of billions of dollars from Amazon and Google, including conditional commitments. Based on that, Anthropic aims to speed up efforts to secure computing power to support demand for Claude.

ㆍAfter Amazon, Google also makes large investment in Anthropic ... up to $40 billion ㆍAmazon to invest up to an additional $25 billion in Anthropic ... to provide AI chips

As AI agents spread, the AI infrastructure landscape, once centered on GPUs, is shifting toward CPUs. Beyond inference, many tasks such as logic processing, file management, network calls and code execution are handled by CPUs, which is cited as a factor increasing the role of CPUs in AI agents. In this context, Meta drew attention by signing a deal to adopt Amazon Web Services' (AWS) in-house Graviton CPUs at scale, considering AI demand.

ㆍIn the era of AI agents, why CPUs are gaining weight ㆍMeta to adopt AWS Graviton CPUs at scale ... to handle AI agent workloads

Big tech companies are also accelerating moves to help companies manage the growing number of AI agents effectively. The companies are taking different approaches to AI agent management and governance.

ㆍHow to manage countless AI agents in a company? ... AWS and Google Cloud take different approaches ㆍAWS adds managed agent harness to AgentCore ... simplifying AI agent deployment ㆍ"This year is the first year of AI agents ... goals, data, guardrails and speed will decide real-world success"

Adobe unveiled CX Enterprise, an AI agent platform aimed at helping corporate customers automate digital marketing and other tasks. It plans to respond proactively to changes in the software market led by AI.

ㆍAdobe unveils CX Enterprise ... "Automating digital marketing with AI agents"

Oracle strengthened cooperation with Google Cloud and introduced an Oracle AI Database Agent that lets users query databases with natural language without SQL. Tencent released a global version of its OpenClaw-based AI agent QClaw. Data cloud company Snowflake updated Snowflake Intelligence and Cortex Code. Snowflake Intelligence is a context-aware AI agent for business users, allowing them to ask questions about corporate data and gain insights about the business.

ㆍOracle and Google Cloud release AI agent that queries Oracle DB in natural language ㆍTencent releases global version of OpenClaw-based AI agent QClaw ... links with WhatsApp and Telegram ㆍSnowflake upgrades AI agent platform ... expands integration with Gmail, Jira and Salesforce

Canadian AI company Cohere and German AI company Aleph Alpha have agreed to merge, according to reports. Aleph Alpha was once in the spotlight as a German version of OpenAI, but abandoned development of large AI language models like ChatGPT and shifted to enterprise-focused AI applications. Cohere is pursuing the same direction. Customer service AI agent startup Sierra acquired French startup Fragment. Fragment helps companies integrate AI into workflows.

ㆍIntercontinental AI company merger ... Canada's Cohere and Germany's Aleph Alpha to combine ㆍSierra acquires French AI startup Fragment

Agentic AI startup FortyTwoMaru will work with the Army Logistics Command on efforts for AI transformation (AX) in defense logistics. SK AX decided to integrate and manage overall IT operations for Daishin Securities over the next 7 years and gradually expand financial infrastructure operations through AI agents. Hancom plans to commercialise within this year a "twin-type agentic OS" that boosts productivity by replicating users' work patterns. LG AI Research and Nvidia executives decided to strengthen a technology alliance to expand the 'K-EXAONE' ecosystem. The companies will also broaden cooperation, including jointly developing specialized models by combining LG's AI model EXAONE and Nvidia's Nemotron open ecosystem.

ㆍFortyTwoMaru to cooperate with Army Logistics Command on 'defense AX' ㆍSK AX and Daishin Securities to deploy agentic AI in financial infrastructure operations ㆍHancom to commercialise 'twin-type agentic OS' this year ... "works 24 hours on behalf of users" ㆍLG AI Research to cooperate with Nvidia to combine EXAONE and Nemotron ecosystems

Motif Technologies, a developer of its own AI foundation model, is seeking differentiation by pushing its own platform instead of Nvidia software tools. It says using Nvidia's general-purpose tools would not allow it to beat big tech companies that have an edge in computing power. Motif is focusing on building and operating LLMs as cost-effectively as possible with its own tools. MegazoneCloud completed a proof-of-concept project with major South Korean architectural design firm Heerim Architects & Planners on an "AI agent system for reviewing building regulations". Enterprise software specialist TmaxSoft confirmed the brand name for its new AI product line under development as 'Continuum AI' and unveiled its brand identity.

ㆍMotif: "Developing our own foundation model with our own tools instead of Nvidia software ... a bet with higher odds against big tech" ㆍMegazoneCloud develops multi-agent-based AI for reviewing building regulations with Heerim Architects & Planners ㆍTmaxSoft to launch new AI brand 'Continuum AI' ... targeting the enterprise AI market

Google's expanded rollout in South Korea of its AI-enabled browser, 'Gemini in Chrome', has put Naver, the country's top portal, to a full test of its AI response.

ㆍGoogle targets the 'portal gateway' ... Naver's 'AI search' put to the test

As a scenario that AI will replace enterprise software, the so-called "SaaSpocalypse" narrative, gains traction, the counterattack from existing software companies is intensifying. Leading software companies are not only launching AI agent platforms but also actively rebutting the limitations of the SaaSpocalypse narrative.

ㆍ"There is no end to SaaS" ... software industry begins counterattack against AI-driven threat narrative ㆍ[Tech Insight] "Building enterprise software with LLMs would cost 10 times more than existing platforms"

As AI agents spread, usage-based pricing and performance-based pricing models — where fees are charged only when AI completes work properly — are also spreading rapidly in the enterprise software market.

ㆍ"Fit for the AI era" ... spread of performance-based software pricing