Major companies are increasingly entering the Neocloud market, which provides AI-focused cloud computing services, rapidly changing the industry landscape.

With Meta following SoftBank into the neocloud market, a space formed around startups is being reshaped into one that also includes big tech companies.

Attention is focused on whether this could shake up, to some extent, the cloud landscape led by Amazon Web Services (AWS), Microsoft and Google Cloud.

SoftBank Group and its telecom subsidiary SoftBank will start a neocloud business from the next fiscal year, renting AI computing resources to U.S. companies. Meta is also stepping up efforts to run a cloud business using data centres built with large-scale capital investment. Meta's move is similar to SpaceX. SpaceX signed a contract with Anthropic in early May and sold the full computing capacity of its Colossus1 data centre. It later signed similar leasing deals with Google and Reflection AI.

SoftBank declares entry into the U.S. AI cloud computing market; establishes SB Neo. Meta pushes ahead with cloud infrastructure business; also considers selling AI models.

Nvidia is also accelerating support for neocloud companies. Nvidia is providing financial guarantees to smaller cloud companies that make a living by leasing its GPUs, in exchange for sharing part of their revenue.

Nvidia provides funding guarantees to AI cloud companies and takes part of their revenue.

Moves by leading companies to build forward deployed engineer (FDE) organisations, as part of efforts to target the enterprise AI market, are gaining pace. Global big cloud companies AWS and Microsoft are also investing large sums to build FDE organisations.

Both the AI 'Big 2' and the cloud 'Big 3' build their own FDE organisations; major enterprise AX push. An AWS-style FDE organisation emerges; $1 billion to be投入.

Other moves by companies in and outside South Korea related to AI are summarised below.

OpenAI has limitedly released three new GPT-5.6 models: Sol, Terra and Luna. The top-end model Sol is specialised in cyber security. OpenAI will also launch a hardware device related to its AI coding tool Codex on July 15.

OpenAI limitedly releases three GPT-5.6 models; top model specialised in cyber security. OpenAI signals July 15 launch of a macropad dedicated to coding tool Codex.

Anthropic has begun early-stage work to develop its own AI chip. It is reported to be in talks with Samsung Electronics over a manufacturing partnership. Anthropic also released its mid-range model Claude Sonnet 5. Sonnet 5 offers performance close to Opus 4.8 while being cheaper to use.

Anthropic also reviews developing its own AI chip; discusses partnership with Samsung Electronics. Anthropic unveils Claude Sonnet 5; lowers price with Opus 4.8-level performance.

Google unveiled its image-generation model Nano Banana 2 Lite. Nano Banana Lite features improved speed and lower costs than the previous version.

Google unveils fast and low-cost image generation model Nano Banana 2 Lite.

Microsoft will integrate its consumer and enterprise Copilot into one in August. The unified app will include an AI coding tool and a new agent, Autopilot, that carries a separate fee. Salesforce unveiled Agentforce Help Agent, designed to reduce the burden of building and operating AI agents for customer service. Agentforce Help Agent applies a pay-per-resolution model.

Microsoft to integrate personal and enterprise Copilot in August; also streamlines unused features. Salesforce unveils Agentforce Help Agent; applies performance-based pricing.

Meta released a new app, Pocket, that makes small games and interactive apps using AI prompts. Users can create small interactive apps and games with text prompts and can run results made by other users directly from a feed. Meta also developed Vistara, a technology that installs DDR4 memory removed from older servers into new servers and shares memory across applications using its own CXL (Compute Express Link) chip without latency issues.

Meta quietly releases Pocket app that makes games with AI. Meta develops Vistara chip technology that reuses memory from older servers.

Adobe will introduce a joint global initiative with professional networking platform LinkedIn called AI Essentials for Marketers.

Adobe announces global AI skills initiative for marketing professionals with LinkedIn.

Chinese AI company Z.ai officially launched Zcode, a free desktop coding tool dedicated to its flagship GLM-5.2 model. Zcode competes directly with Cursor, Claude Code, GitHub Copilot and Google Antigravity. Meituan introduced the AI model LongCat-2.0, previously known by the code name Owl Alpha. Benchmark results show it scored 59.5 on SWE-Bench Pro, ahead of GPT-5.5's 58.6. It scored 70.8 on Terminal-Bench 2.1 and 77.3 on SWE-Bench Multilingual. DeepSeek unveiled DSpark, an open-source technology that speeds up large language model response times.

Z.ai launches Zcode coding agent for GLM-5.2; faces off with Cursor and Claude Code. Meituan releases LongCat-2.0 as open source; trained with 50,000 China-made ASICs. DeepSeek unveils DSpark, open-source technology that boosts AI inference speed by up to 85 percent.

Kuaishou Technology's AI video subsidiary Kling raised about $2 billion from investors. Kling is competing in the video AI model market with Google, New York-based Runway AI and ByteDance, releasing a series of AI models used for making films, advertisements and social media content.

Kuaishou AI video subsidiary Kling raises $2 billion; valued at $18 billion.

Data and AI specialist Mobigen held its '2026 Mobigen Media Day' on July 2 and unveiled Graphio 2.0, a dynamic ontology-based data and AI app platform, while emphasising a leap as a data and AI specialist. AI infrastructure specialist Dinoticia disclosed a paper and source code on KV cache compression technology, a temporary memory space that stores on GPUs so large language models do not recalculate previously read context. Rebellions will acquire SqueezeBits, a company specialising in AI inference optimisation. Rebellions is speeding up a shift into an integrated AI infrastructure company that provides software optimisation and inference serving on one platform, in addition to NPU hardware, through the acquisition.

Mobigen says dynamic ontology is the answer to solving AI slop; launches Graphio 2.0. Dinoticia discloses KV cache compression technology; says it resolves AI computing bottlenecks. Rebellions to acquire SqueezeBits; expands into AI inference optimisation.

Physical AI company CarbonSix, targeting manufacturers, raised $40 million in Series A funding from Korean and U.S. venture capital firms including DSC Investment, LB Investment, IMM, KDB, SV Investment, Cotensia and ASQ. French industrial automation solutions company Schneider Electric will acquire industrial AI company Cognite for $3.1 billion.

Physical AI company CarbonSix raises Series A funding of 60 billion won. Schneider Electric to acquire industrial AI company Cognite for $3.1 billion.

AI HR tech company Human Consulting Group (HCG) unveiled a year-end tax settlement AI chatbot based on generative AI technology for its payroll outsourcing service. Greeting, a recruitment management solution operated by Doodlin, launched an MCP integration feature that lets customers connect their recruitment data to AI tools to carry out hiring tasks.

HCG adds year-end tax settlement AI chatbot to payroll outsourcing service. Greeting launches MCP integration feature to support connecting recruitment data to AI.

Anthropic will deploy Claude functions to Salesforce Slack, a major enterprise communications platform. Microsoft is also developing a feature that allows Claude to be used in Teams, which competes with Slack. Inside Salesforce, there is also a different view. Some worry that the Claude tag is like bringing an enemy into the home.

Enemy or ally...Anthropic's Claude enters major B2B SaaS services.

Demand is increasing among companies for Chinese open-source AI models that are relatively cheaper due to AI costs while delivering performance close to OpenAI and Anthropic.

Leading U.S. tech companies increasingly use Chinese open-source AI models.

Palantir CEO Alex Karp criticised Anthropic and OpenAI's token pricing models, pointing to surging AI costs. "The industry should not underestimate the speed at which China is developing AI models," Karp said. "Many companies are shifting toward building and training more efficient proprietary tools themselves rather than using general-purpose AI models," he said.

Palantir CEO says companies should turn to open-weight models instead of wasting tokens.

As industry moves accelerate toward developing AI models optimised for robots, two camps are vying for technological leadership. One is VLA (vision-language-action models) derived from large language models. The other is the world models camp, trained mainly on video to predict what will happen in the real environment when robots take action.

Who will lead robotics AI models?...World models vs VLA rivalry.