Meta has launched its paid artificial intelligence (AI) coding service, Muse Code, marking the company’s entry into the developer market.

On Aug. 5, Meta unveiled its latest Muse Spark 1.2 model alongside Muse Code, an AI coding agent. Muse Spark 1.2 is an updated version of the previous model optimized for software development tasks. According to Meta, the model was trained with significantly increased compute resources and a broader range of training environments, improving its capabilities in code generation, complex debugging and codebase comprehension.

While Muse Spark is a general-purpose model, Muse Code is a standalone terminal-based AI coding agent tool. Meta said it trained Muse Spark 1.2 and Muse Code together. Muse Code, the company’s first AI coding agent, is powered by the latest Muse Spark 1.2 model and has been released in beta for macOS and Linux.

One notable aspect of the announcement was the complete absence of references to open source. The shift is unusual for Meta, which has positioned itself as a leading advocate of open AI ecosystems in recent years.

Whereas the company previously pursued indirect monetization through the expansion of the open ecosystem around its open-weight Llama models, Muse Code represents an attempt to generate revenue directly from an in-house business.

Meta highlighted several technical features of Muse Code, including autonomous development workflows, persistent asynchronous background agents and local event logs.

An autonomous workflow feature allows the system to handle planning, coding, testing and result verification in a single process after receiving a user request—a standard capability among advanced AI coding tools. However, persistent asynchronous background agents represent Meta's key strategic move as a late entrant to the AI coding market.

Persistent background agents distribute workloads across multiple AI agents operating simultaneously in the background during a development session. While most competing AI coding tools also support background agents, those systems typically create new helper agents for each task.

In contrast, Muse Code keeps specialized background agents continuously active (warm) throughout an entire session. According to Meta, maintaining agents in a warm state rather than creating them for individual tasks helps avoid redundant information gathering.

This feature could significantly address practical inefficiencies in code development, as creating a new sub-agent for every task forces the system to repeatedly reread the same files.

Under this feature, large-scale tasks are distributed and processed in parallel by sub-agents operating within isolated Git worktrees. This architecture preserves the advantages of parallel execution while preventing state contamination, such as file deletion or modification by other agents on the same files.

Generally, a Git repository has only one working directory, but a Git worktree is a Git feature that enables multiple separate directories—each checking out different branches—to share a single repository (.git history) simultaneously. Each sub-agent operates within its own worktree (an independent directory with an independent branch), ensuring physical isolation from file changes made by other agents. Once work is complete, branches can be merged sequentially, confining conflict management strictly to the integration phase.

Although the technology itself is not new, many products still process subtasks sequentially within a single worktree or rely on file locking and execution ordering for parallel processing. Adopting worktree isolation as a core architecture increases implementation complexity and cost, requiring branch creation and management, merge strategies, conflict handling and cleanup logic.

For that reason, technology media outlet VentureBeat praised Meta’s decision to offer the architecture as a built-in capability at the harness level, despite its relatively limited adoption across the industry.

Meta also highlighted its local event log system, which records model calls, tool usage, approvals and code edit histories to enable task resumption even after crashes. The append-only event log allows developers to replay their entire work history from start to finish to resume work following a failure. This substantially resolves operational issues and serves as a foundation for audit trails.

Overwrite-style event logs make it impossible to determine prior actions, the process taken to reach the current state, or pending tasks. In contrast, foreign media outlets have praised Muse Code's append-only event log system for preserving the entire history sequentially for reconstruction at any time. This forms the foundation for developer efficiency, operational problem-solving, failure recovery and root-cause analysis (RCA) audit trails.

However, Muse Code does have weaknesses. First, its initial release is restricted to a terminal interface. Competing products such as Anthropic’s Claude Code offer terminal, IDE, desktop and web interfaces, while OpenAI’s Codex is supported across CLI tools, IDE extensions and cloud services.

Meta also released benchmark results comparing Muse Spark 1.2 and Muse Code against competing AI models, including Grok 4.5, Claude Opus 5, GPT-5.6 Terra and Gemini 3.6 Flash. The choice of comparison models sparked controversy, as Meta avoided top-tier models from competitors in favor of second-tier options.

According to Meta, Muse Spark 1.2 and Muse Code (harness) achieved a score of 82.9% on Terminal Bench 2.1, narrowly outperforming OpenAI’s GPT-5.6 Terra and Codex at 81.8% and SpaceX (xAI)’s Grok 4.5 and Grok Build at 81.6%. However, it trailed Anthropic’s Claude Opus 5 and Claude Code, which scored 86.7%.

On the Deep SWE 1.1 benchmark, Muse Spark 1.2 ranked behind Claude Opus and OpenAI GPT-5.6 Terra. This drew criticism from the developer community, with some arguing that the system would show lower performance when compared against GPT-5.6 Sol.

AI tech news site Kingy AI further claimed that Meta’s reported benchmark scores were slightly higher than actual verified figures, publishing a side-by-side comparison between Meta’s claims and verified evaluation scores.

The timing of the announcement has also raised suspicion. Muse Code was unveiled less than a month after the release of Muse Spark 1.1. Observers interpret this as an attempt by Meta to regain momentum after Moonshot AI’s Kimi K3 stole the spotlight with overwhelming performance immediately following the ambitious Muse Spark 1.1 launch.

Despite these concerns, Muse Code’s pricing structure is viewed as a major advantage that offsets doubts about its market positioning. Meta is attempting to compensate for its late entry into the coding tool market with an aggressive pricing policy.

Critics note that the pricing strategy is more disruptive than the technology itself. Under the Standard Tier, Muse Spark 1.2 API access for Muse Code costs $1.25 per million input tokens and $4.25 per million output tokens, with cached inputs priced at $0.15 per million tokens. Meta also pledged not to use prompts or generated outputs for model training.

A second pricing option, the Contributor Tier, is available at just $0.10 per million input tokens and $0.20 per million output tokens. In exchange, developers must allow Meta to use their code for future model training.

Financial pressure appears to be driving this aggressive pricing strategy. While Meta reported second-quarter revenue of $60.8 billion, up 28% year over year, operating income fell 8% to $18.8 billion, and operating margin declined from 43% a year earlier to 31%. Capital expenditures (CAPEX) reached $31.08 billion for the quarter alone, with full-year guidance reaching up to $145 billion.

Meanwhile, entering the developer market effectively means entering the enterprise market. Alexander Wang, chief AI officer (CAIO) of Meta Superintelligence Labs and head of Muse Code development, emphasized the product's enterprise capabilities, noting that requests for Zero Data Retention (ZDR) features have already begun arriving. Remarks from CAIO Alexander Wang suggest that ZDR is currently implemented in Muse Code as an optional capability available to enterprises upon request.

ZDR is an agreement where the AI service provider immediately deletes user inputs (prompts) and outputs (responses) from servers upon completion of API call processing, storing them in no form whatsoever. While the Standard Tier already includes a commitment not to use prompts and data for training, ZDR goes one step further.

Under standard tiers, logs are temporarily stored on servers for a set period before deletion, whereas ZDR prevents storage entirely from the outset, eliminating risks associated with data leaks, hacking or future legal disputes. This is essential for regulatory compliance and intellectual property (IP) protection, both of which are major hurdles for enterprise adoption of AI coding tools.

Organizations handling sensitive source code representing trade secrets or operating under strict legal requirements regarding data retention on third-party servers demand ZDR capabilities. Without ZDR, deployment in highly regulated industries remains difficult regardless of model performance, making the feature an essential requirement for AI models and coding tools targeting the enterprise market.

Canadian enterprise AI company Cohere held its first media briefing in South Korea on Aug. 5 at The Westin Josun Seoul, unveiling plans to make the country a key hub for its Asia-Pacific expansion. Frank O'Dowd, Chief Revenue and Commercial Officer, and Hugh Jang, President for Asia-Pacific, outlined the company's regional strategy.

O'Dowd said South Korea is the most dynamic and competitive AI market in the Asia-Pacific region, adding that success in Korea would provide a strong foundation for expansion into other regional markets. Based on that assessment, Cohere plans to establish a Korean subsidiary by the fourth quarter of this year and hire a double-digit number of technical and sales personnel. The company also plans to establish a separate legal entity in Japan and deepen cooperation with Fujitsu, while expanding support for public-sector customers in Singapore.

Cohere's business in South Korea has expanded through its partnership with LG CNS. The company described LG CNS as its first key partner in the market, helping localize Cohere's AI models and agentic AI platform for Korean language and industry requirements. Cohere said it plans to broaden its domestic ecosystem beyond LG CNS to include additional IT service providers and AI model companies.

Two themes stood out in Cohere's presentation. First, the company spent considerably more time discussing its AI agent platform, North, than its proprietary generative AI model family, Command. Second, it positioned sovereign AI as a core competitive advantage.

Cohere was originally known for developing the enterprise large language model (LLM) Command. Founded in 2019 with the goal of building generative AI for businesses, the company counts Aidan Gomez among its three co-founders. Gomez, formerly of Google Brain, was a co-author of the landmark 2017 paper "Attention Is All You Need," which introduced the Transformer architecture that underpins modern AI and LLMs.

From the outset, Cohere targeted enterprise customers. It launched API services for enterprise text generation and vector embeddings in 2021 and later developed Command in 2022. Until the launch of North in January last year, the company's primary focus was AI models. At its first Korean press event, however, the majority of the discussion centered on North, its AI workspace platform.

North is an enterprise AI agent platform that connects internal corporate data and business systems so AI agents can perform real-world work tasks. Combined with Cohere's Embed and Rerank engines, it enables retrieval-augmented generation (RAG)-based search across large volumes of enterprise data, including documents, emails, databases and wikis, allowing users to find relevant information and generate answers and reports. The platform also supports enterprise AI agents capable of autonomously planning and executing multi-step tasks and workflow automation.

When Cohere launched North early last year, it positioned the product as an enterprise alternative to Microsoft and Google. North competes directly with Microsoft Copilot Studio and Google Vertex AI Agent Builder.

All three platforms connect enterprise data such as documents, emails and presentation files with AI systems to create customized internal AI agents.

Microsoft Copilot Studio is tightly integrated with the Microsoft 365 ecosystem, making it highly accessible for enterprises running Windows-based environments. Google Vertex AI Agent Builder allows organizations to build agents integrated with Google Workspace on Google Cloud Platform infrastructure.

In contrast, Cohere said North is not tied to any specific hyperscale cloud ecosystem and can be deployed across any cloud platform or internal infrastructure chosen by the customer. The company argues that this flexibility makes North particularly suitable for organizations that seek independence from large technology vendors and require stronger security controls and customization, especially in financial services and other highly regulated industries.

According to Cohere, the company has adhered to four principles since its founding: enterprise control of data, multi-cloud support, on-premises deployment capability, and strong security and privacy protection. Command and North can be deployed in public clouds, private cloud environments, customer-owned data centers, or air-gapped environments disconnected from the internet. The company cited this deployment flexibility as its primary differentiator and the foundation of its sovereign AI strategy.

Cohere argued that enterprises and government agencies should be able to maintain direct control over both their data and AI operating environments. The company said its ability to support AI deployment in private cloud and on-premises environments, in addition to public cloud infrastructure, enables customers to meet data sovereignty and security requirements.

According to international media reports including TechCrunch, overseas customers using North have adopted the platform's agent capabilities in on-premises environments to balance strict security requirements with business productivity gains. Royal Bank of Canada (RBC), for example, worked with Cohere to develop and deploy North for Banking, a dedicated platform tailored to financial services. All financial data remains within RBC's own data centers and is not transmitted to external cloud environments.

LG CNS is also cited as one of North's key early customers. Cohere and LG CNS signed a partnership agreement in March 2025, and Cohere later stated on its corporate blog that LG CNS had deployed North to validate the effectiveness of AI agents in large-scale enterprise environments. According to the company, the platform is being used as a foundation for secure, customized AI agent initiatives at Korean enterprises that are highly sensitive to data leakage risks and seek alternatives to dependence on major technology vendors.

Cohere also said North provides transparent citations to source documents and visibility into how responses are generated, allowing enterprises to trace AI-generated outputs. The company said these capabilities have made the platform particularly attractive to financial institutions and large enterprises where controlling hallucinations is critical.

O'Dowd said many organizations have adopted generative AI but have not achieved the return on investment they initially expected. Rather than relying on the largest available AI models, he said, companies should select models suited to specific business objectives and optimize both performance and cost.

As part of that strategy, Cohere said North supports not only the company's own Command models but also a range of third-party AI models. The company also emphasized that it charges customers based on the number of annual users rather than token consumption, making AI operating costs more predictable for enterprises.

Cohere's decision to place North ahead of Command appears to reflect a recognition that competing solely on model performance is no longer sufficient. When Command was first introduced, model-to-model performance competition was still a meaningful differentiator. Today, however, the company lacks a model that directly rivals the performance of OpenAI's ChatGPT or Anthropic's Claude.

Instead, Cohere argues that enterprises should have broad freedom to choose the most appropriate models for their environments, including open-weight models, and describes that model choice as another component of sovereign AI. The message closely mirrors comments recently made by Palantir Chief Executive Officer Alex Karp.

In terms of market positioning, strategic direction and enterprise deployment scope, Cohere North sits between Palantir AIP and OpenAI's ChatGPT Enterprise and ChatGPT Work offerings. The distinctions among the three platforms, however, are becoming increasingly blurred.

OpenAI is expanding beyond AI assistants toward AI agent platforms and enterprise workflow automation, while Palantir is broadening AIP beyond operational automation into knowledge-worker functions such as document processing and natural-language interfaces. Cohere occupies a position that gives it opportunities to expand in both directions, but also leaves it exposed to competitive pressure from both sides.

Bigwave Robotics and Omron Electronics Korea have signed a memorandum of understanding (MOU) to cooperate in the robotics business, aiming to establish a framework for deploying safe humanoid robots in industrial environments.

Bigwave said that while humanoid robot pilot projects are expanding, existing robot safety standards do not adequately account for characteristics such as bipedal locomotion and AI-based autonomous decision-making. As a result, the company said there is a growing need to integrate risk assessment and safety design from the earliest stages of humanoid robot deployment.

Under the agreement, Bigwave's humanoid robot supply and system integration capabilities will be combined with Omron Electronics Korea's industrial safety consulting and safety control technologies to establish new safety standards for humanoid robots.

The two companies plan to jointly manage the entire process, from identifying customer demand and proposing robot specifications to risk assessment, safety hardware design and installation, validation, and documentation. Bigwave will propose humanoid robots and systems tailored to customer manufacturing processes and oversee on-site integration.

Omron Electronics Korea will support risk assessments and Performance Level required (PLr) analysis based on ISO 12100 and ISO 13849-1 standards, as well as the deployment of safety solutions including safety programmable logic controllers (PLCs), emergency-stop switches, and laser scanners. When necessary, the company will also integrate safety hardware such as light curtains, which stop equipment operation when a worker enters a monitored area, and door switches to improve applicability in industrial settings.

Bigwave said customers will be able to receive robot supply, safety consulting, safety hardware deployment, validation documentation, and on-site implementation as part of a single project. To accelerate humanoid robot adoption in manufacturing and logistics facilities, the company will establish a demonstration environment for safety solutions at its data center, where the partners will conduct joint proof-of-concept (PoC) projects and field trials.

The companies also plan to standardize processes from initial consultation through on-site deployment and secure real-world implementation references through joint seminars and facility assessments.

AI platform company INFINIQ has begun developing an intelligent data processing technology that converts handwritten military logistics documents into AI-based digital data.

The company announced that it has completed the project kickoff for the development of an OCR-based intelligent data processing and structuring system for unstructured handwritten documents in the military logistics sector under the second round of the 2026 SME Technology Development Support Program. Delivery is scheduled for December.

The project's core objective is to automatically recognize unstructured logistics documents—including supply requisition forms, maintenance records and inventory status reports—using AI-powered optical character recognition (OCR) and convert them into structured data that can be directly linked to the Defense Logistics Integrated Information System (DELIIS).

INFINIQ said most military logistics forms are currently handwritten and later re-entered manually by personnel, resulting in significant labor consumption and data-entry errors. The company added that conventional OCR solutions relying on external cloud APIs have limitations in accurately recognizing military-specific codes such as National Stock Numbers (NSNs) and unit abbreviations.

To address these challenges, INFINIQ plans to deploy a three-layer hybrid recognition engine combining precision OCR, visual language model (VLM)-based contextual understanding and ontology-based logical validation. OCR will recognize handwritten text and special codes, the VLM will correct misidentified content based on context, and the ontology layer will verify data consistency according to military logistics domain rules.

The company said the system is designed to achieve character recognition accuracy exceeding 95%, a character error rate (CER) below 5%, and processing times of less than five minutes per document. INFINIQ emphasized that the platform will operate independently within a fully isolated on-premises network without external internet connectivity, and will also incorporate a five-layer security framework that includes data encryption, role-based access control and automatic masking of personal information.

According to the company, the system will enable personnel currently assigned to manual data entry to focus on analytical and decision-making tasks while reducing data entry and verification time by more than 50%. The company added that the accumulated maintenance data could later be analyzed using time-series techniques to predict optimal maintenance schedules, which could reduce maintenance costs, prevent equipment failures and improve overall logistics operations.

To ensure technical completeness and domain relevance, the Korea Military Academy will participate as an advisory institution. The academy will provide development facilities and collaborate throughout the project on ontology design and validation.

INFINIQ also plans to integrate the structured logistics data generated by the project with its ontology-based decision support platform, AURON. Combined with AURON, the data can be searched and analyzed through semantic relationships linking maintenance histories and inventory status, and the platform can assess impact ranges and maintenance priorities based on relationships among equipment, units and supplies. The company said it ultimately plans to visualize the data through dashboards and expand the system into a military-specific AI platform that can be utilized across units, facilities and operational functions.

CJ ONE, the lifestyle membership platform operated by CJ OliveNetworks, is launching its August Wonder Series promotion in partnership with Mega MGC Coffee.

The company previously held the promotion with Tous Les Jours in July and will offer benefits through Mega MGC Coffee in August.

The August Wonder Series consists of Wonder Week with Mega MGC Coffee, Wonder Raffle on Aug. 11 and Wonder Quiz on Aug. 21.

Wonder Raffle will offer prize drawings, while Wonder Quiz will provide opportunities to earn CJ ONE points and benefits from major CJ brands.

From Aug. 1 to Aug. 5, CJ ONE is running a teaser event for Wonder Week. Members who sign up for event notifications through the CJ ONE app during the period will receive CJ ONE points.

The main Wonder Week event will run from Aug. 6 to Aug. 10.

During the event, all CJ ONE members can receive a 50% discount coupon for a Mega MGC Coffee iced Americano through the CJ ONE app at 10 a.m. each day.

Members who did not earn or redeem CJ ONE points at Mega MGC Coffee between Jan. 1 and July 31, 2026, will receive an additional 50% discount coupon if they make their first point transaction during the Wonder Week promotion.

Global AI audio research startup ElevenLabs has made its new dubbing model, Dubbing v2, available through its ElevenAPI platform, allowing developers to integrate the technology directly into their products and workflows. The model was first introduced through a user interface version in May and is now being offered via API access.

Unlike conventional dubbing systems that follow a speech recognition → translation → speech synthesis pipeline using cloned voices, Dubbing v2 employs a direct speech-to-speech (audio-to-audio) architecture. ElevenLabs said the model generates speech output directly from the source speech, enabling it to replicate the speaker's voice and more faithfully reflect the original's emotional expression and performance while producing natural multilingual dubbing.

The company also said the model is designed to align spoken dialogue naturally with video timing, improving synchronization accuracy. However, Dubbing v2 does not provide AI lip-sync processing for video content or real-time multilingual dubbing for live-streamed broadcasts. The model supports 92 languages, including Korean.

ElevenLabs said the primary target users for Dubbing v2 include content creators, media companies and global enterprises seeking to deliver high-quality content to international audiences.

The company added that professional film- and television-grade dubbing has traditionally cost hundreds of dollars per minute, whereas Dubbing v2 aims to provide high-quality dubbing at a more affordable price point.

MICUBE Solution, a provider of smart factory and autonomous manufacturing artificial intelligence (AI) solutions, will build an AI autonomous manufacturing platform for cosmetics original development manufacturer (ODM) Cosmecca Korea.

The project is being carried out under the Ministry of SMEs and Startups' "AI Applied Product Rapid Commercialization Support Program," operated by the AI Innovation Promotion Team of the Korea Technology and Information Promotion Agency for SMEs (TIPA). MICUBE is one of the selected solution providers for the initiative.

MICUBE will develop five core AI solutions to be deployed at Cosmecca Korea's manufacturing facilities. The goal is to create an intelligent autonomous manufacturing environment spanning the entire cosmetics production process, from research and development to formulation, production and quality inspection.

The solutions include: an AI Multi-Vision Quality Inspection system that uses deep-learning-based image analysis to respond immediately to product diversification and process changes; a Recipe Generation AI system that analyzes body and skin information alongside raw material and ingredient data to recommend optimal cosmetic formulations and validate them in a virtual environment; a Similar Complaint Prevention AI system that leverages large language models (LLMs) to analyze historical quality and customer complaint data, predict potential defects and provide natural-language question-and-answer capabilities; a Generative AI Data Integration platform that connects data across R&D, production and quality processes to support real-time decision-making; and a Digital Twin-Based Autonomous Control system that combines virtual manufacturing environments with LLMs to optimize processes in real time while providing AI-generated control recommendations and decision rationales.

Regarding the Recipe Generation AI solution, MICUBE said it has already validated a mathematical structure-based optimization algorithm in the secondary battery industry, where it analyzes large numbers of raw material combinations to identify optimal formulations. The company plans to apply that technology to cosmetics recipe development. It added that it has carried out numerous AI projects in industries such as food and chemicals, where ingredient blending and recipe optimization are critical.

The project will be implemented over two years. During the first year, the companies will establish AI core models and data infrastructure. In the second year, they plan to enhance the AI models, conduct field validation and commercialize a customized autonomous manufacturing system. The companies also intend to expand deployment of the solutions to Cosmecca Korea's global production sites.

MICUBE said it aims to successfully demonstrate a cosmetics-focused AI integration framework that combines generative AI and digital twins through the project. The company plans to expand the framework as a standard AI transformation (AX) model for small and mid-sized manufacturers and broaden the application of its autonomous manufacturing solutions to industries including pharmaceuticals and biotechnology.

Chinese AI startup DeepSeek and robotics manufacturer Unitree are partnering to develop AI models for robots.

According to Reuters and other media reports on Aug. 6, DeepSeek will invest 140.8 million yuan in Unitree and receive 933,399 shares, equivalent to a 2.31% stake, through a strategic allocation tied to Unitree's initial public offering (IPO) in Shanghai. The companies plan to use the partnership to develop robotics AI. DeepSeek will contribute AI model development capabilities, while Unitree will provide expertise in mechanical engineering, motion control and embodied intelligence.

Specific details about the robotics models under development have not been disclosed. However, the South China Morning Post reported that the two companies will jointly develop not only AI models but also embodied AI technologies, while granting each other preferential access. Unitree will give priority to DeepSeek when procuring AI models and model-training services, while DeepSeek will prioritize Unitree when purchasing humanoid robots, quadruped robots and robotics platforms.

Unitree's robotics data and hardware platforms are expected to become key assets for DeepSeek's robotics AI ambitions. For Unitree, access to advanced robotics AI models could strengthen its competitive position against AgiBot. More broadly, the partnership could help China combine its strengths in humanoid robot hardware, supply chains and manufacturing capacity with frontier AI models.

Reuters noted that while Chinese humanoid robots are relatively inexpensive and capable of complex movements, they still lack the autonomy needed to perform real-world tasks without human intervention. The combination of Unitree's physical-world data and DeepSeek's AI expertise could improve competitiveness in the humanoid robotics model market.

The "brains" of humanoid robots are typically powered by Vision-Language-Action (VLA) models, World Models (WMs) or Robot Foundation Models (RFMs). Prominent examples include Gemini Robotics from Google DeepMind, GR00T from Nvidia, Helix from Figure AI and π0 from Physical Intelligence.

Chinese-developed robotics models include proprietary systems used by robot makers such as Unitree and AgiBot, as well as ByteDance's GR-3 and GR-RL, Tencent's Hy-Embodied-VLM and Alibaba's RynnVLA-001.

The evolution of robotics AI depends heavily on physical-world data used for training. Given China's rapidly growing humanoid robot shipments, the industry could benefit from a virtuous cycle of model development, data collection and model retraining. Considering Unitree's market position, data collected from its deployed robots may significantly accelerate DeepSeek's robotics AI development.

According to market research firms, the global humanoid robot market is currently dominated by China's AgiBot and Unitree.

Omdia estimates that AgiBot shipped 5,168 units in 2025, representing 39% of the global total of 13,000 units, while Unitree shipped 4,200 units, accounting for 32%. Counterpoint Research estimates a larger market of roughly 16,000 units, with AgiBot holding a 31% share and Unitree 27%.

By contrast, China's CCID (China Center for Information Industry Development) ranked Unitree as the market leader with a 32.4% share. Unitree has stated that its actual humanoid robot sales and shipments exceeded 5,500 units in 2025, a figure included in its IPO prospectus.

Even if AgiBot remains the leader, the shipment gap between the two companies is only around 330 units according to Omdia's estimates. In an early-stage market totaling approximately 13,000 to 18,000 units annually, such differences are small enough for rankings to vary depending on methodology.

AgiBot's developer, AgiBot Innovation Technology, was founded with the goal of developing general-purpose embodied AI robots. The company has built a vertically integrated portfolio covering robot hardware, components, robotics AI models, VLA and embodied AI models, robot data collection platforms and training datasets.

In March 2025, AgiBot introduced its GO-1 embodied AI foundation model, followed by WholeBodyVLA in December. It also built a large-scale robotics data platform called AgiBot World, which reportedly contains more than one million robot manipulation trajectories across 217 tasks.

As a result, Unitree has generally been viewed as a company with strong robot manufacturing capabilities and a cost advantage, while AgiBot has been seen as a provider of integrated robot hardware, data and embodied AI. The DeepSeek-Unitree partnership could blur that distinction.

Unitree is also developing embodied AI models internally and is reportedly training robot behaviors in simulation environments.

A report published in April by the Mercator Institute for China Studies (MERICS) said Unitree has been using Nvidia Isaac Lab to train robots for complex behaviors. Unitree's IPO prospectus also states that 48% of the funds raised through the IPO will be invested in embodied AI model training. The company appears poised to accelerate its internally developed embodied AI efforts by combining them with DeepSeek's models and training capabilities.

The competitive focus in humanoid robotics is increasingly shifting toward intelligence rather than locomotion alone. Beyond walking, running and finger dexterity, the key challenge is enabling robots to understand and execute human instructions autonomously. A task such as "Bring three red pencils from the blue basket on the desk and place them on the dining table" requires the integration of multimodal AI, reasoning, VLA models, world models and robot control systems.

Unitree aims to strengthen those capabilities through DeepSeek, while DeepSeek gains entry into a new category of AI models. DeepSeek's rise disrupted the dominance of high-cost U.S. AI models, leading to terms such as the "DeepSeek Moment" and "DeepSeek Shock." However, DeepSeek's existing systems are fundamentally large language models (LLMs). Expanding into robotics requires access to physical-world training data, which Unitree can provide.

DeepSeek has also been viewed as relatively weaker in multimodal integration. The partnership with Unitree could help strengthen those capabilities, as robotics models rely heavily on spatial understanding derived from camera-based perception before taking physical actions.

Meanwhile, Unitree is scheduled to begin IPO subscriptions on Aug. 10 on Shanghai's STAR Market before commencing trading. On Aug. 7, the company set its IPO price at 150.80 yuan per share, implying a valuation of approximately 61 billion yuan. Unitree plans to issue about 40.45 million new shares and raise roughly 6.1 billion yuan.

LG Uplus said during its second-quarter 2026 earnings conference call that it plans to pursue a growth strategy centered on two pillars in the second half of the year: AI data centers (AIDC) and AI contact centers (AICC).

According to the company, revenue from its enterprise infrastructure division rose 8.6% year over year to 464.4 billion won in the second quarter. Within the segment, AIDC revenue increased 28.9% to 124.1 billion won, driven by growth in colocation services. The company said AIDC was the primary driver of growth in the enterprise infrastructure business.

LG Uplus is currently building a 200-megawatt (MW) AIDC campus in Paju, which it described as the largest facility of its kind in the Seoul metropolitan area. The company plans to open the first data center building in the first half of next year and said it has accelerated the construction schedule to meet rapidly growing demand for AIDC infrastructure.

The company also plans to secure additional power capacity through design-build-operate (DBO) projects while simultaneously expanding into regional locations. “Although it is difficult to precisely forecast GPU-based AI demand, we have secured a phased and scalable AIDC pipeline that can respond flexibly to changes in demand,” the company said.

During the conference call, Jung Young-hoon, head of Enterprise AI Business at LG Uplus, said demand in the AIDC market can be broadly divided between inference services concentrated in the Seoul metropolitan area and training and non-real-time inference workloads located in regional areas. He added that the company is investing a total of about 2 trillion won, including an additional 1.3 trillion won for the second phase of the Paju AIDC project.

Addressing questions about financing for AIDC construction, the company stressed that the projects are being funded without external capital raising.

Yeo Myung-hee, chief financial officer (CFO) of LG Uplus, said capital expenditures (CAPEX) related to AIDC will increase this year compared with last year, but will remain within the company’s EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization) framework.

“We are improving capital efficiency and reducing pressure on cash flow by combining our own facilities with lighter models such as leasing and DBO projects,” Yeo said. “As a result, we plan to pursue our medium- to long-term AIDC targets without separate external financing.” She added that investment plans could change depending on market conditions and that the company would communicate any changes promptly.

For its AICC business, LG Uplus said it will continue enhancing its solutions by incorporating operational expertise and customer pain points gained as an actual user running customer service centers with more than 4,000 agents.

During the first half of the year, the company focused on deployment-based projects in the financial sector and secured reference customers. In the second half, it plans to expand its product lineup with large language model (LLM)-based offerings in addition to existing products such as Consultation Advisor and AutoQA. The company said it aims to accelerate tangible results from AI-driven new businesses by leveraging the competitiveness of its internalized AICC solutions.

Revenue from the solutions business, which includes AICC and smart mobility services, rose 6.8% year over year to 134.6 billion won in the second quarter, supported by growth in messaging relay services.

The conference call also included questions regarding profitability differences between conventional internet data centers (IDC) and AIDC, as well as potential business and role overlap between LG Uplus and LG CNS within the LG Group’s AI strategy.

The company said profitability at newly built AIDCs, including the Paju facility, is expected to exceed that of traditional IDC operations. Data center revenue is generally divided into space rental fees and electricity charges. Space rental fees command higher pricing because investments in liquid cooling and power infrastructure are required to support AI servers and ensure operational stability.

LG Uplus also noted that demand continues to significantly exceed supply, pushing space rental unit pricing higher. While electricity fees are typically charged on a metered basis according to usage, the higher utilization rates of AI servers are also increasing overall electricity-related revenue, the company said.

Regarding potential overlap in business and roles with LG CNS within the group, the company avoided direct comparisons but said both firms possess capabilities in the design, construction and operation of AIDCs, while their business models differ according to their respective strengths and customer bases.

The company said LG Group is pursuing a “One LG” strategy to advance AI infrastructure and create new value. “The group will strengthen competitiveness in AI infrastructure by combining cooling, power operations and network capabilities across affiliates, while enhancing the value chain through integration with AI technologies such as EXAONE from LG AI Research,” it said.

Within that framework, LG Uplus emphasized that it serves as a core operator handling AI infrastructure operations. As a provider capable of delivering both data center and telecommunications network services, the company said it is expanding its maximum data center capacity through both self-owned facilities and external DBO projects, which it began pursuing last year.

“Based on our AIDC infrastructure in key metropolitan locations, network capabilities and data center operating expertise, we plan to continue expanding AI infrastructure to meet growing demand,” LG Uplus said. “Through this approach, we will maximize synergies across the group, manage operational risks through demand-driven investment and improve investment efficiency.”

OpsNow, an AI-powered multi-cloud management platform provider, said on Aug. 6 that it has simultaneously obtained ISO/IEC 27001:2022 certification for information security management systems and ISO/IEC 27701:2025 certification for privacy information management systems. The certifications cover the development, operation and delivery of its cloud management platform.

ISO/IEC 27001:2022 is a globally recognized information security standard awarded after an assessment of 93 control measures across four areas: organizational, human, physical and technological security. ISO/IEC 27701:2025, meanwhile, was revised in October 2025 and transitioned from an extension of ISO/IEC 27001 into an independent management system standard.

Organizations certified under the previous 2019 version are required to transition to the updated standard by October 2028. OpsNow said it received certification under the latest 2025 version, validating its management framework in its role as a personally identifiable information (PII) controller. The certifications are valid for three years.

With the new certifications, OpsNow said an independent external audit has verified that its management systems for platform development and operations, as well as its privacy governance framework as a PII controller, comply with international standards.

OpsNow's multi-cloud management platform provides services ranging from cloud infrastructure resource management and multi-cloud cost optimization to cloud security posture management (CSPM) and AI-based operational automation. The company said security is a core competitive factor because the platform directly handles customers' critical infrastructure, cost and configuration data.

OpsNow expects the certifications to accelerate its expansion into global markets. The company said the internationally recognized management frameworks will help satisfy due diligence requirements commonly requested by customers in the United States, Europe and Asia, strengthening its ability to win overseas business. It also expects stronger momentum in the enterprise market, particularly among large corporations, financial institutions and public-sector organizations where security regulations and compliance requirements are stringent.

Trimble, a global provider of 3D building information modeling (BIM) solutions, said it has successfully deployed its Trimble Connect and Trimble Connect Dashboard platforms at Kolon Global, establishing a BIM-based digital integrated management system for steel frame construction that enables real-time monitoring of fabrication, shipment, delivery and installation progress.

Kolon Global said it has been diversifying its business portfolio beyond housing projects into general building construction and high-tech facilities, while expanding its BIM processes from reinforced concrete (RC) structures to steel and precast concrete (PC) structures. As part of that transition, the company identified the need to reduce data discrepancies and communication issues associated with 2D drawing- and spreadsheet-based management systems, while shortening the performance aggregation process—which previously took more than a week—to quickly identify schedule delays and project risks.

Through Trimble Connect and Trimble Connect Dashboard, Kolon Global said it can now manage the entire steel construction process, from fabrication to installation, through a unified real-time digital platform.

When partner companies upload shop models created with Tekla software to Trimble Connect, the status of each steel component is displayed through color-coded indicators on a 3D model. By linking QR codes assigned to individual components and shipping invoices, fabrication, shipment, delivery and installation data are updated in real time, enabling construction sites, headquarters and subcontractors to track project progress using the same dataset.

Following implementation, the company said performance aggregation processes that previously required more than a week have been replaced by a real-time monitoring system. The platform also provides visibility into material deliveries, installable inventory and actual performance against plans, allowing delays to be identified early and enabling faster corrective actions such as adjusting production volumes or revising installation schedules.

In addition, quantity information from 3D shop models can be compared and verified against subcontractor progress payment claims, helping prevent overbilling and enabling more accurate settlements through component-level verification. The company also said the system has improved decision-making speed by allowing project owners and supervisors to review construction progress using the same data standards without requiring additional reporting work.

Kolon Global plans to expand the system beyond steel construction to precast concrete processes and eventually deploy a standardized digital construction management framework across all of its steel and PC project sites.

LG Uplus reported record quarterly earnings on Aug. 6, posting consolidated operating profit of 344.5 billion won for the second quarter. The figure increased 13.1% from a year earlier and 26.5% from the previous quarter, marking the highest quarterly operating profit in the company's history.

Operating margin, based on service revenue, improved by 1.1 percentage points year over year to 11.2%.

Revenue declined 3.9% from a year earlier to 3.69 trillion won, primarily due to weaker handset sales.

Service revenue, which reflects the company's core telecommunications business, increased 2.0% to 3.08 trillion won. Net profit rose 0.3% to 217.7 billion won, while EBITDA, a measure of cash-generating ability, climbed 5.9% to 1.03 trillion won. EBITDA margin reached 33.6%.

Corporate infrastructure was the company's strongest growth segment. Revenue from the business rose 8.6% year over year to 464.4 billion won. Within the segment, AI data center (AIDC) revenue surged 28.9%, driven by higher colocation sales.

Smart home revenue increased 4.3% to 663.8 billion won, supported by growth in IPTV and broadband subscribers as well as higher average revenue per user (ARPU). Mobile revenue rose 0.9% to 1.66 trillion won. Total subscriber lines increased 5.2% year over year to 31.47 million, while the 5G penetration rate among handset subscribers reached 84.9%.

On the cost side, operating expenses declined 5.4% from a year earlier, contributing to improved profitability. Capital expenditures (CAPEX), however, rose 21.5% to 477.8 billion won as the company accelerated AI infrastructure investments, including the construction of its Paju AI data center.

LG Uplus also expanded shareholder returns. The company declared an interim dividend of 270 won per share for 2026, up 8.0% from a year earlier. In addition, it approved an additional 90 billion won share repurchase program on July 29 and plans to retire the repurchased shares.

Since announcing its corporate value enhancement plan in November 2024, LG Uplus has canceled approximately 180 billion won worth of treasury shares.

The company said it intends to proactively respond to growing infrastructure demand driven by the expansion of AI services.

The Paju AI data center will occupy a site measuring 73,712 square meters and is designed with a maximum power capacity of 200 megawatts. The first server building is scheduled for completion in June 2027, while the second and third buildings are expected to be completed in November 2028.

LG Uplus also said it is pursuing the global expansion of its AI call agent service, ixi-O, strengthening its cybersecurity business through the acquisition of security specialist PAGO Networks, and redesigning company-wide business processes using AI technologies.

Alibaba has officially launched Qwen 3.8-Max, its latest artificial intelligence model. First introduced as a preview version on July 20, Qwen 3.8-Max features 2.4 trillion parameters and supports a context window of up to 1 million tokens. Alibaba describes it as the most powerful flagship model in the Qwen family.

Alibaba said Qwen 3.8-Max ranked fifth on the Text Arena benchmark and second on the Vision Arena benchmark, demonstrating advanced capabilities in coding, real-world task execution, research and long-horizon problem solving. The company said the model was designed as a multimodal foundation model with native visual intelligence capabilities.

Qwen 3.8-Max is available through Alibaba Cloud's Model Studio as an application programming interface (API) service. The model weights are scheduled for release next week, and the model is also accessible through Alibaba's all-in-one workplace AI agent platform, QwenWork.

Built on Qwen 3.5 architecture, Qwen 3.8-Max combines Sparse Mixture-of-Experts (Sparse MoE) architecture with a hybrid attention mechanism. Alibaba said this architecture enables large-scale models to generate inference results efficiently.

Under this architecture, only 95 billion of the model's 2.4 trillion parameters are activated during inference. This allows the model to deliver performance on complex tasks while reducing computational costs.

Alibaba also highlighted Qwen 3.8-Max's autonomous coding and long-horizon task execution capabilities, claiming the model can work independently for extended periods without human intervention. In internal testing, the company said the model autonomously completed a real-world software engineering project over a 16-day period.

According to Alibaba, when tasked with building a self-evolving agent framework from scratch, Qwen 3.8-Max established an engineering loop that integrated user feedback, community best practices and self-generated testing data. The model then repeatedly generated code, conducted testing, created previews and analyzed logs before completing a self-evolving agent framework called "oh-my-cli."

The framework has been released as open-source software on GitHub. Qwen 3.8-Max also ranked fourth on the Frontend Code Arena benchmark.

Alibaba further said the model demonstrated problem-solving capabilities in highly specialized fields, including designing novel methodologies that surpassed results reported in existing research papers.

As a multimodal foundation model, Qwen 3.8-Max can transform static input data into dynamic and interactive knowledge structures, according to Alibaba. The company said the model can process hundreds of pages of documents, entire television series or 100-hour livestreams and convert them into searchable and interactive knowledge bases.

Alibaba said Qwen 3.8-Max can be used to edit raw personal footage into polished vlogs or generate immersive educational animations from text prompts alone. According to the company, the model can also reconstruct front-end web projects from a single user interface screenshot, transform 2D floor plans into detailed 3D interior visualizations, and create interactive games through natural-language instructions.

Woorinet, a provider of wired and wireless telecommunications equipment, has joined the South Korean government's Artificial Intelligence Data Center (AIDC) Alliance and is pursuing a national research and development (R&D) project to secure core technologies for Data Center Interconnect (DCI) networks linking AI data centers.

Launched under the leadership of the Ministry of Science and ICT (MSIT), the AIDC Alliance is a cooperative framework designed to strengthen South Korea's global competitiveness in AI data center technologies through collaboration among industry, academia, research institutes and government organizations across the entire value chain, including AI semiconductors, data centers, networking, power, cooling and software. The alliance is organized into demand, supply and infrastructure divisions. Woorinet has joined the supply division and will participate in the commercialization of next-generation DCI technologies.

As AI models continue to grow in scale, distributed training and distributed inference environments that simultaneously utilize graphics processing unit (GPU) resources across multiple data centers are becoming more common. This trend is increasing the need for backend network infrastructure capable of connecting AI computing resources.

Woorinet said it was selected in May as the lead organization for the Institute of Information & Communications Technology Planning & Evaluation (IITP)'s national R&D project titled "Development of AI Data Center Transmission Technology for Horizontal Expansion of AI Resources Across Data Centers." Working with the Electronics and Telecommunications Research Institute (ETRI), Sangmyung University and Pukyong National University, the company is developing ultra-high-speed transmission technologies between AI data centers and pursuing next-generation DCI technologies that would allow distributed AI resources to be utilized as a single AI infrastructure.

For AI data centers, DCI technology requires ultra-high bandwidth, lossless and ultra-low-latency communications, and advanced security capabilities. Woorinet said technologies based on the Ultra Ethernet Consortium (UEC) standard that can connect data centers located tens to hundreds of kilometers apart without data loss will become a key factor in determining AI infrastructure competitiveness.

To address these networking requirements, Woorinet said it is developing next-generation DCI technologies based on 16-terabit-per-second Packet Optical Transport Network (POTN) systems and wavelength-division multiplexing (WDM) optical transmission technologies capable of reliably transporting massive volumes of data.

The company also emphasized that it is strengthening its technological capabilities for building ultra-high-speed, ultra-low-latency transmission environments between AI data centers through proprietary optical transmission and network control technologies, including POTN, Packet Transport Network (PTN), Reconfigurable Optical Add-Drop Multiplexer (ROADM) and Transport Software-Defined Networking (T-SDN).

SpaceX reported stronger-than-expected revenue growth in its first quarterly earnings release since going public in June, but investor attention quickly shifted to the company's massive artificial intelligence (AI) infrastructure spending and continued losses in its AI and space businesses. The stock fell about 7% in after-hours trading following the earnings announcement.

For the second quarter of 2026, SpaceX reported revenue of $7.814 billion, up 92% from a year earlier, and a net loss of $541 million. While the company remained unprofitable, losses narrowed from the prior-year period. Earnings per share (EPS) came in at negative $0.09, beating the consensus estimate of negative $0.26.

Starlink remained the company's largest growth engine. Revenue from the Connectivity segment, which includes Starlink, reached $4.291 billion, exceeding analyst expectations of $3.83 billion. Subscriber numbers doubled year over year to 12 million, while segment revenue increased 66%.

The AI segment, which includes xAI, Grok and social platform X, generated approximately $2.6 billion in revenue, representing year-over-year growth of 247%. The company attributed the expansion to the integration of xAI and growing demand for AI computing services. However, the AI division posted an operating loss of $1.26 billion, although losses narrowed from the previous quarter.

SpaceX's third business segment, Space—which includes launch services and the Starship program—generated revenue of $962 million, up 29% from a year earlier, while recording an operating loss of $542 million. Although the AI segment delivered the fastest growth rate, Starlink's Connectivity business was the company's only profitable division.

While revenue and earnings exceeded expectations, investors focused on the scale of SpaceX's capital expenditures (CAPEX). Second-quarter capital spending surged to approximately $18.4 billion, compared with about $2.8 billion a year earlier. Most of the spending was directed toward AI data centers and computing infrastructure. SpaceX argued that the investments are necessary to support long-term growth, but investors expressed concern about the company's cash consumption.

During the earnings call, Chief Executive Officer (CEO) Elon Musk said SpaceX could generate $1 trillion in annual revenue by 2030 and suggested there was a "non-zero chance" of reaching that milestone as early as 2029. Musk also predicted that Starlink could eventually become one of the world's largest internet service providers.

"At some point, Starlink may carry a majority of internet traffic" in countries where the service is authorized, he said.

The earnings call drew significant attention as the company's first public earnings presentation following its IPO. However, investor and media reaction was notably cautious. TechCrunch noted that Musk frequently made more aggressive claims than other executives during the call, while senior management often provided more measured interpretations.

One example involved Starlink's future market position. After Musk said Starlink could handle a majority of global internet traffic within a decade, President and Chief Operating Officer (COO) Gwynne Shotwell later described the outlook more conservatively, saying Starlink could account for a "significant portion" of global internet traffic in the coming years.

A similar contrast emerged regarding revenue expectations. Chief Financial Officer (CFO) Brett Johnson said annual recurring revenue (ARR) could potentially reach $100 billion by year-end. About 20 minutes later, Musk described the figure as "not a question mark" and suggested it could be even higher.

Johnson attributed the ARR outlook to faster returns on new computing infrastructure investments, $6.7 billion in additional cloud-service contracts scheduled to begin in the October quarter, and revenue contributions from Cursor, which SpaceX recently acquired.

Investors remain skeptical about Musk's long-term targets. The $1 trillion revenue goal was first introduced during the company's IPO roadshow in June, when management discussed 2031 as a possible target year. During the latest earnings call, Musk accelerated that timeline by one year to 2030.

Based on SpaceX's reported 2025 revenue of $18.67 billion, reaching $1 trillion by 2030 would require approximately 54-fold growth over five years, implying a compound annual growth rate (CAGR) of roughly 120%.

SpaceX's current trajectory is stronger than that baseline suggests. With quarterly revenue already exceeding $7.8 billion, annualized revenue for 2026 could approach $32 billion or more. Even so, Wall Street projections remain far below Musk's target.

During the IPO roadshow, Goldman Sachs estimated SpaceX's 2030 revenue at roughly $470 billion, while Morgan Stanley projected approximately $330 billion. Both firms served as lead underwriters for the IPO, and even their forecasts were widely viewed as aggressive. As a result, some investors question whether the $1 trillion target represents a genuine internal objective or a narrative designed to support market sentiment.

Despite the stronger-than-expected quarter, the stock reaction underscored investor skepticism. SpaceX shares had risen 9.43% during regular trading on Aug. 4 to close at $125.33 ahead of the earnings release, but subsequently fell about 7.5% to 8% in after-hours trading.

Musk's trillion-dollar revenue vision remains far more ambitious than prevailing market expectations, but investors seeking clearer paths to profitability appeared unconvinced by promises tied to distant milestones and long-term projects.

POSCO DX has announced its transformation into an AI-Native Company, built on two strategic pillars: physical artificial intelligence (AI) for industrial operations and agentic AI for enterprise workflows.

The company said it is focusing on expanding physical AI to build intelligent factories where people, AI and industrial equipment or robots work together seamlessly. The strategy centers on applying physical AI technologies to automate and optimize the operation of large-scale industrial equipment through robotization.

POSCO DX said the technology has already been deployed in selected operations. It has applied physical AI to steel mill equipment weighing tens of tons, including overhead cranes that transport steel products, reclaimers that move raw materials in storage yards and ship unloaders, improving worker safety and productivity.

The company is also converting the tacit knowledge of highly experienced operators into core datasets for physical AI, enabling AI to collaborate with human operators in control rooms while managing robots and industrial equipment.

At a press briefing on July 29, Cho Seok-joo, executive vice president and head of the AX Research Institute at POSCO DX, outlined the company's progress in developing physical AI technologies.

POSCO DX is also expanding the deployment of industrial robots in hazardous and labor-intensive environments. The company said it provides end-to-end robotic automation services covering consulting, design, implementation and operations. Together with POSCO, it has developed a platform for managing heterogeneous robot fleets to optimize robot operations.

The company is also designing optimized robotic systems through simulation technologies and developing Vision-Language-Action (VLA) models for humanoid and other mobile robots to improve locomotion and manipulation capabilities.

As examples of practical deployment, POSCO DX highlighted robots that remove impurities from molten zinc baths in galvanizing processes, robotic band cutters used in cold rolling operations and automated guided vehicles (AGVs) capable of transporting steel products weighing up to 65 metric tons. The company is also deploying robotic systems for raw material sampling and logistics automation at plants producing environmentally friendly materials and automotive components.

For office and management functions, POSCO DX is promoting agentic AI through its proprietary Agentee platform. According to the company, Agentee manages the entire lifecycle of AI agents—from recruitment and evaluation to development, promotion and retirement—in a manner similar to managing human employees. The platform enables domain-specific "AI employees" to automate office and business operations.

POSCO DX said its AX consulting services first analyze customer workflows to determine how responsibilities should be divided between employees and AI agents. Based on those workflows, Agentee either creates customized AI agents or reuses existing ones. The company said the platform standardizes every stage from design to deployment, enabling rapid development of enterprise AI agents.

POSCO DX demonstrated several internal use cases, including finance, human resources, labor management and project management. In one example, an AI agent automates monthly pre-closing financial risk inspections by scanning all accounting accounts, identifying abnormal entries and missing journal records, and generating risk reports. The process, which previously required about three hours, now takes approximately one minute. The company estimates that expanding AI agents across all financial closing activities could improve productivity by at least 30%.

POSCO DX currently operates approximately 200 AI agents across its own organization and other POSCO Group companies. It plans to establish standardized AI agent models for the group before commercializing the validated platform for external customers.

The company said it is the only domestic provider with full-stack capabilities spanning operational technology (OT) on factory floors and enterprise information technology (IT) systems.

"Leveraging our expertise across both industrial operations and office environments, we aim to become the company best positioned to deliver physical AI and agentic AI services by integrating advanced AX technologies," POSCO DX said.

At the same briefing, Kim Byung-seok, head of the AI Workforce Task Force at POSCO DX, presented the company's agentic AI strategy.

Separately, POSCO Group has adopted a Mission-oriented AI Transformation (AX) strategy centered on three pillars—process, work and value—to accelerate AI adoption across the group.

At the POSCO CEO Investor Day held on July 2, the group unveiled its Triple-Core growth strategy spanning industrial resources (steel), strategic resources (including lithium, cathode and anode materials, and rare earth elements) and energy resources (LNG and renewable energy). The group also announced plans to expand into physical AI for process industries by leveraging decades of experience in factory automation, intelligent manufacturing and large-scale industrial data.

The POSCO CEO Investor Day is the group's annual investor relations event, where the chairman and senior executives present long-term growth strategies, future business plans and financial targets to shareholders, institutional investors and securities analysts.

Autonomous trucking startup Mars Auto has signed a memorandum of understanding with the Korea Post Logistics Support Foundation and logistics company Logisquare to accelerate the adoption of autonomous driving technologies for commercial freight transportation. The partnership will expand paid autonomous freight services within Korea Post's logistics network while using real-world driving data to enhance the company's end-to-end (E2E) autonomous driving AI model.

The three-way agreement is part of the South Korean government's national AI initiative, "Software-Defined Vehicle (SDV) Transformation and AI-Based End-to-End Autonomous Driving Model Advancement." In November last year, Mars Auto was selected by the Ministry of Trade, Industry and Resources and the Korea Evaluation Institute of Industrial Technology (KEIT) to lead a project developing E2E AI-based autonomous freight transportation technology for heavy-duty trucks. The project has a total budget of 18.2 billion won and is being carried out by a consortium of 13 organizations, including the Korea Post Logistics Support Foundation and Logisquare.

Under the partnership, Mars Auto will install its Copilot autonomous driving system on postal freight vehicles operated by the Korea Post Logistics Support Foundation, while Logisquare will oversee operations for autonomous commercial freight routes. By collecting diverse real-world driving data from heavy-duty truck drivers, the partners aim to improve E2E AI performance while increasing operational efficiency and safety through fuel savings of more than 10% and reduced driver fatigue during long-distance and nighttime operations.

Postal freight vehicles equipped with Mars Auto's Copilot system will operate on major highways and local roads connecting 22 regional mail processing centers and approximately 3,700 post offices across South Korea. Because the system is not limited to a specific vehicle manufacturer or truck model, it can collect driving data from a broad range of commercial vehicles. The data will be transmitted in real time over SK Telecom's wireless network to the National AI Project training infrastructure located in Yeongdeungpo, Seoul. Copilot has already been installed on 20 postal freight vehicles, with additional deployments planned.

Mars Auto's autonomous driving technology relies on AI and camera-based sensors rather than LiDAR or high-definition (HD) maps. The company believes the quantity and quality of real-world driving data are critical to improving AI performance and said the partnership will support its goal of commercializing Level 4 fully autonomous driving.

Separately, on July 20, Mars Auto became the first company in South Korea to receive government approval to launch an autonomous trailer business under the Ministry of Land, Infrastructure and Transport's mobility regulatory sandbox program.

AI infrastructure software company Xiilab announced on July 29 that it has joined the AI Data Center (AIDC) Alliance, a public-private consultative body led by South Korea's Ministry of Science and ICT.

The AIDC Alliance was established by the government together with industry, academia and research organizations to foster AI data centers as a national strategic industry and a new export growth engine. The alliance discusses policies and institutional improvements related to AIDC development, including power infrastructure, site development and financing.

The South Korean government plans to build a total of 18.4GW of AI data center capacity by 2035 while expanding cumulative investment to more than 1 quadrillion won. It also aims to develop a domestic full-stack AI ecosystem integrating locally developed neural processing units (NPUs), cloud platforms, power and cooling systems, and operational software.

Xiilab said it plans to deploy its AstraGo platform during the operational phase of AI data centers to improve GPU utilization through Kubernetes-based GPU scheduling, automatic GPU partitioning and multi-node distributed training.

The company said it has established expertise in AI infrastructure services, including AI cluster design, reference architecture (RA)-compliant deployment, performance validation and operational software. Xiilab added that it recently achieved Nvidia Elite Partner status in the software category.

Kolon Benit is expanding its enterprise software business with a portfolio of in-house solutions covering business planning, consolidated accounting, financial disclosure and manufacturing digital transformation (DX). The company aims to broaden its customer base by targeting growing demand for regulatory compliance and workflow automation in finance and manufacturing.

Kolon Benit's proprietary software portfolio includes r-PLANNA for business planning and profit forecasting, BENIT SIGMA for consolidated accounting, DSDFlow for electronic financial disclosures and r-CoCoAna, a manufacturing DX platform. The company plans to progressively integrate artificial intelligence (AI) capabilities into these products to further automate business planning, profit forecasting, financial consolidation, regulatory reporting and manufacturing operations.

The r-PLANNA solution helps companies analyze projected costs and profitability during the business planning process. Built on SAP S/4HANA, it performs repeated scenario-based forecasting and analysis by incorporating variables such as exchange rates, selling prices, raw material costs and production volumes. Kolon Benit said the solution uses a proprietary calculation engine based on the same mechanism as SAP systems, enabling highly reliable forecasts. In customer validation projects, forecast production costs matched actual costs with approximately 99% accuracy.

BENIT SIGMA and DSDFlow support financial reporting workflows from group consolidation through regulatory disclosure. BENIT SIGMA collects and consolidates financial data from multiple subsidiaries to produce consolidated financial statements while automating complex accounting processes, including intercompany transactions and ownership relationships. The solution supports both SAP and non-SAP enterprise resource planning (ERP) environments and has accumulated more than 100 customers since its launch in 2009.

To prepare customers for the implementation of IFRS 18 in 2027, Kolon Benit recently introduced the next-generation BENIT SIGMA Pro. The new solution provides dual reporting capabilities that allow users to compare financial statements prepared under existing accounting standards with those based on IFRS 18. It also includes annotation functions for generating Management-defined Performance Measures (MPMs) specified by corporate management. The company plans to further automate repetitive consolidation, financial reporting and data validation tasks by incorporating AI technologies.

DSDFlow converts financial data generated during the consolidation process into formats required for submission to South Korea's Financial Supervisory Service's Data Analysis, Retrieval and Transfer (DART) system while supporting preparation and validation of disclosure documents.

In manufacturing, Kolon Benit is strengthening its business with the r-CoCoAna platform. The system collects and integrates real-time data from production equipment and manufacturing processes, providing a unified platform for managing production, quality, inventory, equipment and energy operations using data-driven analytics.

The platform combines a Connected Data Platform (CDP), which links data from both legacy and modern manufacturing equipment, with a Manufacturing Operations Management (MOM) system that integrates production planning, process execution, quality management and equipment management. It also utilizes technologies such as vision AI to automatically detect product defects and analyze equipment and process data.

Kolon Benit said it has built manufacturing DX expertise through r-CoCoAna since establishing a dedicated manufacturing business organization in 2014. Over the past three years, the company has carried out projects for manufacturing affiliates within the Kolon Group, including Kolon Industries, Kolon Life Science and Kolon Pharmaceutical, covering data collection and analysis, equipment monitoring, process optimization and quality stabilization.

The company said those projects enabled it to build end-to-end execution capabilities spanning factory assessments, process improvement planning, system implementation and operational stabilization. Going forward, Kolon Benit plans to expand its integrated manufacturing DX package—which combines consulting, data collection and analytics, system implementation and AI-based operations—to external customers to help reduce costs, improve productivity and enhance product quality.

Microsoft shares rose more than 8% in after-hours trading following the release of its fiscal 2026 fourth-quarter (April–June) results. The gains were driven by stronger-than-expected growth in the company’s Azure cloud business, lower-than-anticipated capital expenditure (Capex) plans, and continued confidence in its cash-generation outlook. Microsoft also issued fiscal 2027 first-quarter guidance that exceeded market expectations for both revenue and cloud growth.

Quarterly revenue reached $90.01 billion, surpassing the consensus estimate of $87.62 billion. Revenue from Azure and other cloud services increased 43%, exceeding analyst expectations. StreetAccount had projected growth of 40.2%, while Visible Alpha forecast 39.98%.

Azure is reported within Microsoft’s Intelligent Cloud segment, which includes Azure and other cloud services, server products and related cloud services, and enterprise services. Microsoft does not disclose standalone Azure revenue. In its earnings report, the company said Azure and other cloud services revenue grew 43% during the quarter.

Chief Executive Officer (CEO) Satya Nadella disclosed that Azure’s annual revenue exceeded $100 billion for the first time. CEO Nadella also emphasized that paid licenses for Microsoft 365 Copilot had surpassed 30 million.

He further stated that Microsoft expects to continue generating strong cash flow throughout fiscal 2027 (July 2026–June 2027).

Beginning in fiscal 2027, Microsoft will extend the estimated useful life of its data centers and office buildings from 15 years to 25 years. The longer depreciation period will result in some future data center lease agreements being classified as operating leases rather than finance leases, reducing the amount recorded as Capex under accounting rules. Microsoft emphasized that excluding this accounting change, its actual data center investment plans remain unchanged.

The move has helped ease investor concerns that surging data center spending and AI services could undermine the profitability of Microsoft’s core productivity software business.

Azure’s 43% growth also countered speculation that Google Cloud had been taking market share from Amazon Web Services (AWS) and Azure after Alphabet reported an 82% surge in Google Cloud revenue last week. The latest results indicate Microsoft remains firmly among the leaders in the cloud infrastructure market.

Microsoft’s commercial remaining performance obligations (RPO) reached $678 billion at quarter-end. RPO represents contracted revenue that has not yet been recognized. The figure increased by approximately $51 billion from $627 billion in the previous quarter. The company said all of the quarter-over-quarter increase came from customers other than frontier AI model developers such as OpenAI, suggesting demand for Microsoft’s cloud and enterprise software offerings is broadening across a diverse customer base.

The number of paid Microsoft 365 Copilot licenses rose from 20 million in the previous quarter. Analysts had expected an average of 26.9 million licenses, meaning Microsoft exceeded expectations by more than 3 million customers.

Microsoft forecast fiscal 2027 first-quarter (July–September) revenue of between $89.85 billion and $90.95 billion. The midpoint of $90.4 billion is above the consensus estimate of $89.66 billion compiled by LSEG. The company also projected Azure growth of 45% on a constant-currency basis, well above Visible Alpha’s forecast of 40.92%.

Microsoft expects first-quarter fiscal 2027 Capex of $50 billion and calendar-year 2026 Capex of $175 billion. Both figures are below previous expectations of $56.02 billion and $190 billion, respectively.

Large-scale data center investments continue to weigh on free cash flow (FCF). Microsoft reported quarterly FCF of $19.6 billion, calculated by subtracting $35.8 billion in cash capital expenditures from operating cash flow of $55.4 billion. Although FCF declined 23% from a year earlier, it significantly exceeded the Visible Alpha consensus estimate of $13.44 billion.

Total Capex for the quarter reached $41 billion, up more than 70% year over year but below market expectations of $42.37 billion. The figure included $5.6 billion in finance leases related to data center sites and other facilities. Cash spending on property and equipment totaled $35.8 billion. Total Capex in the previous quarter was $31.9 billion.

Additional investment could still lie ahead. In filings with the US Securities and Exchange Commission (SEC), Microsoft disclosed $329.1 billion in future data center lease commitments that have not yet commenced. Those agreements are scheduled to begin between fiscal 2027 and fiscal 2033, with lease terms ranging from one to 20 years.

Microsoft is also working to reduce its reliance on OpenAI technology by adding Anthropic models to its services and developing proprietary AI models. At the same time, the company is leveraging long-standing enterprise customer relationships to expand sales of its $30-per-month Copilot offering. A major agreement signed with Accenture earlier this year is one example.

Reuters noted that Microsoft has been among the weakest-performing members of the “Magnificent Seven”—Microsoft, Apple, Nvidia, Alphabet, Amazon, Meta and Tesla—so far this year, but the latest earnings report suggests the company may be positioned for a rebound. Microsoft shares had declined approximately 18% year to date before the earnings release.

Because the quarter marked Microsoft’s fiscal 2026 fourth quarter, the company also reported full-year results for fiscal 2026, which covered the period from July 2025 through June 2026.

For fiscal 2026, Microsoft reported revenue of $331.8 billion, up 18% year over year, or 16% on a constant-currency basis. Operating income rose 21% to $155.2 billion, or 19% on a constant-currency basis. Net income reached $133.7 billion, increasing 31% under generally accepted accounting principles (GAAP), or 22% on a non-GAAP basis excluding gains and losses related to its OpenAI investment. On a constant-currency basis, non-GAAP net income increased 20%. Diluted earnings per share (EPS) were $17.95, up 32% year over year under GAAP.

Both fiscal 2026 annual revenue and fourth-quarter revenue increased 18% from the previous year. However, the fact that annual and quarterly growth rates were identical does not necessarily indicate that growth was evenly distributed throughout the year. Microsoft also separately disclosed non-GAAP results excluding the impact of its OpenAI investment. During fiscal 2026, accounting gains related to that investment increased GAAP net income by $4.963 billion and diluted EPS by $0.67. The company noted that this accounting benefit does not represent cash earnings received from OpenAI.