Disney is testing AI-driven search and content discovery features on ESPN and Disney+ platforms.

Doubao has proudly announced a significant enhancement to its video call functionality, incorporating the cutting-edge native audio and video full-duplex large model, SeedRealtime. This innovative model facilitates real-time video communication, continuous environmental awareness, and natural voice exchanges, allowing for instantaneous comprehension of audio-visual data without the necessity for discrete processing steps. Post-upgrade, Doubao's video calls now seamlessly integrate audio and visual cues to grasp user requirements on the fly, accurately identify speakers, and engage in proactive interactions, thereby minimizing instances of misinterpretation or disruptions. The video call feature, now fortified with this advanced model, has been fully deployed. Users are invited to upgrade to the latest version of the Doubao APP to enjoy this complimentary enhancement. Doubao is evolving from a mere question-and-answer AI into an intelligent assistant that proactively offers support. Looking ahead, Doubao envisions extending its real-time multimodal capabilities to a range of products, including AI glasses, thereby further enriching the user experience.

Microsoft employs a two-pronged approach, combining investment with in-house research. In the fiscal year 2026, Microsoft's commercial partnership with OpenAI generated a staggering $24.1 billion in revenue, constituting roughly 70% of its total AI business income. Notably, Microsoft has poured $11.9 billion into OpenAI, reaping an impressive return on investment of around 10 times. Beyond OpenAI, Microsoft has also ventured into AI firms like Anthropic and Mistral. Specifically, its investment in Anthropic yielded $3.2 billion in revenue during the fourth quarter of fiscal year 2026.

The collaborative framework between Microsoft and OpenAI has evolved from an exclusive partnership to a more competitive one, as OpenAI has inked a cooperation pact with Amazon. Amazon, too, has made investments in both OpenAI and Anthropic. The substantial uptick in Amazon's net profit during the second quarter can largely be attributed to its stake in Anthropic, which has pledged to allocate at least $100 billion towards AWS technology over the forthcoming decade.

To mitigate dependence on a singular model, both Microsoft and Amazon have significantly ramped up their AI-related capital outlays and are fervently pursuing the development of their proprietary research models. Notably, Microsoft has unveiled the MAI model family, marking a significant stride in its AI endeavors.

On August 7, reports emerged that the National Energy Administration had unveiled the '15th Five-Year Action Plan for Workplace Safety in the Electric Power Sector'. This plan outlines a strategy to bolster the utilization of 'AI+' technologies in safety management, pioneer new approaches for equipment fault prediction and health monitoring, seamlessly integrate artificial intelligence into intelligent safety tools, and delve into research on decision-support technologies grounded in large-scale AI models.
Simultaneously, there will be a concerted push to accelerate independent R&D endeavors for vital power equipment, coupled with an emphasis on the advancement of novel protective materials. Dedicated initiatives will be launched to conquer core component technologies for power equipment, propelling forward breakthroughs in pivotal areas such as power chips and ultra-high voltage (UHV) components.
Furthermore, the plan advocates for the promotion of innovative safety and quality control technologies in power construction projects. This includes exploring the establishment of an intelligent regulatory framework and harnessing the power of AI, big data, and other cutting-edge tools to fortify remote supervision and quality assurance measures for major power projects.

Media reports have indicated that, in a recent show, singer Wang Feng disclosed that following the comprehensive integration of AI technology within his company, the workforce was downsized from 1,100 employees to merely 400, a revelation that has ignited widespread debate. According to the Tianyancha App, Wang Feng is linked to over 20 businesses, with a number of them, including Nanjing Guojiang Cultural Development Co., Ltd. and Jiangxi Feier Technology Co., Ltd., remaining active and operational.

On August 7, Firmus, an Australian company specializing in artificial intelligence infrastructure, revealed that it had successfully closed a strategic equity funding round worth US$2 billion. The roster of investors included tech giant NVIDIA, investment firm Coatue, various funds managed by Blackstone Group, as well as other investment entities under Blackstone's umbrella. Global trading and technology powerhouse Jane Street also joined the funding round. Firmus explained that the capital raised in this round would be earmarked for expediting the development of its 'Southgate AI Factory' project in Australia and laying the groundwork for subsequent expansion into other Asia-Pacific markets.

On August 7, the large model concept stocks listed on the Hong Kong stock market experienced a notable upswing. Among them, MINIMAX-W stood out with a remarkable surge of over 8%, while Zhipu also posted a solid gain of more than 6%. Notably, since its inclusion in the Hong Kong Stock Connect on August 6, MINIMAX-W has witnessed an impressive cumulative increase of over 30% within just two days, reflecting strong investor confidence and market enthusiasm in this sector.

On August 7, Nobitron's (02635.HK) stock price was recorded at HK$11.88, marking a 4.39% increase, with a trading volume reaching HK$40.4846 million. The previous day, August 6, the company disclosed that it had recently entered into a software sales agreement with a premier domestic provider of AI computing chips and intelligent computing infrastructure, with the total contract value amounting to 50.8 million yuan (inclusive of tax). As per the agreement, Nobitron will supply a range of its in-house developed software products and complementary technical services. These offerings include the NBK Industry Application Platform V2.0 series software, the GIS Positioning Map Assisted Positioning Management System V1.0, the Intelligent Robot Inspection and Electrical Operation Assistance System V1.0, as well as the AI Operations and Maintenance Big Data Analysis Software. This partnership signifies a significant leap forward in the commercialization of Nobitron's AI industrial software products within intelligent computing center environments.

On August 7, both the App and PC versions of QianWen underwent significant functional enhancements, introducing a suite of new features: Thinking and Research, Scheduled Tasks, Office Assistant, Agent Plaza, and Voice Calls. Additionally, the platform now supports Alibaba's cutting-edge flagship model, Qwen3.8-MAX, which is accessible to all users at no cost. Among these updates, the 'Thinking and Research' feature represents a substantial upgrade from the previous 'Deep Thinking' module, boasting improved reasoning and information integration abilities. The Office Assistant is designed to tackle complex office tasks with autonomy, breaking down objectives, employing various tools, and delivering results efficiently. It facilitates seamless cross-device collaboration between mobile and PC platforms. For professional users, the Office Assistant also provides premium options, offering additional usage quotas for a fee.

On August 7, 2026, OpenAI announced an update to the ChatGPT experience. On the paid side, the GPT-5.6 Sol model used by Plus and Pro users has been re-tuned for higher factual accuracy and more focused responses, with the addition of a slider function to control the "depth of thought." On the free side, the default model has been switched to GPT-5.6 Luna, with unlimited text conversations and a new Think button to handle questions requiring more reasoning.

According to Reuters, citing sources with knowledge of the matter, Alibaba intends to introduce a revenue-sharing model for major commercial users of its next-generation Qwen open-source AI model, though the exact revenue-sharing ratio remains to be finalized. This move follows similar steps taken by Moonshot AI for its Kimi K3 model, which mandated commercial agreements for entities with annual sales surpassing $20 million. At present, developers using the model on Alibaba Cloud’s computing platform are required to pay, while no charges are imposed for running open-source models in their own data centers.

According to AI Pulse, Seedance 2.5, ByteDance's advanced video generation large model, is scheduled to commence its API public beta test at 2 PM today. Earlier, the pricing for the model's API was unveiled, revealing a cost of 42 yuan per million tokens for video input and 70 yuan per million tokens for non-video input. These prices reflect an increase of 14 yuan and 24 yuan, respectively, when compared to the rates for Seedance 2.0.

According to an exchange announcement, Chinese AI startup DeepSeek has invested 140.8 million yuan (approximately $20.8 million) in robotics manufacturer Unitree Technology (688836.SS), with the two parties set to jointly develop AI models for humanoid robots. DeepSeek has acquired 933,399 shares in Unitree Technology, representing 2.31% of the total strategic placement shares in its Shanghai IPO.

DeepSeek, renowned in the past for compelling global AI firms to slash prices due to its exceptional cost-efficiency, is now set to revise its approach. Lately, the company has embarked on two parallel endeavors: firstly, initiating a fresh funding round with the objective of amassing roughly 50 billion yuan, projecting the company's valuation to soar to 500 billion yuan; secondly, declaring a substantial surge in API service pricing.

In July 2026, the artificial intelligence firm Moonshot AI unveiled its latest-generation model, Kimi K3. Boasting a staggering 2.8 trillion parameters, it emerged as the world's inaugural 3-trillion-parameter model with open weights. This model incorporates a Mixture of Experts (MoE) architecture, supports a context window of up to 1 million tokens, and demonstrates exceptional proficiency in programming and visual comprehension tasks. It secures the third position globally in terms of overall intelligence. Tesla's CEO, Elon Musk, commented beneath the evaluation report, describing it as "remarkable." This marks his second public endorsement of the team's work.

According to a report by the Financial Times, Google is transferring the reins of its AI operations from London to Silicon Valley. The sweeping leadership overhaul announced on Wednesday is set to cement the influence of co-founder Sergey Brin, following his return to the company. This move entails a significant reconfiguration of Google's AI leadership structure, with a notable concentration of power in California. The shift aligns with Google's strategic push to commercialize the Gemini model and bridge the competitive gap with rivals.

DeepMind founder Demis Hassabis is relinquishing his day-to-day operational duties to assume the roles of chairman and Alphabet's chief scientist. Meanwhile, the mantle of daily management will be passed on to the newly appointed Senior Vice President, Koray Kavukcuoglu. The primary objectives of this reshuffle are to expedite decision-making processes, bolster commercialization efficiency, and effectively counter the mounting pressure from competitors, notably OpenAI.

On August 5 (local time), SpaceX took to social media to announce its partnership with NVIDIA in developing the Starmind AI1 satellite computing payload. Each Starmind satellite will be outfitted with NVIDIA's Rubin GPU and Vera CPU, enabling it to deliver data center-level computing capabilities in space. NVIDIA has also verified this collaboration, highlighting that AI infrastructure is venturing into uncharted territories. Concurrently, China is stepping up its efforts in this domain, with the space-based computing power industry transitioning from theoretical exploration to on-orbit verification and initial networking phases. According to a research report by Changjiang Securities, space-based computing power is on track to become a substantial complement to China's computing power resources, with the industry poised for significant breakthroughs.

On August 6, Disney announced that its ESPN and Disney+ platforms are testing AI-powered search and content discovery features designed to help users access sports information and entertainment content more quickly by describing their needs in natural language. Specifically, ESPN is testing an AI-assisted search function called 'ESPN Search' with select users, allowing them to ask sports-related questions in natural language. Based on ESPN's sports content and data ecosystem, the function provides answers, related content, statistics, and recommendations, drawing from sources such as ESPN articles, videos, research data, internal knowledge bases, and sports data systems.

On August 7th, information from the Korea Economic Daily revealed that Koo Kwang - mo, the Chairman of LG Group, is slated to have a meeting with Jensen Huang, the CEO of NVIDIA, at the Silicon Valley headquarters in the upcoming week. The two parties are likely to deliberate on areas of cooperation, including embodied artificial intelligence (Note: "physical AI" is translated into the more commonly used term "embodied AI" in the AI field, which emphasizes the interaction between AI and the physical world) and smart factories. Moreover, they might also engage in in - depth discussions about the specifics concerning LG's acquisition of Blackwell GPUs.

According to sources familiar with the matter, OpenAI's highly anticipated new device will feature a unique shape and movable components designed to give it distinct personality traits, with a price tag that may exceed $300. The sources, who requested anonymity due to confidentiality requirements, stated that the product is essentially a display-less smart speaker shaped like a donut and roughly the size of a hockey puck, allowing users to easily hold it in one hand and move it around the home. OpenAI plans to release this product in 2027 to explore new markets.