AI News Report — 2026-07-03 (Fri)

[Overview] The biggest focus has been on reports that OpenAI is in negotiations to grant a 5% stake to the U.S. administration, and a new phase of capital ties between the government and major AI companies has emerged. Anthropic is accelerating negotiations with Samsung to manufacture custom chips, expanding its Seattle base, and responding to infrastructure and geopolitics. Microsoft launches new $2.5 billion AI implementation division, intensifying competition for enterprise adoption.

[Anthropic / Claude trends] It has been reported that the company is negotiating with Samsung to manufacture custom AI chips, which appears to be a move to break away from dependence on NVIDIA. The company has leased 113,000 square feet of office space in Seattle's South Lake Union, allowing it to expand its workforce. Meanwhile, WSJ reports that emails showing how relations with Pentagon deteriorated have been made public, and friction in the government procurement field has come to the fore. There are reviews that report Claude Sonnet 5's superiority over ChatGPT when comparing practical assistants, and the product evaluation is generally good.

[Trends in OpenAI] Multiple media outlets (Axios, CNN, Forbes) have reported that negotiations are underway to grant a 5% stake to the Trump administration, drawing attention to the unprecedented structure of an equity relationship with the government. Sam Altman has proposed the idea of a ``US-Led International Forum'' to strengthen ties with the administration in conjunction with a US-led international framework for AI policy. There are reports that the neutrality of OpenAI/Anthropic models on Cursor is being questioned after the acquisition of SpaceX, and tensions over platform neutrality continue.

[Google AI trends] Nano Banana 2 Lite and Gemini Omni Flash (video generation) have been added to expand the image and video generation lineup. There have been reports that the data collected by Gemini is being used extensively for purposes other than advertising, reigniting privacy concerns. Some analysis suggests that Alphabet's adoption of the Dow average and Gemini's lack of AI capacity are behind the expansion of its partnership with Meta.

[Trends of other AI companies] (Meta, Microsoft, xAI, DeepSeek, etc.) Microsoft has established a new $2.5 billion, 6,000-person AI engineering implementation department as a "Frontier Company," and is also reviewing its enterprise Copilot. While Meta is making a full-scale entry into the cloud AI business and is taking on Amazon and Google, Zuckerberg himself has admitted that ``AI agent development is slower than expected,'' revealing the gap between expectations and reality. There have been reports that a new model from China is one-sixth the cost of Anthropic/OpenAI in terms of performance, and the expression "Mini DeepSeek Moment" is being used. xAI is in a phase where its model neutrality via SpaceX's Cursor is being tested, and reports have surfaced that Musk is developing a device to replace mobile phones (though he denies it).

[AI celebrities' statements and trends] Sam Altman (OpenAI CEO) — Says that AI will “reshape the material conditions of human life,” and proposes a U.S.-led international AI framework and a 5% equity grant to the administration. Elon Musk (xAI Founder / Tesla CEO) — While he has said that AI can solve the debt crisis by improving productivity, the Treasury Department's calculations are skeptical. He denies reports of SpaceX's mobile replacement device. Dario Amodei (Anthropic CEO) — Emails showing deteriorating relations with Pentagon were published in a WSJ report, making visible friction in government procurement. Daniela Amodei (Anthropic President) — In an interview with Le Monde, she said, ``We need to accept both the positive and negative aspects of AI.'' Mark Zuckerberg (Meta CEO) - In an exclusive interview with Reuters, he admitted that ``AI agent development is slower than expected,'' but said that even after mass layoffs, ``I don't think AI will replace human workers.'' Demis Hassabis (Google DeepMind CEO) — Named UKtech50 2026 Best of the Year. A column in the Times newspaper said, ``Ignore the prophets of AI pessimism.''

[Breakthroughs and innovations in AI technology] Reports have been confirmed in multiple media that a new AI model from China costs one-sixth that of Anthropic and OpenAI, and the competitive landscape in terms of performance and cost ratio is changing. Multimodal expansion of image and video generation continues, including Google's Nano Banana 2 Lite and Gemini Omni Flash. There are also reports that Etched has come out of stealth with a valuation of $5 billion to compete with NVIDIA, and competition in the inference-specific chip market is widening.

[AI regulation/policy/social impact] The FTC has proposed a policy to prohibit "deceptive steering" in AI products and services, and is working to strengthen regulations from the perspective of consumer protection. Draft high-risk classification guidelines for the EU AI Act have been published, requiring responses before a July 23 deadline. Reports of OpenAI's stock giving to the government have sparked a new regulatory and governance debate over the relationship between AI companies and the government.

[Integrated analysis: What is happening now] The biggest structural change is the report that OpenAI will give a 5% stake to the Trump administration, meaning that AI companies that were previously ``subject to regulation'' are moving to the stage of ``being incorporated into the government's shareholder structure.'' At the same time, Anthropic's relationship with the Pentagon has deteriorated, and the relationship between AI companies and the US government is not monolithic, and there are differences in temperature between companies. On the infrastructure side, companies are shifting the focus of their investments from "model development" to "implementation and supply networks," with Anthropic negotiating with Samsung chips, Meta entering the cloud business, and Microsoft establishing a $2.5 billion implementation division. Behind this is the ``delay in agent development,'' which Zuckerberg himself admitted, and it appears that each company's response to the plateau in performance improvement has been dispersed into infrastructure, chips, and implementation support. On the other hand, the fact that the costs of Chinese models are becoming lower, threatening the cost advantage of major Western companies, is a theme that should be continuously monitored as an axis of geopolitical competition.

[Future points of interest] 1. Whether there is a formal agreement between OpenAI and the government regarding negotiations for granting 5% equity, and the actual status of deregulation and procurement preferential treatment associated with it. 2. Future developments in the Anthropic-Pentagon relationship and the materialization of chip manufacturing negotiations with Samsung. 3. The impact of the cost advantage of the "Mini DeepSeek" model from China on the pricing strategies of major Western companies.

[Overview] Regarding Anthropic, the observation of collaboration with Samsung on custom AI chips and Alibaba's policy to ban Claude Code are running parallel, highlighting the tug-of-war over reliability. OpenAI quietly introduces GPT-5.6 while strengthening compliance with European regulations. At Meta xAI, management has repeatedly made statements admitting delays in the progress of AI agents.

[Anthropic / Claude trends] Tech Times reported that Anthropic is reportedly in talks with South Korea's Samsung for custom chips, and is considering a 2nm process. On the other hand, Alibaba is said to have a policy of prohibiting the use of Claude Code internally, citing backdoor risks as the reason. Details of Fable 5's jailbreak countermeasure framework have also been made public, and the company continues to promote its safety.

[Trends in OpenAI] OpenAI has announced its support for building a "trusted AI ecosystem" in Europe, and has put regulatory compliance at the forefront. IBM reports that although GPT-5.6 has been launched, the pace of deployment is cautious. A lawsuit has surfaced over the partnership with Oracle, alleging that investors were "misled" about OpenAI-related transactions.

[Google AI trends] Sundar Pichai promotes the use of Gemini with Google Workspace ads. Gemini personal image generation is now free in the US, and Nano Banana 2 Lite and Gemini Omni Flash have been rolled out to developers. Google Drive has also added an AI document search function for mobile, and the product's functionality has been significantly expanded.

[Trends of other AI companies] Meta admits that its agent-based AI did not progress as expected, and the Watermelon model reportedly uses 10 times more computing resources than its predecessor, the Avocado. He is also facing an IP lawsuit from a textbook author. xAI's exhibition of Grok-equipped smartphones became a hot topic, but Musk denied reports of AI device development at SpaceX. Tesla limits employees' AI usage budget to $200 per week, excluding Grok. Microsoft plans to invest $2.5 billion in its consulting business after a prototype of its new Copilot-based OS ``Project Aion'' was leaked. It is reported that DeepSeek continues to move closer to U.S. and Chinese AI companies with low costs.

[Remarks and trends of AI celebrities] Sam Altman (OpenAI CEO) reportedly called for the establishment of an international AI standards forum led by the United States, and also made a remark about predicting job losses. Dario Amodei (Anthropic CEO) criticized Pentagon's attitude towards introducing AI, calling it a "uniformly defensive attitude", and the tense exchange was revealed in a published email. Elon Musk (xAI Founder) denied reports that SpaceX was developing an AI device, but said that he would use engineers from Starship and Starlink to develop Grok. Mark Zuckerberg (Meta CEO) admitted in an internal town hall that progress on agent-based AI is slower than expected. Sundar Pichai (Google CEO) reportedly sold $9.8 million in GOOGL stock. It was reported that Yann LeCun (Meta Chief AI Scientist) is working on developing more flexible AI models at AMI Labs.

[Breakthroughs and innovations in AI technology] The fact that Meta's new model Watermelon uses 10 times more computing resources than the previous generation symbolizes the trend of increasing computing costs. While investment in hardware is accelerating, with Taiwan being pointed out as the center of the AI revolution and AI semiconductor stocks continuing to perform well, reports of clear breakthroughs in models are limited.

[AI Regulation/Policy/Social Impact] FTC begins soliciting public comments on proposed policy statement on accuracy and ideological manipulation of AI output. The New York State Bar Association has published an article analyzing the necessity of Assembly Bill 8546. In Idaho, a new education law includes an AI framework. Discussions regarding the role of AI are becoming more active in the legal world.

[Integrated analysis: What is happening now] In terms of power, Anthropic is conspicuously moving toward building its own infrastructure by collaborating with Samsung on chips, while Alibaba's ban policy symbolizes a geopolitical divide in trust. OpenAI is shifting its focus from introducing models to regulatory compliance and creating international standards, and Altman's comments can be read as an attempt to secure leadership in governance. At Meta and xAI, the management has continued to acknowledge the delay in agent AI, and it is possible that the company has entered a phase of adjustment from "anticipating expectations." The fact that the increase in computational resource input (Watermelon) and the rise in semiconductor stock prices are running parallel to each other suggests that model performance improvement is returning to dependence on scale. On the regulatory front, the FTC's increased commitment to accuracy and fairness represents increasing social control pressure on AI output in the United States.

[Future points of interest] - Is there an official announcement regarding custom chip talks between Anthropic and Samsung? - Impact of Meta xAI agent AI delays on product roadmap - Regulatory pressure from the FTC's proposed AI accuracy policy on the industry's output verification process