AI Daily News Weekly Report — 2026-W28

AI industry meta-analysis report (past 7 days/July 2026)

Summary: The most important developments this period

The biggest incident was the U.S. government's sudden change in restrictions on AI exports to China. Export restrictions that were temporarily imposed on Anthropic/OpenAI's latest models (Fable 5, GPT-5.6, etc.) have been lifted, and Anthropic is reported to have cleared an important hurdle toward IPO. Meanwhile, OpenAI, led by Sam Altman, is negotiating to offer a 5% stake to the US government, which is linked to speculation that SoftBank's IPO plans will be delayed. The second focus is Alibaba's ban on internal use of Anthropic products (including Claude Code), with Anthropic expressing suspicions of a "distilled attack," and the US-China AI friction is expanding not only to export restrictions but also to enterprise usage. Third, an unverified benchmark that claimed Meta's "Watermelon AI" was on par with GPT-5.5 in performance caused a stir, and Zuckerberg himself acknowledged delays in AI agent development. Fourth, Tesla's unusual move to impose a usage cap of $200 per week on all AI tools except Grok has become a hot topic. Fifth, DeepSeek was reported to have completed a $7 billion round of funding at a valuation of more than $5 billion, confirming once again that low-cost AI models originating from China are making inroads into the stronghold of top American companies.

Anthropic / Claude trends

Anthropic had the highest amount of coverage at 65 cases, with the focus being on IPO-related activities. Export restrictions on the Fable 5 model have been lifted, it has been reported that it has surpassed OpenAI with a valuation of $96.5 billion, and talks with Samsung to manufacture custom AI chips are progressing at the same time. Meanwhile, it faced a new $75 million lawsuit for copyright infringement, and Alibaba banned its internal use due to security concerns. On the business front, it has been reported that the company has started offering an "AI workbench" for scientists and is moving into original drug discovery, making it clear that the company is aiming for vertical integration beyond the framework of providing models. In terms of human resources and regulations, he is increasing his involvement in policy, such as proposing mandatory AI safety testing for Congress.

Trends in OpenAI

The topic of OpenAI centers around building relationships with the US government. SamAltman proposed a 5% stake to the government, underscoring his closeness to the Trump administration. GPT-5.6 is limited to government-approved partners and is heavily influenced by export regulations. While IPOs are expected to be valued at around $1 trillion, concerns about delays in projects led by SoftBank have also emerged, and market expectations and practical uncertainties coexist. Regarding UK investment, there are problems with building external trust, such as failure to conduct on-site inspections. Due to the intensifying price competition with Anthropic, it was also reported that subscription prices may be lowered.

Trends in Google AI/DeepMind

Google focuses on product development news, and the expansion of Gemini integration is noticeable. Integration of AI ordering function into Google Maps, addition of Computer Use function for AI agents to Gemini 3.5 Flash, and deployment of in-vehicle Gemini in the automobile industry (Renault, GM Cadillac/Corvette) are progressing. On the other hand, it has been reported that there has been a supply/demand crunch between Google and Meta over the provision of Gemini to Meta, and behind the scenes of the AI development competition, the bargaining over the allocation of compute resources has become apparent. On the marketing front, an ad praising AI by the founding fathers of the Revolutionary War caused controversy, and the company's poor branding strategy was exposed. DeepMind has announced a research partnership with A24 and the launch of Nano Banana 2 Lite.

Trends of other major companies

Meta claims that its "Watermelon AI" is on par with GPT-5.5, but the benchmark details remain undisclosed and unverified, raising questions about its reliability. Zuckerberg himself has admitted that the development of AI agents has been delayed, and is shifting his focus to the cloud business (competing with AWS and Google Cloud). xAI/Tesla introduced a $200 weekly cap on AI tool use, with only Grok being an exception, and the asymmetry of resource allocation under the Musk regime became a hot topic. While Microsoft continues to expand its investments, including equipping Copilot with GPT-5 and Claude Sonnet 5 and establishing a $2.5 billion new AI business division, Frontier Co., corporate opposition to Copilot's implementation measures (requests to disable AI during Teams meetings) has also surfaced. DeepSeek is said to have completed funding of over $7 billion and reached a valuation of $50 billion, and continues to approach the strongholds of Anthropic and OpenAI in terms of cost efficiency.

Statements and trends of AI celebrities

Sam Altman's main topics were government stock offer negotiations and G7 attendance (along with Dario Amodei and Demis Hassabis). Dario Amodei publicly commented on Alibaba's alleged distillation attack, making his wary attitude toward China clear. While Elon Musk denied rumors of a SpaceX AI device (smartphone-type prototype), the departure of all 11 founding members of xAI led to Yann LeCun calling xAI a failure. LeCun himself has reiterated his theory that current AI "lacks understanding of the real world." Mark Zuckerberg admits that he is behind in the development of AI agents, but his management stance is divided, with the image of him making a business call on a jet ski while wearing Meta Ray band glasses becoming a hot topic. Jensen Huang continues to take a bullish stance on the humanoid robot market (estimated at $40 trillion).

Technological breakthrough

In terms of new models, it was reported that Google's Gemini 3.5 Flash (with Computer Use function), Meta's Watermelon AI, and DeepSeek V4 will be introduced with a peak hour billing system. In terms of benchmarks, competition among Chinese models is intensifying, as seen in the comparison article "GLM-5.2 vs DeepSeek V4 vs Kimi K2.6." ByteDance has announced the discovery of a new scaling law, shattering the debate over the sustainability of the AI boom. Hermes MoA is reported to have shown a higher score than individual models with a Mixture of Agents configuration that combines GPT, Claude, and DeepSeek, and is attracting attention as a move that suggests a shift from dependence on a single frontier model to a model routing ensemble architecture. Together AI has announced that the open source inference market has surpassed the $1 billion mark after raising $800 million in funding, and the numbers are beginning to confirm the structural change from the closed model.

Regulation/Policy/Geopolitics

While the U.S.'s AI export restrictions to China have shifted toward relaxation, and the focus has been on lifting restrictions on Anthropic and OpenAI models, there are also analyzes (Foreign Policy in Focus) that discuss why the relaxation cannot be maintained, and questions remain about the consistency of the policy. AI-related legislation has been enacted in several states in the country. Illinois has enacted an AI anti-bullying and deepfakes law in schools, Florida has enacted an AI data center law, and New York has enacted an AI-generated advertising regulation law, and regulations continue to be taken at the state level. Thailand has begun drafting a comprehensive AI law. Internationally, it has been pointed out that OpenAI's negotiations to accept government funding may set a new precedent regarding the relationship between AI companies and states.

Changes in power balance

Due to the relaxation of export regulations, there is an increasing competition between Anthropic and OpenAI in terms of distance from the government. Anthropic has taken the lead in lifting regulations and preparing for an IPO, but OpenAI is seeking to deepen its ties with the government through direct provision of shares, and the differences in their approaches may affect future regulatory operations and valuations. Meta/xAI's inferiority in model development has become apparent, and confusion continues as Meta moves into cloud business and xAI leaves internally. This structure is an extension of the ``narrowing down of frontier model development'' that has been going on since the first half of this year, and is progressing in tandem with the rise of cost-efficient and open source groups such as DeepSeek and Together AI. Tensions over compute allocation between Google and Meta show that the AI boom is shifting its focus from competition for model performance to competition for infrastructure, power and chip supplies.

Key points for the next period

First, we should closely monitor the progress of OpenAI's government investment negotiations and the impact on the IPO timing and valuation. Second, it is necessary to follow how the alleged distillation attack between Alibaba and Anthropic will develop and whether it will be reflected in the implementation of export restrictions to China. Thirdly, the focus will be on whether Meta Watermelon AI's official benchmark will be released or not, and whether the actual state of xAI's organizational management after all founding members have left will be revealed.