Chinese AI companies are rapidly accelerating their new model releases, aiming to compete directly with US rivals like OpenAI and Google. Firms such as Alibaba and Moonshot AI are focusing on making their AI affordable and open-source to attract users, particularly in emerging markets. Moonshot AI recently launched Kimi K2.5, which now leads all open-source AI models on the Artificial Analysis Intelligence Index v4.0 with a score of 47 points, positioning it just behind top US models from OpenAI, Anthropic, and Google. This model is also remarkably cost-effective, costing only $371 to run all evaluations.
Moonshot AI's Kimi K2.5, valued at US$4.3 billion after a US$500 million funding round, is open-source, multimodal, and features an "agent swarm" utilizing up to 100 subagents for complex tasks. This advancement, alongside Alibaba's updated Qwen model and Tencent's plan to open-source its Hunyuan model, demonstrates China's progress in narrowing the AI development gap with the US, despite ongoing chip restrictions. Meanwhile, Microsoft has unveiled its Maia 200 AI chip, designed for efficient AI inference in its cloud services. This move aims to enhance performance and energy efficiency while reducing Microsoft's reliance on external chip suppliers like Nvidia.
NVIDIA is also enhancing resource management with its Run:ai v2.24, introducing a time-based fairshare feature for Kubernetes clusters. This ensures more equitable GPU access among teams by tracking past usage, preventing smaller, frequent jobs from monopolizing resources. In the realm of security, a Zscaler report highlights a 91% surge in enterprise AI usage in 2025, but alarmingly, 100% of observed AI systems had critical vulnerabilities. The report also noted 18,033 TB of company data transferred to AI apps like ChatGPT and Grammarly, resulting in 410 million data privacy violations. This underscores the growing security risks, as exemplified by a CISA acting director accidentally uploading "for official use only" documents to a public ChatGPT.
The broader AI sector is entering a new phase, shifting focus to robust infrastructure and clear policy. This includes prioritizing AI chip production, data center expansion, and electricity supply, alongside removing innovation barriers. Astronomers are already leveraging AI's power, with an AI-assisted technique identifying over 700 rare cosmic objects, including 27 potential gravitational lenses, in Hubble data in just two and a half years. Separately, the European Space Agency's AnomalyMatch AI scanned nearly 100 million Hubble images in two and a half days, discovering over 800 previously unknown "astrophysical anomalies."
On the financial front, Asian stock markets saw gains driven by strong performances in technology and AI-related companies like Samsung and SK Hynix. Investors are closely monitoring the sales backlogs of cloud giants such as Microsoft, Amazon, and Google, which reflect future sales from multiyear deals with AI companies like OpenAI. Oracle's backlog, for instance, surged by $460 billion, largely due to a $300 billion commitment from OpenAI. However, concerns linger about the full materialization of these large, single-customer deals. Cisco CEO Chuck Robbins warns that while the AI boom will be bigger than the internet, it could also lead to a market bubble, similar to the dotcom bust, urging workers to embrace AI's transformative impact.
Key Takeaways
- Moonshot AI's Kimi K2.5 leads open-source AI models on the Artificial Analysis Intelligence Index v4.0 with 47 points, costing only $371 for evaluations, and the company is valued at US$4.3 billion.
- Chinese AI companies like Alibaba, Moonshot AI, and Tencent are rapidly releasing affordable, open-source AI models to compete with US rivals such as OpenAI and Google.
- Microsoft launched its Maia 200 AI chip for efficient AI inference in cloud services, aiming to reduce reliance on external suppliers like Nvidia.
- NVIDIA Run:ai v2.24 introduced time-based fairshare for Kubernetes clusters, improving equitable GPU resource allocation among teams.
- Enterprise AI usage increased by 91% in 2025, but a Zscaler report found 100% of observed AI systems had critical vulnerabilities, with 18,033 TB of data transferred to AI apps leading to 410 million privacy violations.
- An acting director at CISA accidentally uploaded "for official use only" government documents to a public ChatGPT, triggering an internal investigation.
- AI-assisted techniques discovered over 800 new cosmic anomalies, including gravitational lenses, in Hubble Space Telescope data, significantly accelerating astronomical discoveries.
- The AI sector is shifting focus to building robust infrastructure, including increased AI chip production and data centers, and establishing clear policy frameworks.
- Cloud giants like Microsoft, Google, and Oracle are seeing large sales backlogs; Oracle's surged by $460 billion, largely due to a $300 billion commitment from OpenAI.
- Cisco CEO Chuck Robbins warns the AI boom, while bigger than the internet, could create a market bubble and advises workers to embrace AI's transformative impact on jobs.
Chinese AI companies speed up new model releases
Chinese AI companies like Alibaba and Moonshot AI are quickly releasing new AI models to compete with US rivals such as OpenAI and Google. These firms focus on making their AI affordable and open-source to gain users, especially in emerging markets. Moonshot AI recently launched Kimi K2.5, which claims to outperform leading US models in some areas. Alibaba updated its Qwen model, and Tencent announced it will open-source its Hunyuan model. This strategy aims to build large application ecosystems around Chinese AI technologies.
Moonshot AI's Kimi K2.5 leads open-source AI models
Moonshot AI's Kimi K2.5 model now leads all open-source AI models on the Artificial Analysis Intelligence Index v4.0. It scored 47 points, placing it just behind top US models from OpenAI, Anthropic, and Google. Kimi K2.5 also offers great value, costing only $371 to run all evaluations, much less than its competitors. This shows Chinese open-source AI is advancing quickly, offering powerful options at a lower price. The gap between Chinese and US open-source AI capabilities is widening.
Moonshot AI's Kimi K2.5 closes AI gap with US
Moonshot AI released its new Kimi K2.5 model, which analysts say narrows the AI development gap with the US. This model is open-source, multimodal, and much cheaper to run than leading US models. Kimi K2.5 also features an "agent swarm" that uses up to 100 subagents for complex tasks. Despite US chip restrictions, Chinese companies like Moonshot AI use efficient architectures to keep pace. Moonshot AI was recently valued at US$4.3 billion after a US$500 million funding round.
AI finds hundreds of cosmic oddities in Hubble data
Astronomers used an AI-assisted technique to find many rare astronomical objects in Hubble Space Telescope data. The AI identified 27 potential gravitational lenses, with 17 already confirmed. This new method found over 700 such objects in just two and a half years, which is more than in the past 50 years. These discoveries, mostly gravitational lenses, help scientists better understand dark matter and dark energy. Dr. Sherry H. Wu and Pablo Gómez led this powerful research, published in The Astrophysical Journal.
AI discovers 800 new cosmic anomalies in Hubble archives
Scientists at the European Space Agency used an AI model called AnomalyMatch to search Hubble Space Telescope archives. The AI scanned nearly 100 million images in just two and a half days. It found over 800 previously unknown "astrophysical anomalies." These included merging galaxies, gravitational lenses, and jellyfish galaxies, with some objects defying classification. This project, led by David O’Ryan and Pablo Gómez, shows how AI can greatly help astronomers find new discoveries in vast amounts of space data.
NVIDIA Run:ai improves GPU sharing with new fairshare
NVIDIA Run:ai v2.24 introduced a new feature called time-based fairshare for Kubernetes clusters. This feature helps share GPU resources more fairly among different teams. It solves the problem where small, frequent jobs could take all available extra resources, leaving larger jobs waiting. The system now tracks past GPU usage to give teams a more balanced share over time. This ensures everyone gets fair access to GPU power for their projects, while guaranteed quotas and priority tiers still work as before.
Microsoft unveils Maia 200 AI chip for cloud services
Microsoft launched its new AI chip, Maia 200, designed for efficient AI inference tasks. This chip will help Microsoft run trained AI models in its cloud services more economically and with better performance. Maia 200 is more energy-efficient than older solutions and will first be used in Microsoft data centers in the US. By developing its own hardware, Microsoft aims to reduce its reliance on external chip suppliers like Nvidia. This move will lead to faster and more affordable AI services for Azure customers and users of Microsoft's AI products.
Cisco CEO warns AI boom will create winners and losers
Chuck Robbins, CEO of Cisco Systems, believes the AI boom will be bigger than the internet but also warns of a market bubble. He expects some companies will fail, similar to the dotcom bust of the late 90s. Robbins advises workers to embrace AI, as it will change or eliminate some jobs, especially in customer service. He also highlighted AI's risks, like better cyber attacks and scams, but noted Cisco uses quantum technology to boost security. Robbins thinks the UK has a good chance to become an AI superpower.
CISA acting director accidentally leaks data to ChatGPT
Madhu Gottumukkala, the acting director of CISA, accidentally uploaded "for official use only" government documents to a public ChatGPT last summer. This happened shortly after he joined the agency and received special permission to use the chatbot, which most DHS staff cannot access. The incident triggered internal cybersecurity warnings and led to a DHS investigation. While CISA stated his use was temporary and aligned with AI orders, critics question Gottumukkala's leadership and his reported attempt to remove CIO Robert Costello.
Asia stocks rise on strong AI tech company gains
Most Asian stock markets saw gains on Wednesday, driven by strong performance in technology and AI-related companies. This rise happened before major US tech company earnings reports and a Federal Reserve interest rate decision. South Korea's KOSPI jumped, with Samsung and SK Hynix seeing significant increases. Hong Kong's Hang Seng also surged, led by its tech sector. Investors are watching for signs of continued AI-related growth from companies like Microsoft and Apple, which are reporting earnings soon.
AI use grows 91% but security risks rise
A new report from Zscaler shows that enterprise AI usage jumped 91% in 2025 across over 3400 applications. However, this rapid growth brings serious security threats, as 100% of observed AI systems had critical vulnerabilities. Many companies also do not track their AI models, creating a wider attack surface. The report found that 90% of systems were compromised in under 90 minutes. Additionally, a huge amount of company data, 18,033 TB, was transferred to AI apps like Grammarly and ChatGPT, leading to 410 million data privacy violations.
AI enters new phase with focus on infrastructure and policy
Artificial intelligence is moving into a new phase, shifting from simple experiments to widespread use in 2026. This change means that building strong infrastructure and creating clear rules are now top priorities. Policymakers need to focus on increasing AI chip production, building more data centers, and ensuring enough electricity. They also aim to remove barriers to AI innovation and expand its use in government. While AI offers many opportunities, it also brings challenges like job changes and security risks that require careful management.
Cloud giants face investor worry over AI sales backlog
Investors are closely watching the sales backlog of major cloud companies like Microsoft, Amazon, and Google. This metric, called remaining performance obligations, shows future sales from multiyear deals, especially with AI companies like OpenAI. Since ChatGPT launched in late 2022, Microsoft and Google have added more to their backlogs than Amazon. Oracle's backlog also surged by $460 billion, largely due to a $300 billion commitment from OpenAI. However, investors worry if these promised sales will fully materialize, especially with such large deals tied to single customers.
Sources
- One year after DeepSeek, Chinese AI firms from Alibaba to Moonshot race to release new models
- Kimi K2 Becomes Top Open Model On Artificial Analysis Index
- Moonshot’s newest release narrows US-China AI model development gap: analysts
- Artificial intelligence reveals mysterious ‘astrophysical anomalies’
- Scientists let AI loose on Hubble’s archives
- Ensuring Balanced GPU Allocation in Kubernetes Clusters with Time-Based Fairshare
- Maia 200, Microsoft pushes its own AI hardware into the next generation
- AI boom will produce victors and carnage, tech boss warns
- US cyber defense chief accidentally uploaded secret government info to ChatGPT
- Asia stocks jump on AI-led gains ahead of ’Mag 7’ earnings; Fed on tap By Investing.com
- AI Security Threats Loom as Enterprise Usage Jumps 91%
- The Next Phase of AI: Technology, Infrastructure, and Policy in 2025–2026
- Microsoft, Amazon Give Bubble-Wary Investors Another Metric to Worry About
Comments
Please log in to post a comment.