Beijing considers restricting overseas access to China’s most powerful AI models ↗
Chinese authorities are discussing controls that could prevent overseas users from accessing the country’s most advanced AI models. Alibaba, ByteDance and Z.ai reportedly attended meetings led by China’s Ministry of Commerce.
Officials also considered tougher penalties for leaking proprietary AI technology and restrictions on foreign investment in domestic AI startups. Chinese open models have become popular because they are capable and relatively inexpensive, so closing the gates could leave plenty of international developers suddenly checking their infrastructure budgets.
DeepSeek is reportedly developing its own AI chip ↗
DeepSeek is working on an inference chip designed to run its AI models without relying as heavily on Nvidia or Huawei hardware. The company has reportedly increased recruitment of chip-design engineers and held discussions with foundries, memory suppliers and other technology partners.
Inference chips handle the everyday work of generating responses after a model has been trained. Building one successfully would give DeepSeek greater control over costs and capacity, although designing the chip may prove considerably easier than finding somewhere capable of manufacturing it at scale.
Meta launches Muse Image across Instagram and WhatsApp ↗
Meta is rolling out Muse Image, the first image-generation model produced by Meta Superintelligence Labs. The model can understand complicated prompts, work from uploaded photographs and let users edit images through sketches and annotations.
Muse Image will support more than 30 new effects for Instagram Stories and allow users to generate images while chatting with Meta AI on WhatsApp. Facebook and Messenger integration is planned, while an early version of Meta’s Muse Video generator is also being previewed. Apparently, merely posting photographs was becoming far too straightforward.
Ukraine prioritises AI models it can operate independently ↗
Ukraine plans to favour AI models that can be installed on its own servers rather than systems that remain under the control of external providers. The policy is intended to ensure that government, business and military services cannot suddenly lose access because a technology company or foreign government changes the rules.
Ukraine currently uses Gemini within its Diia government application but removes personal information before sending requests because it does not control the underlying model. It is also developing a locally operated system based on Google’s open Gemma model, scheduled for release in the autumn.
Bank of England warns AI could threaten financial stability ↗
The Bank of England says AI is becoming a growing risk to the financial system. Its concerns include heavily leveraged investors betting on AI companies, technology businesses borrowing enormous sums to fund infrastructure and increasingly capable cyberattacks against banks.
A sharp reassessment of AI’s profitability could trigger falling technology shares, with concentrated investments and borrowed money potentially making the decline worse. Britain’s banking system remains resilient, according to the Bank, although it noted that existing regulatory frameworks were not designed for autonomous AI agents.
Amazon seeks another $25 billion to fund its AI ambitions ↗
Amazon is looking to raise $25 billion through a bond sale as it continues spending heavily on AI infrastructure. Investor demand reportedly reached $62 billion, demonstrating that the market has not completely lost its appetite for financing extremely expensive data centres.
Amazon, Alphabet, Microsoft and Meta are collectively expected to spend more than $700 billion on AI this year. Technology giants traditionally funded expansion from their enormous cash reserves, but the AI infrastructure race is now sufficiently costly that even Big Tech has started reaching for the credit card.
SpaceXAI and Cursor prepare their first jointly developed model ↗
SpaceXAI and Cursor are reportedly preparing to release their first jointly developed AI model. The launch was delayed briefly while engineers worked on improving efficiency, with the model expected to compete in some areas with advanced systems from Anthropic and OpenAI.
SpaceX agreed in June to acquire Anysphere, the company behind Cursor, in a $60 billion all-stock transaction. Combining Cursor’s popular coding tools with SpaceXAI’s computing infrastructure could create a considerably stronger competitor in enterprise AI — because rockets, satellites and social media apparently were not keeping Elon Musk busy enough.
FAQ
Why is China considering restrictions on overseas access to advanced AI models?
Chinese authorities are reportedly discussing controls that could limit foreign access to the country’s most powerful AI models. The proposals may also include tougher penalties for leaking proprietary technology and restrictions on foreign investment in Chinese AI startups. Such measures could affect international developers that rely on capable, lower-cost Chinese open models.
What would a DeepSeek AI chip be used for?
DeepSeek’s proposed chip would focus on inference, the process of generating responses after an AI model has been trained. Developing its own hardware could reduce the company’s dependence on Nvidia or Huawei products and provide greater control over capacity and operating costs. Manufacturing the chip at sufficient scale may remain a significant challenge.
How does Meta Muse Image work on Instagram and WhatsApp?
Muse Image can generate pictures from detailed prompts, use uploaded photographs and support edits through sketches or annotations. On Instagram, it will power more than 30 new effects for Stories. WhatsApp users will be able to create images while chatting with Meta AI, with Facebook and Messenger support planned later.
Why does Ukraine prefer AI models it can run on its own servers?
Ukraine wants to reduce the risk of losing access when an external provider or foreign government changes its policies. Locally operated AI models also provide greater control over sensitive government, business and military services. Ukraine currently removes personal information before using Gemini and is developing a domestic system based on Google’s open Gemma model.
How could AI investment create risks for financial stability?
The Bank of England has highlighted risks from leveraged investors, large infrastructure debts and more capable cyberattacks against financial institutions. A sudden decline in expectations for AI profitability could cause technology shares to fall sharply. Concentrated investments and borrowed money could amplify those losses, while autonomous AI agents may create challenges not fully covered by existing regulations.
Why are major technology companies borrowing money for AI infrastructure?
Building data centres and computing capacity requires exceptionally large amounts of capital, even for companies with substantial cash reserves. Amazon is reportedly seeking $25 billion through a bond sale, while demand from investors reached $62 billion. Borrowing allows technology companies to continue expanding AI infrastructure without funding every project directly from existing cash.