On May 29, 2025, DeepSeek released an update to its R1 reasoning model, intensifying competition with U.S. rivals. The updated R1 model, weighing in at 685 billion parameters, was released on Hugging Face. DeepSeek also gained attention on May 28 after its chatbot app topped the Apple App Store charts.

Nvidia reported a $4.5 billion charge in Q1 due to licensing requirements affecting H20 AI chip sales in China and anticipates an $8 billion revenue hit in Q2 due to these restrictions. Nvidia's Q2 revenue forecast missed market estimates due to U.S. export restrictions impacting AI chip sales to China.

NVIDIA is developing a new budget AI chip based on its Blackwell architecture for the Chinese market, expected to be priced between $6,500 and $8,000, as reported on May 29, 2025.

On May 28, 2025, London-based AI lab Odyssey launched a research preview of an AI model transforming video into interactive worlds, initially focusing on film and game production. The interactive video responds to inputs in real-time, generating realistic-looking video frames every 40 milliseconds, and is powered by Nvidia H100 GPUs at a cost of $1 to $2 per user-hour.

On May 28, 2025, xAI partnered with Telegram to distribute its Grok chatbot to Telegram users. xAI will pay Telegram $300 million in cash and stock and give Telegram half of any subscription sales made through the messaging app, aiming to tap into Telegram's user base.

On May 29, 2025, Nvidia CEO Jensen Huang argued that keeping Chinese open-source models running on Nvidia chips provides U.S. firms with valuable insight on where the global AI industry is headed. He also expressed concerns about U.S.-China policy, stating that Nvidia risks being cut off from China's AI developer base and that China's chip industry is closing in on the United States' dominance.

As of May 28, 2025, Pony.ai unveiled an agreement with Dubai's Roads and Transport Authority in the United Arab Emirates to trial its robotaxis in the city this year, with full driverless operations planned for 2026. WeRide announced its expansion into Saudi Arabia, with commercial services expected to start in late 2025, and the launch of fully driverless robotaxi trials in Abu Dhabi this month, while Tesla plans a trial in Austin, Texas, by the end of June.

On May 28, 2025, Netflix co-founder Reed Hastings joined Anthropic’s board of directors, expressing his belief in Anthropic’s approach to AI development.

On 2025-05-28, Rillet raised $25M from Sequoia Capital to automate general ledger systems using AI. The funding comes 10 months after a $13.5 million seed and pre-seed round.

On 2025-05-28, it was announced that Context, a startup building an AI-powered office suite, raised $11 million in a seed round led by Lux Capital. Founder Joseph Semrai started working on Context in 2024.

DeepSeek, like other Chinese AI companies, has been affected by U.S. export bans on hardware, forcing it to use less-powerful Nvidia H800 chips. Its AI models are also subject to benchmarking by China’s internet regulator to ensure responses align with “core socialist values,” and the company faces scrutiny from U.S. entities due to data security and propaganda concerns.

NVIDIA and AMD are expected to begin selling new AI chips in China starting in July 2025 to comply with US restrictions. The news was published on May 28, 2025.

On May 28, 2025, AMD announced the acquisition of Enosemi, a startup focused on silicon photonics product development, for an undisclosed amount, to enhance its co-packaged optics offerings for AI systems. AMD also acquired server maker ZT Systems for $4.9 billion in 2025 to expand its AI chip and hardware portfolio and better compete with Nvidia.

Huawei's Supernode 384 architecture, featuring 384 Ascend AI processors and 48 terabytes of high-bandwidth memory generating 300 petaflops of computational power, was presented at the Kunpeng Ascend Developer Conference in Shenzhen on May 28, 2025. The Supernode 384 was designed to address the bottleneck in traditional server architectures for training as the scale of parallel processing grows.

On May 28, 2025, Nvidia has signed new deals in the Middle East, including a 10-square-mile data center site in the United Arab Emirates, and similar deals in Saudi Arabia and Taiwan, to compensate for potential losses in Chinese revenue.

On May 28, 2025, Microsoft announced the beta version rollout of its Copilot for Gaming experience in the Xbox app on iOS and Android, accessible to beta testers. The early version allows the assistant to answer questions about games, provide links to information, and answer questions based on account details, play history, and achievements.

TechCrunch Sessions: AI will be held in Berkeley, California on June 5, 2025, offering a discount on tickets, as published on May 28, 2025.

On May 28, 2025, Google Photos debuted a redesigned editor with new AI tools, including Reimagine and Auto Frame, previously only available on Pixel devices. The new editor gives users helpful suggestions and puts all its editing tools in one place, including the "AI Enhance" option.

On 2025-05-28, MIT Technology Review created the AI Hype Index to summarize the latest buzz about AI, including Meta’s AI friends and how generative AI is helping a Neuralink patient communicate faster. The article mentions that AI could lead to better battery life for iPhones via Google DeepMind's new model, and AI could combine with brain implants to help people communicate when they have lost the ability to speak.

On May 28, 2025, it was published that Sarah Alnegheimish is developing Orion, an open-source machine learning framework and time series library, in Kalyan Veeramachaneni’s Data-to-AI group at MIT's Laboratory for Information and Decision Systems (LIDS). Orion detects anomalies without supervision in large-scale industrial and operational settings and has reached over 120,000 downloads.

On May 28, 2025, it was published that Atomic Canyon received 20,000 GPU hours of compute from Oak Ridge National Laboratory to build new AI models. Atomic Canyon is using sentence embedding and retrieval-augmented generation (RAG) to make nuclear power plant documents searchable.