ChatGPT Toolbox AI Pulse

Huawei Develops Ascend 910D AI Chip

Huawei Technologies is preparing to test its most powerful AI processor yet, the Ascend 910D, which aims to rival Nvidia’s high-end H100 chip by integrating more silicon dies through advanced packaging techniques​.

The company expects to debut sample versions by late May 2025, marking a significant step in China’s push to strengthen its domestic semiconductor industry despite U.S. restrictions on advanced manufacturing tools​.

To maximize performance, Huawei also introduced the CloudMatrix 384 system, which links 384 Ascend chips and can under certain conditions outperform Nvidia’s leading rack solutions​. However, industry analysts note that the Ascend 910D is less power-efficient than Nvidia’s H100, highlighting ongoing challenges in matching both performance and energy metrics​.

The move is part of Beijing’s broader strategy to support domestic AI chip development and reduce reliance on foreign technology amid escalating trade tensions​.

OpenAI’s Forthcoming ‘Open’ AI Reasoning Model

OpenAI is developing a new “open” AI reasoning model under the leadership of VP of Research Aidan Clark, aiming for an early summer 2025 release that tops benchmarks of existing open-source reasoning systems​.

According to sources familiar with the proceedings, the company plans extensive red-teaming and safety evaluations, and intends to publish a detailed model card outlining internal and external benchmarking results​.

Additionally, the forthcoming model will feature the ability to “hand off” complex queries to cloud-hosted models, leveraging OpenAI’s own API infrastructure to enhance its reasoning capabilities when needed​.

This hybrid approach of on-device openness combined with cloud support could set a new standard for accessible yet powerful AI tools, stimulating innovation across research labs and enterprises alike​.

California Pushes Back on Restrictive AI Regulations

California Governor Gavin Newsom has formally warned the California Privacy Protection Agency (CPPA) that proposed regulations on automated decision-making, including AI-driven tools in hiring, healthcare, and lending, could impose up to $3.5 billion in compliance costs in the first year and risk undermining the state’s leadership in technology innovation​.

In a letter to the CPPA board, Newsom argued that such stringent rules might generate unintended legal challenges and hinder investment, aligning with concerns raised by major tech and business groups that fear stifled growth and competitiveness​.

While the CPPA is considering excluding certain generative AI applications from these requirements, board members remain divided between strong consumer protections and industry-friendly positions, with a final decision expected by November 2025​.

Newsom’s intervention underscores a broader debate over balancing AI safety with innovation, potentially influencing other states and federal policymakers as they shape the future of AI governance in the United States​.

Microsoft 365 Copilot Wave 2 Redesign

Microsoft announced a major redesign of its Microsoft 365 Copilot app as part of the Wave 2 spring release, introducing AI-powered Search that integrates data from third-party platforms such as Slack, Google Drive, and Jira to deliver more comprehensive enterprise search results​.

The updated app defaults to a chat-based interface leveraging GPT-4o for a more conversational experience, and adds a new “Create” function to generate multimedia content directly within Office documents​. Furthermore, Copilot Notebooks now allow users to organize and contextualize content and data for specific projects, turning raw information into actionable insights with minimal setup.

A new Agent Store brings first- and third-party AI agents—such as Researcher and Analyst—into the flow of work, enabling users to discover and deploy reasoning-capable tools tailored to different tasks. According to Microsoft, these enhancements aim to usher in the era of “Frontier Firms,” where human–agent collaboration drives efficiency and innovation across workflows.

AI Data Center Costs to Skyrocket by 2030

A recent TechCrunch report highlights that by June 2030, constructing a leading AI data center could require up to $200 billion in capital expenditures and consume approximately 9 gigawatts of power—equivalent to nine nuclear reactors​.

Despite annual improvements in computational performance per watt—averaging a 1.34× increase from 2019 to 2025—the projected growth in AI workloads means that efficiency gains alone will be insufficient to mitigate the escalating energy demands of large-scale training and inference operations​.

The analysis further warns that global data center energy intake could rise by 20% by 2030, potentially straining renewable energy supplies and prompting increased reliance on fossil fuels unless new sustainability measures are adopted​.

Industry stakeholders are exploring innovations such as custom AI chip architectures, advanced cooling solutions, and geographic distribution of data centers to balance performance needs with environmental and economic considerations​.

🔥 Join 10,000+ power users already mastering ChatGPT Toolbox!