October 29, 2025

news  
Google Gemini Competitor, Bezos Earth Fund $2M & $30M AI Grants

The artificial intelligence landscape is rapidly evolving with new hardware, software, and applications emerging across various sectors. Qualcomm has introduced its AI200 and AI250 accelerator cards for data centers, designed to enhance generative AI performance with substantial memory capacity and liquid cooling, slated for release in 2026 and 2027. Microchip is also boosting AI hardware capabilities with its new Switchtec Gen 6 PCIe switches, which double the bandwidth of PCIe 5.0 to accelerate data movement in AI clusters. In the realm of AI models, Chinese startup MiniMax has released its M2 model, which reportedly surpasses Google DeepMind's Gemini 2.5 Pro in intelligence and reasoning tasks, offering a more cost-effective API. Meanwhile, Elon Musk's xAI launched Grokipedia, an AI-powered encyclopedia intended as a Wikipedia alternative, though it experienced an initial crash and appears to host content adapted from Wikipedia. Musk is also funding AI research to analyze ancient Roman history, aiming to uncover new insights. Beyond technology development, AI is finding applications in diverse fields: the Bezos Earth Fund is investing $2 million as part of a larger $30 million initiative to use AI for developing better plant-based food alternatives, partnering with Food System Innovations and Stanford University. On the regulatory front, new AI rules are prompting companies to overhaul their data security plans, emphasizing risk management, data governance, and transparency. However, the proliferation of AI also presents challenges, as demonstrated by the spread of AI-generated fake videos during Hurricane Melissa, which complicated official safety communications. Robotics startup Mbodi, founded by former Google engineers, is developing a system using AI agents to train robots more efficiently, with plans for wider deployment in 2026.

Key Takeaways

  • Qualcomm has launched new AI accelerator cards, the AI200 and AI250, for data centers, featuring high memory capacity and liquid cooling, with availability expected in 2026 and 2027.
  • Microchip's new Switchtec Gen 6 PCIe switches offer double the bandwidth of PCIe 5.0, aiming to improve data transfer speeds for AI hardware.
  • Chinese startup MiniMax released its M2 AI model, which claims to outperform Google DeepMind's Gemini 2.5 Pro in intelligence and reasoning tasks and is available via API at a lower cost.
  • Elon Musk's xAI launched Grokipedia, an AI encyclopedia intended to rival Wikipedia, but the platform crashed shortly after its release and appears to use Wikipedia content.
  • Elon Musk is also funding AI research projects focused on analyzing ancient Roman history to uncover new historical information.
  • The Bezos Earth Fund has awarded a $2 million grant to develop an open-source AI model for sustainable food products, as part of a $30 million AI for Climate and Nature Grand Challenge.
  • New AI regulations are requiring companies to update their data security strategies, focusing on enhanced risk management and data governance.
  • AI-generated fake videos circulated during Hurricane Melissa, highlighting the challenge of misinformation during emergencies.
  • Robotics startup Mbodi, founded by former Google engineers, is using AI agents to accelerate robot training and plans wider deployment in 2026.

Qualcomm launches new AI chips for data centers

Qualcomm announced its new AI inference solutions for data centers, the Qualcomm AI200 and AI250 accelerator cards, and racks. These chips use Qualcomm's NPU technology to provide high performance for generative AI tasks. The AI200 offers a low total cost of ownership for large language models, supporting 768 GB of memory per card. The AI250 will feature a new memory architecture for improved efficiency and lower power use. Both solutions will be available in 2026 and 2027 and include direct liquid cooling and secure computing features.

Qualcomm's AI250 and AI200 chips boost data center AI performance

Qualcomm has launched its AI250 and AI200 chips designed for AI data center workloads, offering significant memory capacity. The AI200 supports up to 768GB of LPDDR memory per card and features direct liquid cooling. The AI250 also provides 768GB memory per card and introduces a new memory architecture using near-memory computing for over 10x higher effective memory bandwidth and lower power consumption. Durga Malladi stated these solutions redefine rack-scale AI inference, offering high performance with a lower total cost of ownership.

Bezos Earth Fund backs AI for better vegan burgers

The Bezos Earth Fund has awarded a $2 million grant to Food System Innovations (FSI) and Stanford University to develop an open-source AI model for sustainable food products. This project aims to accelerate the creation of better-tasting plant-based and blended meat alternatives. The AI model will use FSI's Nectar database, which contains sensory data on plant-based products, to predict taste and optimize ingredient formulations. This initiative is part of the Bezos Earth Fund's $30 million AI for Climate and Nature Grand Challenge, focusing on using AI for environmental solutions.

Fake AI videos spread as Hurricane Melissa nears Jamaica

As Hurricane Melissa, a Category 5 storm, approached Jamaica, AI-generated fake videos began circulating on social media. These misleading videos distracted from important safety information provided by officials. Jamaica's information minister, Senator Dana Morris Dixon, stated that ministers were holding press briefings to share accurate details about the storm. The spread of AI-generated misinformation highlights a growing challenge in disseminating critical information during emergencies.

Microchip's new PCIe switches speed up AI hardware

Microchip has released its Switchtec Gen 6 PCIe switches, built on 3nm process technology, to improve data movement for AI hardware. These switches offer PCIe 6.0 bandwidth, running at 64 gigatransfers per second per lane, which is double that of PCIe 5.0. This advancement addresses the bottleneck in AI clusters where fast data exchange is crucial. The Gen 6 switches also feature enhanced security with hardware root-of-trust and secure boot, along with power efficiency improvements. Microchip provides diagnostic tools and evaluation platforms to help developers utilize the new technology.

Musk's Grokipedia launches, crashes, and hosts Wikipedia content

Elon Musk's AI company xAI launched Grokipedia, an AI-generated encyclopedia intended to rival Wikipedia. The platform, powered by xAI's Grok chatbot, went live but crashed within hours of its release. At launch, Grokipedia hosted nearly 900,000 articles, with many users noting that a significant portion appeared to be adapted from Wikipedia. Musk had previously criticized Wikipedia for 'editorial bias' and aimed for Grokipedia to be a more 'truthful and independent alternative.'

Chinese startup MiniMax releases record-breaking AI model M2

Chinese AI startup MiniMax has launched its M2 model, which achieved a top score on Artificial Analysis's overall intelligence index for open models. With 200 billion parameters and a Mixture-of-Experts architecture, M2 activates only 10 billion parameters per use, enhancing efficiency. The model surpasses Google DeepMind's Gemini 2.5 Pro in performance and excels in reasoning for agentic and coding tasks. M2 is available via API at a lower cost than US competitors, though it consumed a large number of tokens for benchmark evaluations.

Elon Musk funds AI research to rewrite ancient Roman history

Elon Musk is funding new archaeological projects that will use artificial intelligence to explore ancient Rome. The goal of this research is to fill in gaps in historical knowledge and potentially 'rewrite history.' This initiative marks a significant investment in using AI for archaeological discovery and understanding past civilizations.

New AI rules change data security plans

Emerging regulations for artificial intelligence are forcing companies to update their data security strategies. These new rules require enhanced risk management throughout the AI lifecycle, from data collection to deployment. Organizations must also improve data governance to ensure AI training data is accurate and unbiased, while preventing unauthorized access. Increased transparency and accountability are mandated, requiring detailed records of AI operations. Companies are integrating AI-specific security measures, like advanced encryption and anomaly detection, to address unique AI threats.

Mbodi uses AI agents to train robots faster

Robotics startup Mbodi will demonstrate its system for training robots using AI agents at TechCrunch Disrupt 2025. Their cloud-to-edge system uses multiple AI agents that communicate to help robots learn tasks more quickly. Users prompt the software with natural language, and Mbodi's agents break down the request into subtasks for efficient robot training. Founded by former Google engineers, Mbodi aims to solve the challenge of training robots for the physical world's infinite possibilities. The company plans to deploy its software more widely in 2026.

Sources

AI chips data centers Qualcomm AI inference generative AI large language models NPU technology liquid cooling secure computing AI data center workloads memory architecture near-memory computing rack-scale AI inference total cost of ownership Bezos Earth Fund AI for sustainable food plant-based meat alternatives AI model sensory data ingredient formulation AI for climate environmental solutions fake AI videos misinformation Hurricane Melissa social media emergency information Microchip PCIe switches AI hardware data movement PCIe 6.0 AI clusters data exchange hardware root-of-trust secure boot power efficiency Grokipedia xAI AI encyclopedia Grok chatbot Wikipedia editorial bias MiniMax AI model M2 open models Mixture-of-Experts Gemini 2.5 Pro agentic tasks coding tasks API AI research archaeology ancient Rome historical knowledge AI rules data security risk management AI lifecycle data governance AI training data transparency accountability AI security measures AI threats robotics AI agents robot training natural language prompt subtasks physical world