Google Cloud expands AI Hypercomputer while Nvidia ships Rubin instances

Key Takeaways

• Anker unveiled the Thus chip, the world's first compute-in-memory AI audio processor, designed to process data locally in headphones to save battery and reduce heat.
• Thinking Machines Lab partnered with Google Cloud to expand AI Hypercomputer usage, utilizing NVIDIA Blackwell architecture for a 2X increase in training speeds.
• NVIDIA and Google Cloud introduced Vera Rubin-powered A5X instances offering 10x lower inference costs and the ability to scale up to 80,000 GPUs in a single cluster.
• Microsoft and NABTU are providing free AI literacy training to skilled trades professionals across North America through LinkedIn Learning.
• Ohio University approved an AI strategy emphasizing ethics and new degrees, including a Bachelor of Science in Organizational Development.
• General Motors uses AI tools to shave months off the pre-production process, accelerating vehicle design and production timelines.
• Trust issues and workforce readiness challenges are slowing enterprise AI procurement, though 66% of organizations plan to invest in orchestration solutions.
• DHS researchers demonstrated how jailbroken AI can plan terror attacks, while censored models like Anthropic's Claude and OpenAI's ChatGPT refuse harmful queries.
• Legal teams are adopting AI with strict data governance, requiring human oversight and transparent data storage to address ethical and privacy concerns.
• Connecticut passed SB 5, requiring employers to notify individuals before making employment decisions based on AI-driven tools, with obligations starting October 1, 2027.
• Clarifai deleted 3 million OKCupid photos from its training data following an FTC investigation into unauthorized data collection.

Anker unveils Thus chip for local AI in headphones

Anker announced the Thus chip, the world's first compute-in-memory AI audio chip designed for local processing. This technology stores and processes data in one location, similar to how the human brain works, unlike traditional chips that separate storage and computation. The chip integrates computing power directly into NOR flash memory cells, which are faster and require less space than NAND memory. Anker plans to launch the first Thus chip in Soundcore flagship earbuds at an event on May 21st. These earbuds will feature eight MEMS microphones and two bone conduction sensors to enable Clear Calls, providing significantly cleaner audio in noisy environments.

Anker Thus chip enables powerful on-device AI audio

Anker introduced the Thus chip, which uses a compute-in-memory architecture to bring AI directly to headphones and other small devices. This approach eliminates the need to shuttle data between memory and compute cores, saving battery life and reducing heat. The chip is capable of handling several million parameters, allowing for complex noise cancellation that traditional small networks cannot achieve. Anker notes that headphones are challenging environments due to tight space constraints and low power requirements, but the Thus chip is designed to meet these demands. The company plans to expand the chip's use to mobile accessories and IoT devices beyond just audio products.

Anker enters custom silicon race with Thus chip

Anker unveiled the Thus chip, positioning itself alongside Apple and Samsung in the competition to build custom AI processors. This chip processes AI models locally on headphones and IoT gadgets while using significantly less power than conventional processors. By embedding computation directly into memory cells, the chip solves the battery life issues that have plagued previous AI-powered audio devices. Anker CEO Steven Yang stated that this move represents a shift toward on-device intelligence in everyday accessories. The company has not yet disclosed specific product release dates but hinted that Soundcore headphones and Eufy smart home devices are likely candidates.

Anker Day reveals Thus chip for on-device AI

Anker announced its new Thus chip at the Anker Day event on May 21st to bring on-device AI capabilities to its wearables. The chip uses a compute-in-memory architecture to reduce power requirements and integrate computing power directly into NOR flash memory cells. This design requires very little space, making it ideal for small devices like earbuds that have limited room for components. The chip is inspired by the human brain, where storage and processing happen in one place rather than being kept separate. Anker plans to use the chip in upcoming headphones featuring a large neural network for noise cancellation supported by eight MEMS microphones and two bone conduction sensors.

Thinking Machines expands Google Cloud AI infrastructure

Thinking Machines Lab signed a new agreement with Google Cloud to expand its use of the AI Hypercomputer for research and training. The lab will utilize A4X Max VMs equipped with NVIDIA Blackwell architecture, including the NVIDIA GB300 NVL72 system, which is one of the first customers to use this hardware. Early testing showed a 2X increase in training and serving speeds compared to prior generation GPUs. The partnership also leverages Google Cloud services like Kubernetes Engine and Spanner to support massive-scale model orchestration and fine-tuning. Myle Ott, a founding researcher at Thinking Machines, credited the integration for enabling record speeds and reliability for their reinforcement learning workloads.

NVIDIA and Google Cloud advance agentic AI factories

NVIDIA and Google Cloud announced new infrastructure to support agentic and physical AI at Google Cloud Next in Las Vegas. The partnership introduced the Vera Rubin-powered A5X bare-metal instances, which deliver up to 10x lower inference costs and higher throughput than the previous generation. These systems use NVIDIA Rubin GPUs and next-generation Google Virgo networking to scale up to 80,000 GPUs in a single site cluster. The collaboration also includes confidential VMs with NVIDIA Blackwell GPUs and Gemini models running on NVIDIA Blackwell Ultra GPUs. Mark Lohmeyer from Google Cloud emphasized that this integrated stack allows customers to optimize performance, cost, and sustainability for demanding AI workloads.

Microsoft and NABTU offer free AI training to trades

Microsoft and the North America Building Trades Unions (NABTU) are extending free AI literacy training to skilled trades professionals across North America. The partnership delivers no-cost courses on LinkedIn Learning to instructors, apprentices, and journey-level workers, awarding industry-recognized credentials upon completion. This initiative integrates AI education into NABTU's Joint Apprenticeship Training Committee model and works with TradesFutures, a nonprofit that enrolls over 7,700 people annually. The collaboration aims to prepare millions of workers for the AI economy while preserving hands-on expertise. Microsoft Vice Chair Brad Smith stated that people building physical infrastructure deserve a share of the AI opportunity.

Ohio University board approves AI strategy and new programs

The Ohio University Board of Trustees reviewed the university's AI expansion, finances, and new academic programs during meetings in early April. The university's human-first AI strategy emphasizes ethical use, critical thinking, and adaptability, with plans to expand the AI Fellows program. The board approved several new degrees, including a Bachelor of Science in Organizational Development and a Doctor of Nursing Education, alongside modifications to 51 existing programs. Financial reports showed favorable revenue forecasts driven by higher enrollment and increased demand for campus housing. President Lori Stewart Gonzalez highlighted the university's national championship win and its recognition as a Military Friendly School.

GM uses AI to speed up car design and production

General Motors stated that incorporating artificial intelligence into its design process allows it to develop vehicles faster than ever before. The Detroit automaker deployed tools in the past year that shave months off the pre-production process, saving time and money in a competitive global market. These AI tools help streamline the journey from a vehicle's inception to its arrival at dealership showrooms. Designers at GM embrace the technology to accelerate innovation and reduce costs in their engineering workflows.

Trust issues slow down enterprise AI procurement

Many companies are facing trust challenges and workforce readiness issues as they adopt AI for procurement in 2026. A study by Oro Labs found that failures in AI rollouts often stem from unresolved questions about decision ownership and employee trust rather than the models themselves. Despite these concerns, 66% of organizations plan to invest in or upgrade orchestration solutions within the next three years. Adopters reported a median 30% improvement in efficiency and 20 days faster source-to-contract cycle times. The platform is designed to augment human roles rather than replace them, though some workforce displacement is likely as automation expands across data-heavy workflows.

Lawmakers see how jailbroken AI plans terror attacks

House lawmakers attended a closed-door briefing where DHS researchers demonstrated how jailbroken AI can be used to plan terror attacks. The Department of Homeland Security showed models with deactivated safety guardrails that provided step-by-step instructions for harmful activities like bombing or mass shootings. In contrast, censored models like Anthropic's Claude and OpenAI's ChatGPT refused to answer such queries. Representative Andrew Garbarino noted that one model answered how to kidnap a member of Congress in under three seconds. The demonstration highlighted that many AI tools are readily accessible and easy to access, increasing the risk that bad actors could weaponize them.

Legal teams adopt AI with strict data governance

Legal professionals are struggling to trust AI due to compliance risks and data governance concerns. Surveys show that 39% of legal professionals cite ethical concerns and 46% rank data privacy as their top security challenge. To build trust, legal teams must ensure full transparency into data storage, use explainable AI architectures, and maintain secure, ring-fenced environments. Human oversight remains non-negotiable because lawyers are ultimately responsible for outcomes and must validate all AI-generated outputs. Experts recommend focusing on narrow, high-value use cases like document review and contract data extraction rather than deploying AI for entire end-to-end processes.

Connecticut Senate passes strict AI employment rules

The Connecticut Senate approved a sweeping AI policy package that imposes new compliance obligations on employers using AI-driven tools. The revised bill, known as SB 5, establishes a regulatory framework for automated employment-related decision technology used in hiring and performance evaluations. Employers must notify individuals before making employment decisions if an automated system is a substantial factor, disclosing the tool's purpose and data sources. These obligations begin on October 1, 2027, and apply to systems that process personal data to generate scores or recommendations. Violations are treated as unfair or deceptive trade practices enforceable by the Connecticut Attorney General.

Clarifai deletes 3 million OKCupid photos from training

AI company Clarifai has deleted 3 million OKCupid photos it used for facial recognition training after an FTC investigation. The Delaware-based company certified the data deletion to the FTC on April 7 and confirmed it deleted any models trained on the data. The investigation began in 2019 after Clarifai asked OkCupid executives for the data in 2014, which they provided. Clarifai founder Matthew Zeiler had suggested in 2019 that people should get comfortable with tech companies collecting such data. As part of the settlement, the FTC permanently prohibited OkCupid from misrepresenting its data collection and privacy controls.

Sources

• Anker made its own chip to bring AI to all its products
• Anker's 'Thus' chip brings AI to its headphones and other products
• Anker's Custom Thus Chip Brings On-Device AI to Headphones
• Anker Launches 'Thus' Chip to Supercharge On‑Device AI Across Its Products
• Thinking Machines Expands Use of Google Cloud AI Hypercomputer
• NVIDIA and Google Cloud Collaborate to Advance Agentic and Physical AI
• Microsoft and NABTU launch free AI training for skilled trades | ETIH EdTech News
• Ohio University board reviews AI expansion, new programs and finances
• GM says AI helps create cars faster. Why designers embrace the tech
• Procurement AI Hits Trust Wall as Workforce Readiness Falls Behind
• Lawmakers are shown how ‘jailbroken’ AI can plan terror attacks
• Harnessing AI in legal teams: Practical strategies for responsible adoption
• Senate Passes Sweeping AI Mandates » CBIA
• AI company deletes the 3 million OKCupid photos it used for facial recognition training