Google launches TPU 8t as Nvidia ships Rubin GPUs

Key Takeaways

Google launches two new TPUs for AI training and inference

Google introduced two new custom chips called TPU 8t and TPU 8i at Google Cloud Next 2026. The TPU 8t is designed for massive AI model training using a 3D network topology that connects 9,600 chips. It uses a specialized SparseCore accelerator and four-bit floating point to speed up training while reducing memory needs. The TPU 8i focuses on high-speed inference with three times more static random-access memory than the previous Ironwood chip. It also features a custom network called Boardfly ICI that reduces communication delays between chips by up to 50 percent.

Google releases specialized chips to compete with Nvidia

Google is launching the eighth generation of its Tensor Processing Units to compete with Nvidia in the AI hardware market. The new TPU 8t and TPU 8i chips will be available later in 2026. Amin Vahdat, a Google senior vice president, explained that these chips handle training and inference work separately to improve efficiency. The TPU 8t offers high compute throughput for training, while the TPU 8i provides more memory bandwidth for low-latency inference tasks. Google also unveiled tools at its annual conference in Las Vegas to help companies build AI agents for automating tasks.

Google unveils eighth generation TPUs for the agentic era

Google announced its eighth generation of Tensor Processing Units, featuring the TPU 8t for training and the TPU 8i for inference. These chips are built to handle the complex demands of AI agents that reason and execute multi-step workflows. The TPU 8t powers massive model development, while the TPU 8i supports fast, collaborative AI agents with low latency. Both chips use custom hardware to deliver better performance and energy efficiency than previous generations. They are designed in partnership with Google DeepMind to adapt to evolving model architectures at scale.

Google introduces two new TPUs for the agentic era

Google is pushing the idea that the agent era requires new hardware approaches different from previous AI systems. Engineers created the TPU 8t for training and the TPU 8i for inference to make the platform faster and more efficient. The TPU 8t was designed to reduce the time needed to train frontier AI models from months to weeks. It uses innovations like linear scaling with up to a million chips in a single logical cluster. The TPU 8i is designed for inference with less waiting time and runs in larger pods of 1,152 chips compared to 256 for the last generation.

Google unveils chips for AI training and inference

Google is dedicating a separate chip to running AI models and another processor to training them as a strategy to compete with Nvidia. Both chips will become available later this year and are part of the eighth generation of its tensor processing units. Amin Vahdat stated that the rise of AI agents benefits from chips individually specialized to the needs of training and serving. The training chip enables 2.8 times the performance of the seventh-generation Ironwood TPU. Adoption is ramping up with companies like Citadel Securities and all 17 U.S. Energy Department national laboratories using the technology.

Google combines TPUs, NVIDIA Rubin, and Axion CPUs for AI

Google announced its AI Hypercomputer, which unifies performance-optimized hardware for compute, storage, networking, and software. The system houses the latest custom TPUv8 series, Axion Cloud CPUs, and NVIDIA Rubin GPUs to power the Agentic AI era. The TPU 8t training chip offers 121 Exaflops of compute capacity and scales to 9,600 chips with two petabytes of shared memory. The TPU 8i inference chip pairs 288 GB of HBM memory with 384 MB of on-chip SRAM. Both chips support Google's fourth generation liquid cooling technology to sustain high compute densities.

Google unveils 8th generation TPUs for AI Training and Inference

Google unveiled its 8th generation of TPUs, announcing two separate units for AI training and inference. The 8t Superpods now scale to 9,600 chips with 2 petabytes of shared HBM memory and 3x the computational performance per pod. With 10x faster storage access and the new TPUDirect feature, data can be pulled directly into the TPUs for higher utilization. The 8i chip pairs 288GB of HBM memory with 384MB of SRAM, which is 3x the SRAM of the last-generation Ironwood chip. New on-chip Collectives Acceleration Engines reduce on-chip latencies by 5x, delivering 80% more performance per dollar.

Mississippi universities launch AI research partnership

The University of Mississippi is partnering with regional institutions to position the Mid-South as a national hub for applied artificial intelligence research. The initiative includes the University of Mississippi, Mississippi State University, the University of Southern Mississippi, and the University of Memphis. The collaboration focuses on developing AI technologies, training the workforce, and fostering industry partnerships. The partnership is expected to attract new businesses and investments to the region, creating jobs and economic growth. It also aims to address the ethical and societal implications of AI to ensure responsible use.

Pennsylvania launches Keystone AI and Quantum Factory

Pennsylvania launched the Keystone AI + Quantum Factory, a statewide collaboration uniting seven research universities and the Pittsburgh Supercomputing Center. The initiative centers on shared computing infrastructure, workforce development, and coordinated research efforts to accelerate innovation. It supports sectors including energy, manufacturing, agriculture, life sciences, and robotics. Officials say the network is designed to attract investment, retain talent, and drive job creation across the commonwealth. The factory is built on three pillars: world-class research, workforce development, and shared infrastructure.

Israeli students face investigation over AI teacher video

The Israel Police opened an investigation after students in Netanya allegedly created a pornographic video of a middle school teacher using artificial intelligence tools. The suspects are several minors, approximately 14 years old, who used AI tools and digital editing to create the footage. One student is alleged to be behind the creation of the video, while others were partners in its distribution. The primary suspect was released under restrictive conditions, and the police continue to map the scope of exposure. The incident is regarded with severity in the education system and police.

Florida students face charges over AI teacher video

A middle school teacher in Florida is at the center of an incident where students allegedly created a pornographic video using artificial intelligence. The teacher, whose name has not been released, was targeted by several minors, approximately 14 years old, who used AI tools and digital editing to produce the footage. The students involved are facing serious legal consequences, including potential charges related to the creation and distribution of child pornography. The school district has launched an investigation and the teacher has been placed on administrative leave. The community is rallying around the teacher and condemning the actions of the students.

Dells donate $750 million for AI hospital in Texas

Billionaire Michael Dell and his wife Susan Dell gave a $750 million gift to help build the UT Dell Medical Center at the University of Texas at Austin. The planned AI-native hospital is expected to open in 2030 as part of a more than 300-acre advanced research campus. The project will integrate research, clinical care, and advanced computing to improve early disease detection and personalize treatment. The facility will be built from the ground up to incorporate AI rather than retrofitting older infrastructure. The gift ranks among the largest in the history of the university.

IBM launches autonomous security to fight AI attacks

IBM announced a new cybersecurity service called IBM Autonomous Security designed to analyze software exposures and contain threats with minimal human intervention. The service operates at machine speed to match the pace of AI-enabled adversaries who are moving faster than human teams. CrowdStrike data shows AI-enabled attack volume surged 89 percent year-over-year in 2025. IBM's service brings together interoperable digital workers that function across an organization's full security stack. The company also offers an Enterprise Cybersecurity Assessment for Frontier Model Threats to help organizations understand AI-specific exposures.

Cognizant and Google Cloud launch agentic AI for retail

Cognizant and Google Cloud launched Cognizant Agentic Retail CX, a solution powered by Google Cloud's Gemini Enterprise for Customer Experience. The offering gives retailers a purpose-built path from AI investment to contact center impact. It combines Google Cloud's Gemini Enterprise for CX with Cognizant's retail expertise to bring AI-driven capabilities into live customer environments. The solution aims for a 70-85 percent containment rate through robust AI self-service and hyper-personalized recommendations. Early client usage by Ocado Retail demonstrates the solution's impact on customer engagement.

Influencer uses AI to teach history with emotion

Jonathan Laramy, the creator of the 'Chloe vs. History' series, revealed how an AI tool captured human emotion on Fox News Live. The video discusses the use of AI in educational content and how it can replicate emotional responses in historical figures. Laramy explained the process of using the AI tool to bring life to historical characters for his audience. The segment highlighted the potential of AI in making history more engaging and relatable for viewers.

Google Maps adds new AI features for enterprise users

Google unveiled new generative AI features for its mapping and geospatial apps designed with enterprise users in mind. One feature called Maps Imagery Grounding allows users to create realistic scenes in Google Street View using prompts in the Gemini Enterprise Agent Platform. Users can also use Veo to animate these scenes in seconds. A new feature called Aerial and Satellite Insights allows users to analyze imagery stored in Google Cloud's BigQuery. Google also launched two new Earth AI Imagery models trained to identify objects like bridges, roads, and power lines.

Healthcare leaders discuss AI personalization and trust

More than 14,000 people convened in Las Vegas for the annual Adobe Summit to discuss using AI to personalize healthcare delivery. Panelists noted that healthcare has been slower to adopt technology due to regulation and siloed systems. Blue Shield of California implemented price transparency before it became a regulatory mandate to protect consumers. Lesley Spellmeyer from Lilly said AI allows healthcare companies to be more proactive in providing information to patients and providers. Experts agreed that keeping trust requires ensuring there is a human face and leading with clinicians as the best source of information.

Prince's legacy discussed a decade after his death

A look at Prince's career on the 10th anniversary of his death featured a discussion with his longtime lawyer about protecting his music icon's legacy in the age of Artificial Intelligence. NBC News' Chloe Melas reported on the conversation regarding how AI impacts the preservation of historical figures' work. The segment explored the challenges and opportunities of using AI to manage and present Prince's extensive musical catalog. The discussion highlighted the need for careful management of digital assets in the modern era.

Sources

• Two new TPUs to power the next wave of AI training and inference at Google
• Google releases new chips for AI training and inference, competing with Nvidia
• Our eighth generation TPUs: two chips for the agentic era
• Google unveils two new TPUs designed for the "agentic era"
• Google unveils chips for AI training and inference in latest shot at Nvidia
• Google Bets On The Agentic AI Era With Its AI Hypercomputer, Merges 8th-Gen TPUs, NVIDIA Rubin, & Axion CPUs Together
• Google unveils 8th generation TPUs for AI Training and Inference
• Partnership aims to expand artificial intelligence research and workforce, industry collaboration
• Pennsylvania Universities and Commonwealth Leaders Launch Keystone AI + Quantum Factory
• Students produce pornographic video of teacher using artificial intelligence
• Students produce pornographic video of teacher using artificial intelligence
• Michael and Susan Dell surpass 1 billion in donations backing AI driven hospital project
• When AI Attacks AI: Can Autonomous Security Outrun the Agentic AI Threat?
• Cognizant and Google Cloud Bring Agentic AI to Retail's Most Critical Customer Moments with Gemini Enterprise
• Influencer uses AI to teach people about history
• Google Maps is about to get a big dose of AI
• Using AI To Personalize Healthcare–Without Losing Patient Trust
• The future of Prince’s legacy in the age of AI a decade after his death