{
    "success": true,
    "data": {
        "id": 1694484,
        "msgid": "google-launches-specialised-eighth-generation-tpu-to-welcome-the-agentic-ai-era-1776869457",
        "date": "2026-04-22 21:17:59",
        "title": "Google launches specialised eighth-generation TPU to welcome the Agentic AI era",
        "author": "",
        "source": "ANTARA_ID",
        "tags": "",
        "topic": "Technology",
        "summary": "Google Cloud has unveiled the eighth-generation Tensor Processing Units (TPUs), TPU 8t and TPU 8i, specifically engineered to accelerate AI workloads in the emerging Agentic AI era. The TPU 8t is optimised for training complex models on a massive scale, while the TPU 8i focuses on rapid inference to enhance user experiences through AI agents. This dual launch, available commercially by year's end, aims to redefine AI capabilities for developers and global enterprises, with applications spanning scientific discovery to healthcare transformations.",
        "content": "<p>Las Vegas (ANTARA) - Google Cloud has officially introduced the\neighth generation of Tensor Processing Units (TPU), namely TPU 8t and\nTPU 8i, designed to accelerate artificial intelligence (AI) workloads in\nthe Agentic era.<\/p>\n<p>SVP and Chief Technologist for AI and Infrastructure at Google Cloud,\nAmin Vahdat, stated at the event The Future of AI Infrastructure, Google\nCloud Next \u201926, in Las Vegas, United States, on Tuesday, that this year,\ntwo specialised chips are being launched simultaneously to welcome the\nAgentic AI era.<\/p>\n<p>\u201cFor the first time, two custom-made TPUs, 8t and 8i, have been\nentirely designed from scratch to meet current needs. TPU 8t is for\ntraining. This is the main engine. This is the largest scale. TPU 8i is\ntruly built for inference,\u201d said Vahdat.<\/p>\n<p>Vahdat explained that for years, TPUs have supported leading\nfoundational models, including Gemini. The combination of these\neighth-generation TPUs will deliver scale, efficiency, and capabilities\nencompassing training workloads, serving, and agentic needs.<\/p>\n<p>Therefore, 2026 marks a pivotal moment for the technology giant to\nintroduce two highly powerful specialised chips simultaneously, the\nresult of the past two years of development.<\/p>\n<p>Vahdat elaborated that TPU 8t is optimised for training and capable\nof running even the most complex models within a single massive\nintegrated memory pool.<\/p>\n<p>Meanwhile, TPU 8i is specifically designed to enable AI agents to\ncomplete processes extremely quickly, providing a satisfying user\nexperience.<\/p>\n<p>Vahdat emphasised that both chips are purpose-built from scratch for\ntraining and inference efforts, not mere derivatives of one another.\nSpecifications, capabilities, connectivity\u2014all are designed differently\ndue to their specialisation for individual needs.<\/p>\n<p>\u201cIf you need main power for training, you have 8t. If you need\nlightning-fast inference, you have 8i,\u201d Vahdat clarified.<\/p>\n<p>Through the commercial availability of these latest chips by the end\nof the year, Google Cloud aims to redefine the boundaries of AI\ncapabilities for developers and global companies.<\/p>\n<p>The utilisation of this technology, including in Gemini Enterprise\nservices, is expected to accelerate scientific discoveries to\ntransformations in the healthcare sector on an international scale.<\/p>",
        "url": "https:\/\/jawawa.id\/newsitem\/google-launches-specialised-eighth-generation-tpu-to-welcome-the-agentic-ai-era-1776869457",
        "image": ""
    },
    "sponsor": "Okusi Associates",
    "sponsor_url": "https:\/\/okusiassociates.com"
}