{"id":1249,"date":"2026-11-12T02:30:45","date_gmt":"2026-11-11T17:30:45","guid":{"rendered":"https:\/\/itexplore.org\/jp\/columns\/anthropic-claude-fable-mythos-5-2-google-tpu-nominal-growth\/"},"modified":"2026-11-12T02:30:45","modified_gmt":"2026-11-11T17:30:45","slug":"anthropic-claude-fable-mythos-5-2-google-tpu-nominal-growth","status":"publish","type":"post","link":"https:\/\/itexplore.org\/jp\/columns\/anthropic-claude-fable-mythos-5-2-google-tpu-nominal-growth\/","title":{"rendered":"Anthropic\u306eClaude Fable\/Mythos 5.2\u3068Google\u306eTPU\u3001Nominal\u306e\u6025\u6210\u9577"},"content":{"rendered":"<p>\u672c\u65e5\u306e\u6ce8\u76eeAI\u30fb\u30c6\u30c3\u30af\u30cb\u30e5\u30fc\u30b9\u3092\u3001\u5c02\u9580\u7684\u306a\u5206\u6790\u3068\u5171\u306b\u304a\u5c4a\u3051\u3057\u307e\u3059\u3002<\/p>\n<div class=\"wp-block-vk-blocks-alert vk_alert alert alert-warning has-alert-icon\">\n<div class=\"vk_alert_icon\">\n<div class=\"vk_alert_icon_icon\"><i class=\"fa-solid fa-triangle-exclamation\" aria-hidden=\"true\"><\/i><\/div>\n<div class=\"vk_alert_icon_text\"><span>Warning<\/span><\/div>\n<\/div>\n<div class=\"vk_alert_content\">\n<p>\u3053\u306e\u8a18\u4e8b\u306fAI\u306b\u3088\u3063\u3066\u81ea\u52d5\u751f\u6210\u30fb\u5206\u6790\u3055\u308c\u305f\u3082\u306e\u3067\u3059\u3002AI\u306e\u6027\u8cea\u4e0a\u3001\u4e8b\u5b9f\u8aa4\u8a8d\u304c\u542b\u307e\u308c\u308b\u53ef\u80fd\u6027\u304c\u3042\u308b\u305f\u3081\u3001\u91cd\u8981\u306a\u5224\u65ad\u3092\u4e0b\u3059\u969b\u306f\u5fc5\u305a\u30ea\u30f3\u30af\u5148\u306e\u4e00\u6b21\u30bd\u30fc\u30b9\u3092\u3054\u78ba\u8a8d\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/div>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">Claude Fable 5.2\u3068Mythos 5.2\u3092\u767a\u8868<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> Introducing Claude Fable 5.2 and Claude Mythos 5.2<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p><strong>Anthropic<\/strong>\u306f\u3001\u6700\u3082\u56f0\u96e3\u306a\u6280\u8853\u7684\u8ab2\u984c\u306b\u53d6\u308a\u7d44\u3080\u305f\u3081\u306b\u8a2d\u8a08\u3055\u308c\u305f\u30e2\u30c7\u30eb\u3067\u3042\u308b<strong>Claude Fable 5.2<\/strong>\u3068<strong>Claude Mythos 5.2<\/strong>\u3092\u30a2\u30c3\u30d7\u30b0\u30ec\u30fc\u30c9\u3057\u307e\u3057\u305f\u3002\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306f\u3001\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3001\u63a8\u8ad6\u3001\u30d3\u30b8\u30e7\u30f3\u306a\u3069\u306e\u5206\u91ce\u3067\u9855\u8457\u306a\u6539\u5584\u3092\u793a\u3057\u3066\u304a\u308a\u3001\u30ea\u30ea\u30fc\u30b9\u524d\u306e\u30c6\u30b9\u30c8\u3067\u306f\u3001\u30a8\u30f3\u30b8\u30cb\u30a2\u304c\u5b9f\u73fe\u4e0d\u53ef\u80fd\u3060\u3063\u305f\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u79fb\u884c\u3092\u5b8c\u4e86\u3055\u305b\u305f\u308a\u30013\u3064\u306e\u96fb\u5b50\u30c7\u30d0\u30a4\u30b9\u3092\u30bc\u30ed\u304b\u3089\u8a2d\u8a08\u3057\u305f\u308a\u3001\u30d6\u30e9\u30c3\u30af\u30db\u30fc\u30eb\u306e\u7269\u7406\u5b66\u306b\u304a\u3051\u308b\u672a\u89e3\u6c7a\u554f\u984c\u3092\u89e3\u6c7a\u3057\u305f\u308a\u3001\u30d5\u30a7\u30eb\u30de\u30fc\u306e\u6700\u7d42\u5b9a\u7406\u306e\u8a3c\u660e\u309230\u6642\u9593\u3067\u5f62\u5f0f\u5316\u3057\u305f\u308a\u3059\u308b\u306e\u306b\u5f79\u7acb\u3061\u307e\u3057\u305f\u3002<\/p>\n<p><strong>Claude Fable 5.2<\/strong>\u306f\u3001\u60aa\u7528\u3092\u9632\u3050\u305f\u3081\u306e\u5f37\u529b\u306a\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u304a\u3088\u3073\u751f\u7269\u5b66\u7684\u306a\u30bb\u30fc\u30d5\u30ac\u30fc\u30c9\u3092\u5099\u3048\u3066\u304a\u308a\u3001\u3059\u3079\u3066\u306e\u30e6\u30fc\u30b6\u30fc\u304c\u5229\u7528\u3067\u304d\u307e\u3059\u3002\u4e00\u65b9\u3001<strong>Claude Mythos 5.2<\/strong>\u306f\u3001\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u3084\u751f\u7269\u533b\u5b66\u7814\u7a76\u306b\u304a\u3051\u308b\u6709\u76ca\u306a\u30a2\u30d7\u30ea\u30b1\u30fc\u30b7\u30e7\u30f3\u3092\u8ffd\u6c42\u3059\u308b\u7d44\u7e54\u5411\u3051\u306b\u3001\u62e1\u5f35\u3055\u308c\u305f\u30e9\u30a4\u30d5\u30b5\u30a4\u30a8\u30f3\u30b9\u304a\u3088\u3073\u30b5\u30a4\u30d0\u30fc\u691c\u8a3c\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u901a\u3058\u3066\u30a2\u30af\u30bb\u30b9\u304c\u63d0\u4f9b\u3055\u308c\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306f\u3001\u4ee5\u524d\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3068\u6bd4\u8f03\u3057\u3066\u6700\u592730%\u901f\u3044\u51fa\u529b\u751f\u6210\u901f\u5ea6\u3068\u4f4e\u4fa1\u683c\u3092\u5b9f\u73fe\u3057\u3001\u30aa\u30d7\u30b7\u30e7\u30f3\u306e\u9ad8\u901f\u30e2\u30fc\u30c9\u3067\u306f\u30c8\u30fc\u30af\u30f3\u751f\u6210\u3092\u6700\u59272.5\u500d\u306b\u52a0\u901f\u3067\u304d\u307e\u3059\u3002<\/p>\n<p>\u5b89\u5168\u6027\u3068\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u306b\u95a2\u3057\u3066\u306f\u3001\u30b5\u30f3\u30c9\u30dc\u30c3\u30af\u30b9\u30a8\u30b9\u30b1\u30fc\u30d7\u306a\u3069\u306e\u7279\u5b9a\u306e\u61f8\u5ff5\u3055\u308c\u308b\u884c\u52d5\u306b\u5bfe\u3059\u308b\u65b0\u3057\u3044\u8a55\u4fa1\u3092\u542b\u3080\u5e83\u7bc4\u306a\u30ea\u30ea\u30fc\u30b9\u524d\u30c6\u30b9\u30c8\u304c\u5b9f\u65bd\u3055\u308c\u307e\u3057\u305f\u3002\u30e2\u30c7\u30eb\u306f\u307b\u3068\u3093\u3069\u306e\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u6e2c\u5b9a\u3067\u4ee5\u524d\u306e\u30e2\u30c7\u30eb\u3068\u540c\u7b49\u304b\u305d\u308c\u4ee5\u4e0a\u306e\u6027\u80fd\u3092\u793a\u3057\u3001\u9ad8\u30ea\u30b9\u30af\u306e\u8aa4\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u306e\u30ea\u30b9\u30af\u306f\u4f4e\u3044\u3068\u8a55\u4fa1\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u30c6\u30b9\u30c8\u3067\u306f\u3001<strong>Fable 5.2<\/strong>\u3068<strong>Mythos 5.2<\/strong>\u304c\u3001<strong>Opus 5.5<\/strong>\u3001<strong>GPT-6 Astra<\/strong>\u3001<strong>Gemini 4 Carbon<\/strong>\u3068\u3044\u3063\u305f\u4ed6\u306e\u4e3b\u8981\u30e2\u30c7\u30eb\u3092\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3001\u79d1\u5b66\u7814\u7a76\u3001\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u3001\u30bf\u30f3\u30d1\u30af\u8cea\u8a2d\u8a08\u3001\u30ed\u30dc\u30c6\u30a3\u30af\u30b9\u306a\u3069\u306e\u307b\u307c\u3059\u3079\u3066\u306e\u80fd\u529b\u3067\u4e0a\u56de\u3063\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u80fd\u529b\u306f\u975e\u5e38\u306b\u9ad8\u304f\u3001<strong>Anthropic<\/strong>\u306e\u30b3\u30a2\u30b3\u30fc\u30c9\u30d9\u30fc\u30b9\u3078\u306e\u5909\u66f4\u306e90%\u4ee5\u4e0a\u304c<strong>Claude<\/strong>\u306b\u3088\u3063\u3066\u66f8\u304b\u308c\u3066\u3044\u307e\u3059\u3002\u307e\u305f\u3001<strong>Fable 5.2<\/strong>\u306f\u30011,000\u4ee5\u4e0a\u306e\u30b5\u30d6\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3092\u8abf\u6574\u3057\u30668\u65e5\u9593\u3067\u52d5\u4f5c\u3059\u308b\u30aa\u30da\u30ec\u30fc\u30c6\u30a3\u30f3\u30b0\u30b7\u30b9\u30c6\u30e0\u3092\u8a2d\u8a08\u3057\u3001Python\u304b\u3089Rust\u3078\u306e\u91cd\u8981\u306a\u30d7\u30ed\u30c0\u30af\u30b7\u30e7\u30f3\u30b5\u30fc\u30d3\u30b9\u306e\u79fb\u884c\u3092\u5b8c\u4e86\u3055\u305b\u307e\u3057\u305f\u3002\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u5206\u91ce\u3067\u306f\u3001<strong>Claude Mythos 5.2<\/strong>\u306f\u4ee5\u524d\u306e\u30e2\u30c7\u30eb\u3088\u308a\u3082\u591a\u304f\u306e\u8106\u5f31\u6027\u3092\u691c\u51fa\u3057\u3001\u305d\u308c\u3089\u3092\u518d\u73fe\u304a\u3088\u3073\u4fee\u6b63\u3059\u308b\u80fd\u529b\u306b\u512a\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u79d1\u5b66\u304a\u3088\u3073\u5de5\u5b66\u5206\u91ce\u3067\u306f\u3001<strong>Fable 5.2<\/strong>\u306f\u7406\u8ad6\u7269\u7406\u5b66\u306b\u304a\u3051\u308b2\u3064\u306e\u672a\u89e3\u6c7a\u554f\u984c\u3092\u89e3\u6c7a\u3057\u3001<strong>Mythos 5.2<\/strong>\u306f\u8907\u6570\u306e\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30e2\u30c7\u30eb\u3092\u57fa\u76e4\u3068\u3057\u3066\u3001AAV\u30ab\u30d7\u30b7\u30c9\u306a\u3069\u306e\u8907\u96d1\u306a\u30bf\u30f3\u30d1\u30af\u8cea\u69cb\u9020\u3092\u6b63\u78ba\u306b\u4e88\u6e2c\u3059\u308b\u65b0\u3057\u3044\u65b9\u6cd5\u3092\u767a\u898b\u3057\u307e\u3057\u305f\u3002\u6570\u5b66\u306e\u5206\u91ce\u3067\u306f\u30011,000\u4f53\u306e<strong>Fable 5.2<\/strong>\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30c1\u30fc\u30e0\u304c\u3001\u6700\u5c0f\u9650\u306e\u4eba\u9593\u4ecb\u5165\u3067\u30d5\u30a7\u30eb\u30de\u30fc\u306e\u6700\u7d42\u5b9a\u7406\u306e\u8a3c\u660e\u3092\u308f\u305a\u304b30\u6642\u9593\u3067\u5b8c\u5168\u306b\u5f62\u5f0f\u5316\u3057\u307e\u3057\u305f\u3002<\/p>\n<p>\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u304a\u3088\u3073\u30ed\u30dc\u30c6\u30a3\u30af\u30b9\u3067\u306f\u3001<strong>Fable 5.2<\/strong>\u306f\u56de\u8def\u57fa\u677f\u30013D\u30d7\u30ea\u30f3\u30c8\u3055\u308c\u305f\u30a8\u30f3\u30af\u30ed\u30fc\u30b8\u30e3\u3001\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u3092\u5099\u3048\u305f3\u3064\u306e\u52d5\u4f5c\u3059\u308b\u96fb\u5b50\u30c7\u30d0\u30a4\u30b9\u3092\u8a2d\u8a08\u30fb\u88fd\u9020\u3057\u3001\u7269\u7406\u30ed\u30dc\u30c3\u30c8\u306e\u5236\u5fa1\u306b\u304a\u3044\u3066\u5927\u5e45\u306a\u9032\u6b69\u3092\u793a\u3057\u307e\u3057\u305f\u3002\u7814\u7a76\u304a\u3088\u3073\u5206\u6790\u30bf\u30b9\u30af\u3067\u306f\u3001<strong>Fable 5.2<\/strong>\u306f\u3088\u308a\u66d6\u6627\u3067\u30c7\u30fc\u30bf\u91cf\u306e\u591a\u3044\u30bf\u30b9\u30af\u3092\u51e6\u7406\u3067\u304d\u3001\u62bd\u8c61\u7684\u306a\u63a8\u8ad6\u3068\u8a2d\u8a08\u3092\u5fc5\u8981\u3068\u3059\u308b\u30bf\u30b9\u30af\u3067\u3088\u308a\u512a\u308c\u305f\u5224\u65ad\u529b\u3092\u767a\u63ee\u3057\u307e\u3059\u3002<\/p>\n<p>\u4e0d\u6b63\u306a\u84b8\u7559\u306b\u5bfe\u3059\u308b\u5f37\u529b\u306a\u4fdd\u8b77\u6a5f\u80fd\u3082\u5c0e\u5165\u3055\u308c\u3066\u304a\u308a\u3001\u84b8\u7559\u8005\u306b\u3088\u3063\u3066\u53ce\u96c6\u3055\u308c\u308b\u30c7\u30fc\u30bf\u309290%\u304b\u308999%\u524a\u6e1b\u3057\u307e\u3057\u305f\u3002\u307e\u305f\u3001<strong>Enterprise Frontier Safeguards<\/strong>\u306b\u3088\u308a\u3001\u9867\u5ba2\u306e\u30a4\u30f3\u30d5\u30e9\u30b9\u30c8\u30e9\u30af\u30c1\u30e3\u306b\u4fdd\u5b58\u3055\u308c\u305f\u30d7\u30ed\u30f3\u30d7\u30c8\u3068\u5fdc\u7b54\u306b\u5bfe\u3057\u3066\u81ea\u52d5\u5316\u3055\u308c\u305f\u5b89\u5168\u76e3\u8996\u304c\u5b9f\u884c\u3055\u308c\u3001\u30bc\u30ed\u30c7\u30fc\u30bf\u4fdd\u6301\u3068\u4e92\u63db\u6027\u304c\u3042\u308a\u307e\u3059\u3002<strong>Claude Fable 5.2<\/strong>\u306f\u3001<strong>Amazon Web Services<\/strong>\u3001<strong>Google Cloud<\/strong>\u3001<strong>Microsoft Azure<\/strong>\u3092\u542b\u3080\u3059\u3079\u3066\u306e\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u3067\u5229\u7528\u53ef\u80fd\u3067\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/www.anthropic.com\/claude-fable-and-mythos-5-2\" target=\"_blank\" rel=\"noopener\">Anthropic \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> Anthropic's Claude Fable 5.2 and Mythos 5.2 represent a significant leap in AI capabilities across coding, scientific research, cybersecurity, and robotics, emphasizing safety, efficiency, and broad accessibility for complex agentic workloads.<\/li>\n<li><strong>\u8457\u8005:<\/strong> Editorial Staff<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p><strong>Anthropic<\/strong> has upgraded <strong>Claude Fable 5.2<\/strong> and <strong>Claude Mythos 5.2<\/strong>, models designed to tackle the most daunting technical challenges. These models show significant improvements in areas like coding, reasoning, and vision. During prerelease testing, they assisted engineers in completing otherwise-infeasible software migrations, designing three working electronic devices from scratch, solving two open problems in black hole physics, and formalizing the proof of <strong>Fermat's Last Theorem<\/strong> in just 30 hours.<\/p>\n<p><strong>Claude Fable 5.2<\/strong> is available to all users with strong cybersecurity and biology safeguards to prevent misuse. <strong>Claude Mythos 5.2<\/strong>, on the other hand, is accessible to organizations pursuing beneficial applications in cybersecurity or biomedical research through Anthropic's expanded Life Sciences and Cyber Verification Programs. These new models generate outputs up to 30% faster than their predecessors and are available at a lower price, with an optional fast mode that speeds up token generation by up to 2.5 times.<\/p>\n<p>Extensive prerelease safety testing, including novel evaluations for concerning behaviors like sandbox escapes, was conducted to ensure safety and alignment. The models performed as well as, or better than, recent models on most alignment measures, with high-stakes risks from misalignment assessed as low. Benchmarks demonstrate that <strong>Fable 5.2<\/strong> and <strong>Mythos 5.2<\/strong> surpass <strong>Opus 5.5<\/strong>, <strong>GPT-6 Astra<\/strong>, and <strong>Gemini 4 Carbon<\/strong> across nearly all evaluated capabilities, including agentic coding, scientific research, cybersecurity, protein design, and robotics.<\/p>\n<p>Their coding prowess is remarkable, with over 90% of changes merged into Anthropic's core codebase now written by <strong>Claude<\/strong>. <strong>Fable 5.2<\/strong> also designed a working operating system by coordinating over 1,000 subagents and migrated a critical production service from Python to Rust. In cybersecurity, <strong>Claude Mythos 5.2<\/strong> is Anthropic's most capable model, catching more vulnerabilities and being better at reproducing and fixing them than previous models.<\/p>\n<p>In science and engineering, <strong>Fable 5.2<\/strong> solved two open problems in theoretical physics, while <strong>Mythos 5.2<\/strong> found a novel way to accurately predict whole capsid structures for gene therapies by building on open-source protein-folding models. In mathematics, a team of 1,000 <strong>Fable 5.2<\/strong> agents fully formalized the proof of <strong>Fermat's Last Theorem<\/strong> in just 30 hours with minimal human intervention.<\/p>\n<p>For hardware and robotics, <strong>Fable 5.2<\/strong> designed and produced three working electronic devices and showed substantial advances in controlling physical robots, even for tasks it was not explicitly trained on. In research and analysis, early testers found that <strong>Fable 5.2<\/strong> can handle more ambiguous, data-heavy tasks and exhibits better judgment, particularly in abstract reasoning and design.<\/p>\n<p>Strong protections against illicit distillation have been implemented, reducing data harvested by distillers by 90% to 99%. Additionally, <strong>Enterprise Frontier Safeguards<\/strong> enable automated safety monitoring on prompts and responses stored within a customer's infrastructure, compatible with zero data retention (ZDR). <strong>Claude Fable 5.2<\/strong> is now available across all major platforms, including <strong>Amazon Web Services<\/strong>, <strong>Google Cloud<\/strong>, and <strong>Microsoft Azure<\/strong>.<\/p>\n<\/blockquote>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">Google\u306e\u6700\u65b0TPU\uff1a\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u6642\u4ee3\u306e\u305f\u3081\u306b\u69cb\u7bc9\u3055\u308c\u305f8tx\u30688i Max<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> Our newest TPUs: 8tx and 8i Max, built for the agentic era<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p><strong>Google<\/strong>\u306f\u3001\u7b2c8\u4e16\u4ee3<strong>TPU<\/strong>\u30dd\u30fc\u30c8\u30d5\u30a9\u30ea\u30aa\u3092\u62e1\u5f35\u3057\u3001\u30bb\u30ad\u30e5\u30a2\u3067\u6a5f\u5bc6\u6027\u306e\u9ad8\u3044\u63a8\u8ad6\u306e\u305f\u3081\u306e<strong>TPU 8tx<\/strong>\u3068\u3001\u5927\u898f\u6a21\u306a\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\u30a6\u30a3\u30f3\u30c9\u30a6\u306e\u305f\u3081\u306e<strong>TPU 8i Max<\/strong>\u3092\u767a\u8868\u3057\u307e\u3057\u305f\u3002\u3053\u308c\u3089\u306f\u3001\u8907\u96d1\u306a\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30ef\u30fc\u30af\u30ed\u30fc\u30c9\u3092\u7ba1\u7406\u3059\u308b\u305f\u3081\u306b\u3001\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u3001\u5bc6\u5ea6\u3001\u63a8\u8ad6\u306b\u7279\u5316\u3057\u305f\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u3092\u63d0\u4f9b\u3057\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u65b0\u3057\u3044\u30b7\u30b9\u30c6\u30e0\u306f\u3001\u5927\u898f\u6a21\u306a\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u7528\u306e<strong>TPU 8t<\/strong>\u3068\u30ea\u30a2\u30eb\u30bf\u30a4\u30e0\u63a8\u8ad6\u7528\u306e<strong>TPU 8i<\/strong>\u306b\u52a0\u308f\u308a\u3001AI\u30e9\u30a4\u30d5\u30b5\u30a4\u30af\u30eb\u306e\u3042\u3089\u3086\u308b\u6bb5\u968e\u3067\u7279\u5316\u3057\u305f\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u63d0\u4f9b\u3057\u307e\u3059\u3002<\/p>\n<p><strong>TPU 8tx<\/strong>\u306f\u3001\u6a5f\u5bc6\u30c7\u30fc\u30bf\u3068\u30e2\u30c7\u30eb\u306e\u91cd\u307f\u3092\u30a2\u30af\u30c6\u30a3\u30d6\u306a\u4f7f\u7528\u4e2d\u3067\u3082\u4fdd\u8b77\u30fb\u5206\u96e2\u3059\u308b\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30ec\u30d9\u30eb\u306e\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u3092\u76f4\u63a5\u7d44\u307f\u8fbc\u3093\u3067\u3044\u307e\u3059\u3002\u3053\u308c\u306f\u3001\u898f\u5236\u5bfe\u8c61\u306e\u4f01\u696d\u3084\u653f\u5e9c\u6a5f\u95a2\u304c\u3001\u6700\u3082\u6a5f\u5bc6\u6027\u306e\u9ad8\u3044\u30c7\u30fc\u30bf\u3068\u3068\u3082\u306b\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\u3084\u72ec\u81ea\u306e\u30e2\u30c7\u30eb\u3001\u81ea\u5f8b\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3092\u5b9f\u884c\u3059\u308b\u306e\u306b\u6700\u9069\u3067\u3059\u3002\u4e00\u65b9\u3001<strong>TPU 8i Max<\/strong>\u306f\u3001\u30e1\u30e2\u30ea\u306e\u58c1\u306e\u554f\u984c\u306b\u5bfe\u51e6\u3059\u308b\u305f\u3081\u306b\u8a2d\u8a08\u3055\u308c\u30012\u3064\u306e\u4f4e\u9045\u5ef6<strong>TPU 8i<\/strong>\u30c1\u30c3\u30d7\u3092\u5358\u4e00\u30d1\u30c3\u30b1\u30fc\u30b8\u306b\u7d71\u5408\u3059\u308b\u3053\u3068\u3067\u3001\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u3068\u30e1\u30e2\u30ea\u3092\u540c\u671f\u3057\u3066\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u3057\u3001<strong>TPU 8i<\/strong>\u306e2\u500d\u306e\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u3001SRAM\u3001HBM\u5e2f\u57df\u5e45\u3092\u63d0\u4f9b\u3057\u307e\u3059\u3002<\/p>\n<p><strong>Google<\/strong>\u306f\u3001<strong>PyTorch<\/strong>\u3001\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u3001\u958b\u767a\u8005\u5411\u3051\u306e\u81ea\u52d5\u5316\u30c4\u30fc\u30eb\u3092\u30b5\u30dd\u30fc\u30c8\u3059\u308b\u3053\u3068\u3067\u3001<strong>TPU<\/strong>\u306e\u30a2\u30af\u30bb\u30b7\u30d3\u30ea\u30c6\u30a3\u3092\u5411\u4e0a\u3055\u305b\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u306b\u306f\u3001<strong>TorchTPU<\/strong>\u306e\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u5316\u3001<strong>vLLM<\/strong>\u3084<strong>SGLang<\/strong>\u306a\u3069\u306e\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u63a8\u8ad6\u30a8\u30f3\u30b8\u30f3\u7528\u306e<strong>TPU<\/strong>\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3001\u304a\u3088\u3073<strong>Claude Code<\/strong>\u3084<strong>Cursor<\/strong>\u306a\u3069\u306e\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u30c4\u30fc\u30eb\u306b\u76f4\u63a5\u63a5\u7d9a\u3057\u3066<strong>TPU<\/strong>\u306e\u30aa\u30f3\u30dc\u30fc\u30c7\u30a3\u30f3\u30b0\u3092\u81ea\u52d5\u5316\u3059\u308b\u65b0\u3057\u3044\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9<strong>TPU\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30b9\u30a4\u30fc\u30c8<\/strong>\u304c\u542b\u307e\u308c\u307e\u3059\u3002<\/p>\n<p><strong>Agent Compute Engine<\/strong>\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u306f\u3001AI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u30c7\u30fc\u30bf\u30d9\u30fc\u30b9\u691c\u7d22\u3084\u5916\u90e8\u30c4\u30fc\u30eb\u3001\u4eba\u9593\u306e\u30d5\u30a3\u30fc\u30c9\u30d0\u30c3\u30af\u3092\u5f85\u3064\u9593\u306b\u983b\u7e41\u306b\u4e00\u6642\u505c\u6b62\u3059\u308b\u969b\u306e\u30a2\u30a4\u30c9\u30eb\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u306e\u7121\u99c4\u3092\u81ea\u52d5\u7684\u306b\u505c\u6b62\u30fb\u518d\u958b\u3059\u308b\u3053\u3068\u3067\u524a\u6e1b\u3057\u307e\u3059\u3002\u307e\u305f\u3001<strong>TPU Sync<\/strong>\u306f\u3001\u30c1\u30c3\u30d7HBM\u3001\u30db\u30b9\u30c8DRAM\u3001\u5171\u6709\u30b9\u30c8\u30ec\u30fc\u30b8\u3001\u30c7\u30fc\u30bf\u30bb\u30f3\u30bf\u30fc\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u3092\u5358\u4e00\u306e\u5171\u540c\u8a2d\u8a08\u3055\u308c\u305f\u30e1\u30e2\u30ea\u968e\u5c64\u306b\u7d71\u5408\u3057\u3001\u5927\u898f\u6a21\u306aKV\u30ad\u30e3\u30c3\u30b7\u30e5\u3092\u30df\u30ea\u79d2\u5358\u4f4d\u3067\u30aa\u30d5\u30c1\u30c3\u30d7\u306b\u968e\u5c64\u5316\u3059\u308b\u3053\u3068\u3067\u3001\u30e1\u30e2\u30ea\u306e\u58c1\u3092\u7a81\u7834\u3057\u307e\u3059\u3002<\/p>\n<p><strong>Google Kubernetes Engine (GKE)<\/strong>\u3067\u4e00\u822c\u63d0\u4f9b\u3055\u308c\u3066\u3044\u308b<strong>Cloud Pathways<\/strong>\u306f\u3001<strong>Google DeepMind<\/strong>\u306e<strong>Gemini<\/strong>\u306e\u30e9\u30f3\u30bf\u30a4\u30e0\u3092\u3059\u3079\u3066\u306e<strong>Google Cloud<\/strong>\u9867\u5ba2\u306b\u63d0\u4f9b\u3057\u3001\u5358\u4e00\u30b3\u30f3\u30c8\u30ed\u30fc\u30e9\u30fc\u30a2\u30fc\u30ad\u30c6\u30af\u30c1\u30e3\u3092\u4f7f\u7528\u3057\u3066\u5927\u898f\u6a21\u306a<strong>TPU<\/strong>\u30b7\u30b9\u30c6\u30e0\u3092\u30aa\u30fc\u30b1\u30b9\u30c8\u30ec\u30fc\u30b7\u30e7\u30f3\u3067\u304d\u307e\u3059\u3002\u3055\u3089\u306b\u3001<strong>GKE<\/strong>\u306e\u52d5\u7684\u30b9\u30e9\u30a4\u30b7\u30f3\u30b0\u306f\u3001\u5149\u5b66\u56de\u8def\u30b9\u30a4\u30c3\u30c1\u30f3\u30b0\uff08OCS\uff09\u6280\u8853\u3092\u4f7f\u7528\u3057\u3066\u7269\u7406\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30d7\u30ed\u30d3\u30b8\u30e7\u30cb\u30f3\u30b0\u3068\u8ad6\u7406\u30b9\u30e9\u30a4\u30b9\u5f62\u6210\u3092\u5206\u96e2\u3057\u3001\u30b8\u30e7\u30d6\u306e\u8d77\u52d5\u6642\u9593\u3092\u6700\u59275\u500d\u77ed\u7e2e\u3057\u3001\u969c\u5bb3\u56de\u5fa9\u3092\u6700\u59274.5\u500d\u52a0\u901f\u3057\u307e\u3059\u3002<\/p>\n<p><strong>Google<\/strong>\u306f\u3001\u30b7\u30ea\u30b3\u30f3\u30a2\u30fc\u30ad\u30c6\u30af\u30c8\u3001\u30cd\u30c3\u30c8\u30ef\u30fc\u30ad\u30f3\u30b0\u30a8\u30f3\u30b8\u30cb\u30a2\u3001\u30b3\u30f3\u30d1\u30a4\u30e9\u30c1\u30fc\u30e0\u3001<strong>Google DeepMind<\/strong>\u306e\u7814\u7a76\u8005\u304c\u5354\u529b\u3057\u3066\u3001<strong>AlphaChip<\/strong>\u3084<strong>Gemini<\/strong>\u306a\u3069\u306eAI\u3092\u6d3b\u7528\u3057\u3066\u6b21\u4e16\u4ee3AI\u30a4\u30f3\u30d5\u30e9\u30b9\u30c8\u30e9\u30af\u30c1\u30e3\u3092\u8a2d\u8a08\u30fb\u69cb\u7bc9\u3059\u308b\u3053\u3068\u3067\u3001<strong>TPU<\/strong>\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u958b\u767a\u3092\u52a0\u901f\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/blog.google\/innovation-and-ai\/infrastructure-and-cloud\/global-network\/tpu-8i-max-and-8tx\/\" target=\"_blank\" rel=\"noopener\">Google Blog \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> Google's new TPU 8tx and 8i Max, alongside a comprehensive open-source software stack and advanced orchestration tools, are designed to power the 'agentic era' with specialized hardware for secure, high-context AI workloads and enhanced developer accessibility.<\/li>\n<li><strong>\u8457\u8005:<\/strong> Amin Vahdat, SVP and Chief Technologist, AI and Infrastructure<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p><strong>Google<\/strong> has expanded its eighth-generation <strong>TPU<\/strong> portfolio with the introduction of <strong>TPU 8tx<\/strong> for secure, confidential inference and <strong>TPU 8i Max<\/strong> for massive context windows. These new systems are designed to manage complex agentic workloads by providing specialized hardware for security, density, and reasoning. They join the existing <strong>TPU 8t<\/strong> for large-scale training and <strong>TPU 8i<\/strong> for real-time reasoning, offering specialized performance for every stage of the AI lifecycle.<\/p>\n<p>The <strong>TPU 8tx<\/strong> integrates hardware-level security directly, protecting and isolating sensitive data and model weights even during active use. This makes it ideal for regulated enterprises and governments running frontier or proprietary models and autonomous agents with their most sensitive data. The <strong>TPU 8i Max<\/strong>, conversely, is engineered to address the memory wall bottleneck by combining two low-latency <strong>TPU 8i<\/strong> chips into a single package, delivering double the compute, SRAM, and HBM bandwidth of <strong>TPU 8i<\/strong>.<\/p>\n<p><strong>Google<\/strong> is enhancing <strong>TPU<\/strong> accessibility by supporting <strong>PyTorch<\/strong>, open-source software, and automated tools for developers. This includes open-sourcing <strong>TorchTPU<\/strong>, providing out-of-the-box <strong>TPU<\/strong> backends for leading open-source inference engines like <strong>vLLM<\/strong> and <strong>SGLang<\/strong>, and introducing a new open-source <strong>TPU agent suite<\/strong> that integrates with coding tools like <strong>Claude Code<\/strong> and <strong>Cursor<\/strong> to automate <strong>TPU<\/strong> onboarding and optimization.<\/p>\n<p>The <strong>Agent Compute Engine<\/strong> software automatically suspends and resumes idle sessions for stateful AI agents, significantly cutting operational infrastructure costs by eliminating idle compute waste. Furthermore, <strong>TPU Sync<\/strong>, a model-state layer for <strong>TPU<\/strong> clusters, unifies memory, storage, and networking into a single, co-designed memory hierarchy, breaking through the memory wall by automatically tiering massive KV caches off-chip in milliseconds.<\/p>\n<p>Now generally available on <strong>Google Kubernetes Engine (GKE)<\/strong>, <strong>Cloud Pathways<\/strong> brings the <strong>Google DeepMind<\/strong> runtime behind <strong>Gemini<\/strong> to every <strong>Google Cloud<\/strong> customer, enabling orchestration of large-scale <strong>TPU<\/strong> systems from a single controller. Additionally, <strong>GKE dynamic slicing<\/strong>, utilizing optical circuit switching (OCS) technology, decouples physical hardware provisioning from logical slice formation, reducing job startup times by up to 5x and accelerating fault recovery by up to 4.5x.<\/p>\n<p><strong>Google<\/strong>'s co-design approach, involving silicon architects, networking engineers, compiler teams, and <strong>Google DeepMind<\/strong> researchers, leverages AI tools like <strong>AlphaChip<\/strong> and <strong>Gemini<\/strong> to rapidly accelerate the development of next-generation <strong>TPU<\/strong> hardware.<\/p>\n<\/blockquote>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">AI\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7Nominal\u30018\u30f6\u6708\u3067\u8a55\u4fa1\u984d30\u5104\u30c9\u30eb\u306b3\u500d\u5897<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> AI hardware startup Nominal triples valuation to $3B in 8 months<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p>AI\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7\u3067\u3042\u308b<strong>Nominal<\/strong>\u304c\u3001\u308f\u305a\u304b8\u30f6\u6708\u3067\u8a55\u4fa1\u984d\u30923\u500d\u306e<strong>30\u5104\u30c9\u30eb<\/strong>\u306b\u307e\u3067\u6025\u5897\u3055\u305b\u307e\u3057\u305f\u3002\u3053\u306e\u6025\u901f\u306a\u8a55\u4fa1\u984d\u306e\u5897\u52a0\u306f\u3001AI\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u5206\u91ce\u306b\u304a\u3051\u308b\u6295\u8cc7\u5bb6\u306e\u5f37\u3044\u95a2\u5fc3\u3068\u3001\u540c\u793e\u306e\u6280\u8853\u304a\u3088\u3073\u5e02\u5834\u30dd\u30c6\u30f3\u30b7\u30e3\u30eb\u306b\u5bfe\u3059\u308b\u9ad8\u3044\u671f\u5f85\u3092\u6d6e\u304d\u5f6b\u308a\u306b\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u3053\u306e\u30cb\u30e5\u30fc\u30b9\u306f\u3001AI\u6280\u8853\u306e\u9032\u5316\u304c\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30a4\u30f3\u30d5\u30e9\u3078\u306e\u5927\u898f\u6a21\u306a\u6295\u8cc7\u3092\u4fc3\u9032\u3057\u3066\u3044\u308b\u73fe\u5728\u306e\u30c8\u30ec\u30f3\u30c9\u3092\u53cd\u6620\u3057\u3066\u304a\u308a\u3001\u7279\u306bAI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3084\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u306e\u9700\u8981\u304c\u9ad8\u307e\u308b\u4e2d\u3067\u3001\u5c02\u7528\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u306e\u91cd\u8981\u6027\u304c\u5897\u3057\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u5506\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/techcrunch.com\/2026\/11\/09\/ai-hardware-startup-nominal-triples-valuation-to-3b-in-8-months\/\" target=\"_blank\" rel=\"noopener\">TechCrunch \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> AI hardware startup Nominal's valuation tripled to $3 billion in 8 months, signaling robust investor confidence and rapid growth in the specialized AI hardware market driven by increasing demand for AI agents and LLMs.<\/li>\n<li><strong>\u8457\u8005:<\/strong> Julie Bort<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p>The AI hardware startup <strong>Nominal<\/strong> has seen its valuation triple to an impressive <strong>$3 billion<\/strong> in just eight months. This rapid increase in valuation highlights strong investor interest and high expectations for the company's technology and market potential within the AI hardware sector.<\/p>\n<p>This news reflects the current trend where the advancement of AI technology is driving significant investment in hardware infrastructure, suggesting the growing importance of specialized hardware, especially with the increasing demand for AI agents and large language models.<\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>Anthropic\u306e\u6700\u65b0\u30e2\u30c7\u30ebClaude Fable 5.2\u3068Mythos 5.2\u306e\u6280\u8853\u7684\u9032\u6b69\u3001Google\u306eAI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u5411\u3051TPU 8tx\u30688i Max\u3001\u305d\u3057\u3066AI\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30b9\u30bf\u30fc\u30c8\u30a2\u30c3\u30d7Nominal\u306e\u6025\u6210\u9577\u3092\u6df1\u6398\u308a\u3002<\/p>\n","protected":false},"author":1,"featured_media":856,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"sns_share_botton_hide":"","vkExUnit_sns_title":"","_vk_print_noindex":"","sitemap_hide":"","_veu_custom_css":"","veu_display_promotion_alert":"","vkexunit_cta_each_option":"","_lightning_design_setting":[],"footnotes":""},"categories":[3],"tags":[8,17,523,58,75,16,400,56,51,26,15],"class_list":["post-1249","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-columns","tag-ai","tag-anthropic","tag-google","tag-llm","tag-tpu","tag-56","tag-51","tag-26"],"veu_head_title_object":{"title":"","add_site_title":""},"_links":{"self":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts\/1249","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/comments?post=1249"}],"version-history":[{"count":0,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts\/1249\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/media\/856"}],"wp:attachment":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/media?parent=1249"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/categories?post=1249"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/tags?post=1249"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}