clm_a8962adccea24aa48ab7d443279234da_r1

2025-11-28_tpuv7-google-takes-a-swing-at-the::TPU7-0006

① SA Source

• Source: 開啟完整 SA 文章
• Section: Google AI Infrastructure Supremacy: Systems Matter More Than Microarchitecture
• Line hint: 95

Context Before

Source: SemiAnalysis TCO Model ↗, Anthropic and OpenAI

OpenAI hasn’t even deployed TPUs yet and they’ve already saved ~30% on their entire lab wide NVIDIA fleet. This demonstrates how the perf per TCO advantage of TPUs is so strong that you already get the gains from adopting TPUs even before turning one on.

Evidence

In early September, we revealed that one of the big external customers will be Anthropic, ↗ with demand of at least 1 million TPUs.

Context After

As a result, our institutional clients have had plenty of heads up on one of the largest performance dispersions in the AI Trade to date. SemiAnalysis was the first to break all these insights because no other research firm can connect the dots from the fabs to supply chain through the datacenters to the labs. To get access to these insights as and stay ahead of the curve: sales@semianalysis.com

Onto the deal.

② Atomic Claim

SemiAnalysis 先前估計 Anthropic 外部 TPU demand 至少 100 萬顆。

• Epistemic Mode: ATTRIBUTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "tpu_demand",
  "context_nodes": [
    {
      "id": "04_knowledge_base/TPU",
      "label": "TPU"
    }
  ],
  "entity": {
    "id": "02_companies/Anthropic",
    "label": "Anthropic"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "1 million"
    ],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [
      "1 million"
    ],
    "value_text": "at least 1 million TPUs"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	Anthropic	Anthropic
context_0	TPU	TPU

⑤ Human Review

請在 Properties 逐項確認：

• 原文 → Atomic Claim 是否忠實
• Atomic Claim → Semantic Frame 是否忠實
• Canonical Entity mapping 是否正確
• Epistemic mode 是否保留原文語氣
• 最後選擇 review_action

Review state

Markdown 內文不是正式 approval。只有 Apply bridge 寫入的 Decision Ledger event 才是正式決策。