---
格式版本: 2
标题: "AI Chips Need Liquid Cooling"
原文链接: "https://hardwarefyi.substack.com/p/ai-chips-need-liquid-cooling"
发布日期: "2026-08-15"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:scrape:strict_html_body"
发布时间证据: "div class=pencraft pc-reset color-pub-secondary-text-hGQ02T line-height-20-t4M0El font-meta-MWBumP size-11-NuY2Zx weight-medium-fw81nC transform-uppercase-yKDgcq reset-IxiVJZ meta-EgzBVA: Aug 15, 2026"
发布时间校准原因: "规则确认唯一严格发布时间，来源 scrape:strict_html_body"
发布时间校准置信度: "high"
发布时间候选数量: 10
发布时间严格候选数量: 1
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-31T11:53:59+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-31T11:48:46+08:00"
入库时间: "2026-08-31T03:56:34.059Z"
来源平台: "Substack 数据中心相关博客搜索"
搜索渠道: "source_template"
搜索词: "site:substack.com Liquid Cooling"
匹配关键词:
  - "Liquid Cooling"
  - "AI"
  - "Immersion"
  - "performance"
相关厂家:
  - "Google"
  - "Microsoft"
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 54
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "正文是Substack链接汇编，主要相关段落二次概述Google液冷演示，给出超过1GW液冷部署、芯片功耗超过1000W、CDU—机柜歧管—冷板—换热器回路及自2020年达到99.999%可用性等事实。固定知识库未完整出现这些部署与可靠性数据，存在一定历史增量，但均来自所链接的Google演示，当前页面没有独家采访、实测或原创工程增量，且夹杂大量无关内容、发布日期缺失。命中无独家事实的二手汇编/入口型页面否决，当前页面本身不值得替代原始演示保留，最高54分。"
AI质检模型: "gpt-5.6-sol"
AI质检时间: "2026-08-31T11:56:46+08:00"
AI主题相关性: 13
AI来源权威性: 6
AI新颖性: 9
AI技术细节: 14
AI商业部署信号: 7
AI完整性: 5
AI评分提示词版本: "v17-精简生产版"
AI评分提示词SHA256: "48fb9777f386026761b4873eaff30807694fb11e9b352d7c69bf2dfde750cc7d"
AI评分知识库版本: "knowledge_base_v1-20260819+runtime.62"
AI评分知识库SHA256: "b53abf2242d71cbdd0daf92240736320626be4a6f0e164c2fafe18c1a554a4a3"
AI评分知识库检索词: "[\"Google\",\"Liquid Cooling\",\"site:substack.com Liquid Cooling\",\"NPO\",\"Cold Plate\",\"RAS\",\"Microsoft\",\"FYI\",\"CA\",\"GW\",\"CDU\",\"TPU\"]"
AI评分知识库命中: "[{\"id\":\"july-correct-0115\",\"title\":\"锚定 300kW 整机柜演进方向 OAII 社区三项规范联合发布，树立AIDC基础设施建设标准答案\",\"sourceType\":\"labeled_article\",\"time\":\"2026-07\",\"matchedTerms\":[\"NPO\",\"GW\",\"CDU\"],\"rank\":-15.407299121246501},{\"id\":\"july-correct-0033\",\"title\":\"Microsoft, Alphabet, Meta Pivot from Buy to Build in AI\",\"sourceType\":\"labeled_article\",\"time\":\"2026-07\",\"matchedTerms\":[\"Google\",\"RAS\",\"Microsoft\",\"CA\",\"GW\",\"TPU\"],\"rank\":-14.372066867889956},{\"id\":\"historical-jan-apr-05\",\"title\":\"五、AI 光通信需求、可插拔光模块与光互连接口演进\",\"sourceType\":\"curated_item\",\"time\":\"2026-01_to_2026-04\",\"matchedTerms\":[\"NPO\",\"Cold Plate\",\"Microsoft\"],\"rank\":-13.645542720329559},{\"id\":\"historical-jan-apr-02\",\"title\":\"二、Google Cloud Next '26：AI Hypercomputer 与第八代 TPU 发布\",\"sourceType\":\"curated_item\",\"time\":\"2026-01_to_2026-04\",\"matchedTerms\":[\"Google\",\"CA\",\"TPU\"],\"rank\":-13.55652543735205},{\"id\":\"runtime-e8b946c3131877fe7add476b\",\"title\":\"Peeling Apart That Supposed $120 Billion Chip Deal Google Inked With Marvell\",\"sourceType\":\"ai_excellent_article\",\"time\":\"2026-08-27\",\"matchedTerms\":[\"Google\",\"RAS\",\"CA\",\"TPU\"],\"rank\":-10.83382488750567}]"
AI摘要: "Google已部署超过1GW的液体冷却算力基础设施，其方案为直接芯片冷却：冷却液经分配单元送至服务器冷板吸收芯片热量，再经换热器传给建筑水系统，并用于TPU设施。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-08-31T23:37:17.032Z"
采集批次: "2026年8月31日11点48分36秒"
采集批次ID: "20260831-114836-897"
去重键: "https://hardwarefyi.substack.com/p/ai-chips-need-liquid-cooling"
---

*Topics worth your time: Google’s development of liquid-cooled compute at gigawatt scale, a visual guide to control theory, and the history of Taiwan’s power electronics giant. A slightly shorter read this weekend with a few good threads to explore!*

Hardware FYI is brought to you with support from our partners.

---

### Interesting Presentation: Liquid Cooling for High-Performance Compute

Jorge Padilla at Google has a great presentation on the **[development of liquid cooling for Google’s compute infrastructure](https://hc2025.hotchips.org/assets/program/tutorials/HC2025.T1DCRacks.S3.Padilla.pdf)**, tracing the company’s progress from early prototypes to more than 1 GW of deployed liquid-cooled capacity. A useful way to frame the shift is through power-density requirements. Passively cooled devices (think phones, tablets) generally operate at 5–10 W, actively cooled laptops around 15–35 W, and modern high-performance compute chips at more than 1,000 W - well beyond traditional limits of air-cooled systems.

Google’s approach, and more broadly across the industry, uses direct-to-chip cooling. This means:

1. A **coolant distribution unit (CDU)** sends liquid through the rack manifold and onward to each server.
2. There, sealed **cold plates** sit on the processors and pick up their heat.
3. The coolant returns to the CDU, where a **heat exchanger** passes that heat into the building’s water system; the coolant and facility water never mix.

One useful, though slightly informal way to map the broader design space is to ask: how much of the machine are you willing to put in liquid? Phase one cools only the chip, using the cold-plate setup described above for Google’s TPU infrastructure. Phase two puts an entire group of components in a dielectric bath; startups like **[Iceotope](https://www.iceotope.com/technology)** seal coolant inside the server chassis and circulate it across the electronics before transferring the heat to a secondary water loop. Phase three expands immersion to the full rack, and phase four draws the boundary around the data center itself; Microsoft’s **[Project Natick](https://natick.research.microsoft.com/)** is probably the best-known example.

Each step is more aggressive about removing heat from the system, but also changes how the infrastructure is deployed and maintained. Phases two through four are relatively new approaches with varying stages of R&D; for reference, Google’s liquid-cooled fleet has been running at ‘five nines’ uptime (99.999%) since 2020.

---

### Interesting Chart: The Map of Control Theory

Brian Douglas’s **[Map of Control Theory](https://engineeringmedia.com/maps)**. The **[self-balancing inverted pendulum](https://www.youtube.com/watch?v=-UTqKqoh-5A)** is an especially cool example of linear control.

---

### Presented by Jiga: Manufacturing ≠ Transactions

Source custom parts with direct access to the manufacturers making them. No middleman, no black box. **[Jiga](https://www.jiga.io/?utm_source=hardware-fyi&utm_medium=communities&utm_campaign=sponsorships&utm_content=manufacturing-not-transactions)** gives you a vetted network and one accountable partner standing behind every order. You know who's building your parts and communicate with them directly. AI-assisted operations handle supplier matching, quoting, quality verification, and production tracking, so the relationships stay human while the operations run at speed.

---

### Interesting Video: Taiwan’s Power Supply Giant

Asianometry’s **[history of Delta Electronics](https://www.youtube.com/watch?v=nRKxKz2eiOo)**,Taiwan’s power electronics giant behind everyday electronics.

---

### Manufacturing & Startup News

- **[Uplift](https://www.uplift.us/)** launched with **[$7M in seed funding](https://www.linkedin.com/posts/cnitschelm_i-am-proud-to-announce-the-launch-of-uplift-activity-7493667694282436608-_0GZ/)** to build homes on automotive-style production lines, starting with its first deployments over the next year.
- **[Sila](https://silanano.com/)** received a **[conditional $1.4B DoW loan commitment](https://www.businesswire.com/news/home/20260807648387/en/Sila-Receives-Conditional-$1.4-Billion-Loan-Commitment-from-U.S.-Department-of-War-to-Accelerate-Domestic-Battery-Technology-Manufacturing)** to expand silicon-carbon anode production in Washington and build lithium-ion cells for drones and other specialty applications.
- **[Recast Systems](https://www.recastsystems.com/)** launched from stealth with **[cloud-seeding operations](https://x.com/recast/status/2087582521320177932)** in Texas and New Mexico.
- **[Heron Power](https://heronpower.com/)** is turning a former warehouse in Morgan Hill, CA into a **[factory for 5 MW grid converters](https://heronpower.com/news-stories/press-release-heron-power-selects-first-factory-location-to-build-next-generation-power-electronics-in-the-united-states)**, targeting production in late 2027.

---

### Tools From Our Partners

- **[Aris Machina Protos](https://arismachina.com/protos/?ref=hardwarefyi.com)** – Design & validate anything, with the AI-native workspace for industrial science and hardware R&D.
- **[Jiga](https://jiga.io/?ref=hardwarefyi.com)** – Custom parts at digital speed, with trusted shop access.
- **[SolidWorks for Startups](https://app.smartsheet.com/b/form/8da64f67a7564cbeaf2e9a18ffe22900?VAR=HardwareFYI&VAR%20Refered=HardwareFYI)** – Full access to SolidWorks’ product development tools, free for early-stage startups.
- **[Zoo](https://zoo.dev//?utm_source=hardwarefyi&utm_medium=newsletter&utm_campaign=sponsorship)** – AI-native CAD with a modern geometry engine, readable code, and editable feature trees for precise, editable models.

---

### Access Every Link We’ve Ever Shared 🎞️

Refer **three friends** to *Hardware FYI* and unlock our full archive.  
Missed an issue? Hunting for that one link? Every tool, guide, and article we’ve ever shared is in here.

P.S. If you find something interesting—an article, tool, funding round, project, really anything— [send it our way](https://tally.so/r/mVk6L6)!

P.P.S. 20,000+ engineers, founders, and operators read this newsletter. If you want to reach them, [get in touch](https://www.hardwarefyi.com/become-a-sponsor/).