PDF 原檔：報告_OCP2026_Google_主題演講_智慧基礎設施_20261015_original.pdf
講者公開簡報（OCP 官方 Recorded Content 頁提供下載）。文字以 PyMuPDF 逐頁抽取，
### p<n>為頁碼；圖表內容見原檔。
圖片清單（已驗證 2026-10-19）
| 檔名 | 頁 | 分類 | 親眼所見內容 |
|---|---|---|---|
報告_OCP2026_Google_主題演講_智慧基礎設施_20261015_p6.png |
p6 | 真資料圖 | Radical AI System Innovations 八格：TPUs/Axion/GPUs、HBM 288GB co-packaged、Vertical power（Through-substrate BGA）、Deschutes CDU 液冷、ICI multi-TB/s torus、Palomar MEMS OCS、100kW+ 3D pods、1P & 3P GW campuses |
報告_OCP2026_Google_主題演講_智慧基礎設施_20261015_p10.png |
p10 | 真資料圖 | 三項開放資料中心貢獻示意：Diablo 400V Disaggregated Power（Google·Meta·Microsoft）、Project Deschutes CDU Cooling（Google）、Falcon Networking（Google·NVIDIA） |
報告_OCP2026_Google_主題演講_智慧基礎設施_20261015_p12.png |
p12 | 真資料圖 | 電網到晶片電力鏈示意與六項貢獻：1 電網介面（Google·Meta·Microsoft·NVIDIA）、2 SST 中壓→800 VDC（Google·Microsoft·NVIDIA）、3 潔淨備援電力（Google·Microsoft）、4 LVDC 配電（39 contributors，ABB 到 Vertiv）、5 Open Data Center for AI 白皮書（OCP·AMD·Google·Meta·Microsoft·NVIDIA·Oracle）、6 能源碳水指引 |
報告_OCP2026_Google_主題演講_智慧基礎設施_20261015_p18.png |
p18 | 真資料圖 | 極端天氣預測 AI 模型流程圖：物理模擬＋歷史氣象→ML 迴歸→偏誤校正的場址級預測，對應十年（選址）、年度（模組升級）、每週（排程）三個時間尺度 |
原始內容
p1
Intelligent Infrastructure Powering the Intelligence Revolution Partha Ranganathan
p2
Intelligence Infrastructure for Intelligence Intelligent Infrastructure for Intelligence Industry Innovation for Intelligent Infrastructure
p3
Artificial intelligence is transforming all aspects of our life health, software, engineering, education, productivity, creativity, entertainment… We are in extraordinary times -
p4
Extraordinary advances need infrastructure “If AI researchers are space explorers discovering new worlds, we are the ones building the rockets”
p5
Computing, reimagined across the full AI stack
p6
Circa 2003 Silicon accelerators TPUs, Axion & GPUs HBM 288GB co-packaged Vertical power Through-substrate BGA Liquid cooling Deschutes CDU loops ICI Multi-TB/s torus fabric Optics Palomar MEMS OCS Densification 100kW+ 3D pods 3P data centers 1P & 3P GW campuses
p7
System Efficiency (perf, power, cost across generations)
p8
Last year, we called on the community to do more… From ability to agility Agile fungible systems for AI
p9
And the stepped up…
p10
From Atoms to Bytes: Diablo 400V Disaggregated Power Google · Meta · Microsoft Falcon Networking Google · NVIDIA Project Deschutes CDU Cooling Google
p11
Sustainable open data centers: Responsible water stewardship commitments Public Transparency Net Positive Replenishment Community Infrastructure Funding Context-Driven Cooling Alternative Water Sources Demand response with utilities/grids
p12
Grid interface for disturbance performance Google · Meta · Microsoft · NVIDIA SST: medium voltage → 800 VDC Google · Microsoft · NVIDIA Clean backup power technologies Google · Microsoft LVDC power distribution 39 contributors, ABB to Vertiv Energy, carbon & water guidelines Villanova · Google · GSA · Castrol · LiquidCool · Submer · Hscale And many many more… Open Data Center for AI white paper OCP · AMD · Google · Meta · Microsoft · NVIDIA · Oracle
p13
（本頁無可抽取文字，內容在圖中）
p14
Looking ahead to The next big inflection intelligent infrastructure
p15
From “systems for AI” to “ Four vignettes from across the stack 1 2 3 4 ”
p16
AlphaEvolve for evolutionary LLM-guided search New ideas & championship-winning benefits Evolutionary agents to discover new Vignette 1: AI to transform systems hardware https://arxiv.org/abs/2602.22425
p17
Every code change scanned before merge 92% precision with seconds of scan time Mantis open sourced to community Pervasive agentic security check and bug fix in software Vignette 2: AI to transform systems software https://cloud.google.com/blog/topics/systems/using-ai-agents-to-secure-google-infrastructure
p18
Right provision costs of infrastructure vs risks of downtime. Site-specific multi-epochal modeling. Applied to Google data centers. Methodology opened to community. for extreme weather event prediction Vignette 3: AI to transform data center design https://ieeexplore.ieee.org/document/11617528/
p19
Accelerate & amplify data center capacity by automating drudgery & toil; increased safety and speed for data center operations & capacity delivery Vignette 4: AI to transform data center operations
p20
opportunities ahead… Everything tastes better with a sprinkle of AI!
p21
US Igniting Industry Useful Sustainable Impressive
p22
Industry Innovation for Intelligent Infrastructure Call to Action
p23
Tiers of checks & balances to address hate, harm, hallucinations Information quality, Explainable AI Secure by default, private by design AI principles, Responsible AI practices Partnerships with governments, industry, civil society, researchers Partnerships & communities 1 Call to action
p24
Lack of high quality datasets and simulation, data sparsity, +ve/-ve datasets and perturbation for space expansion, generality Data collection & design challenges Objectives, monitoring, perturbation, safety, autotuning; denormalization, latency (human assists), increased search space for codesign, scale, … Systems challenges + ML challenges 2 Call to action
p25
3 Call to action Test Operate Code Deploy Monitor Plan Build Release The 10x problem & rethinking interfaces for the AI-infused workflow Hardware Firmware Low level Interface Scaling Library Tools Tracing Debugging Monitoring Workloads Driver Compiler
p26
4 Call to action Online albums → Single songs → Subscriptions → YouTube shorts Replicate | AI-lift & shift Recombine | AI-augmented Reimagine | AI-first Reinvent | AI-native
p27
Extraordinary AI advances & extraordinary infrastructure
p28
We are in extraordinary times Artificial intelligence is transforming all aspects of science & society
p29
We are in extraordinary times Artificial intelligence is transforming all aspects of science & society
p30
If AI researchers are deciphering the code of life,
p31
Join us on the journey! Intelligent Infrastructure for the Intelligence Era