查看購物車
  • English
  • Japanese
  • Korean
封面
市場調查報告書
商品編碼
2106253

主權人工智慧：在人們對人工智慧泡沫的擔憂中，它正成為人工智慧基礎設施支出的穩定驅動力。

Sovereign AI Emerging as Stable Driver for AI Infra Spending amid Bubble Fears
出版日期: | 出版商: MTN Consulting, LLC | 英文 27 Pages | 訂單完成後即時交付

價格

調查範圍

本報告是首份關於主權AI的調查報告，該議題預計將在未來幾十年內至關重要。調查涵蓋42個國家，其中38個國家在調查期間（2025年1月至今）確認開展了符合標準的活動。中國和美國的項目價值和匯總數據未包含在內。這兩個市場在MTN顧問公司的其他調查中單獨進行追蹤。

本研究基於該公司的 InfraVantage 資料集。該資料集涵蓋 2025 年之後公佈的 195 個項目及其供應商關係，涉及 10 個領域：第三方計算、內部設計計算、網路、傳輸、電力設施、能源供應商、冷卻、儲存/記憶體、陸地和海底光纖以及陸地/建築/電纜。此外，本研究也區分了由國家政策驅動的「主權導向計畫」和出於一般商業目的的超大規模資料中心業者及企業設施擴建計畫。

主要發現

在受訪的42個國家中，有38個國家的人工智慧基礎設施總支出達到9,336億美元。這一數字涵蓋所有所有權類型，例如企業、超大規模資料中心業者雲端服務商、政府、電信業者和主權雲端服務供應商。在構成這總額的195個追蹤項目中，有73個符合主權AI認證標準。具體而言，這些項目由政府或政府附屬機構擁有，擁有明確的主權品牌，或由政策主導的資金支持，這使其區別於典型的超大規模資料中心業者或企業主導的擴張項目。這73個主權AI相關項目的總價值約為3,674億美元，約佔9,336億美元總額的39%。預計這一比例將持續上升。人們對主權AI的興趣濃厚，即使整體人工智慧市場大幅調整，需求也可能持續存在。

符合目標標準的項目形成速度也在加快。在2026年上半年公佈的80個項目中，有29個（佔36%）與主權AI相關。這一比例高於2025年上半年的62個項目中的20個（佔32%）。不同地區主權AI相關項目的比例有差異，歐洲的比例最高，為43%（79個項目中的34個），中東和非洲的比例與之相近，為43%（28個項目中的12個）。北美和南美的比例為41%（17個項目中的7個），而亞洲雖然是項目數量第二多的市場，但其比例最低，僅為28%（71個項目中的20個）。

供應商集中度也在不斷提高。在73個獲得自主人工智慧認證的專案中，NVIDIA在39個專案中被列為主要供應商，遠遠領先AMD（僅6個專案被列為主要供應商）。同時，人工智慧基礎設施的開發惠及眾多產品類型的供應商，為自主人工智慧領域留下了更大的成長空間。

在選擇底層技術和設計設施方面，主權AI設施與商業設施之間沒有根本區別。

在歐洲，將資料中心的廢熱用於區域供熱正變得越來越普遍。位於德國呂貝瑙的Schwarz Digits人工智慧資料中心和位於慕尼黑的德國電信/NVIDIA工業人工智慧雲端平台均採用直接液冷技術，將產生的熱量輸送至市政區域供熱網路。在整個市場範圍內，我們追蹤的十個廢熱再利用項目中，有八個集中在以下國家之一：德國、丹麥、芬蘭、挪威或瑞典。

前景

目前人工智慧基礎設施建設的快速推進引發了人們對人工智慧投資泡沫的擔憂，而這種擔憂並非空穴來風。並非所有已宣布的吉瓦級項目都能按計劃建成。然而，與人工智慧基礎設施市場的整體資本投資相比，對自主人工智慧基礎設施的需求被認為更具永續性。在美國政治環境變化的背景下，其他地區正日益尋求減少對美國公司和美國技術生態系統的依賴。這促使各國政府和區域運營商尋求獲得能夠直接管理的人工智慧運算能力。即使整體超大規模資料中心業者投資週期放緩，這個旨在維護自身主權的自主人工智慧領域預計仍將繼續成長。

目錄

  • 重點：投影片 2
  • 報告摘要：幻燈片 3
  • 市場機會：第5頁
  • 自主人工智慧與商業人工智慧：供應商配置差異：第13頁
  • 主權AI的未來展望：第25頁
  • 關於 MTN 諮詢：第 26 頁

調查範圍

提及的組織：

  • Adani Group
  • ADQ
  • Aker ASA
  • Alibaba
  • Amazon / AWS
  • AMD
  • Arista
  • Backblaze
  • Baidu
  • Bell Canada
  • Bpifrance
  • Broadcom
  • BUZZ HPC (HIVE Digital)
  • Centuria Capital Group
  • Ciena
  • Cisco
  • Cohere
  • Colt
  • CoreWeave
  • Corning
  • Dell
  • Deutsche Telekom
  • Digital Realty
  • EDF
  • Eni
  • Equinix
  • EuroHPC JU
  • EXA Infrastructure
  • Firmus Technologies
  • Fujitsu
  • G42
  • Google
  • GS Group
  • HUMAIN
  • Hypertec
  • Infinera
  • Ireland CRU
  • IT4LIA / CINECA
  • Italy Ministry of University and Research (MUR)
  • Keppel
  • Khazna
  • Kioxia
  • Lightstorm
  • Marvell
  • MERLIN Edged
  • Meta
  • Micron
  • Microsoft
  • Mistral AI
  • Mubadala
  • 紐約州（州長辦公室）
  • Nokia
  • Nscale
  • NVIDIA
  • OpenAI
  • Oracle
  • PJM Interconnection
  • ResetData
  • SanDisk
  • Scality
  • Schneider Electric
  • Schwarz Digits
  • Singtel
  • SK Group
  • SK Hynix
  • SoftBank
  • Start Campus
  • Tata Communications
  • TikTok / ByteDance
  • Vantage Data Centers
  • VAST Data
  • Vertiv
  • Wall Street Journal
  • WEKA
Product Code: DCPC-05082026-1

Scope

This is our kickoff study of Sovereign AI, a topic we expect to be important for decades. The initial study is focused on 42 countries, 38 of which have qualifying activity in our focus timeframe (Jan 2025 – present). We exclude China and the United States from project value and totals; these markets are tracked separately in MTN Consulting’s other research.

The study draws on our InfraVantage dataset. For projects with an announced date of 2025 or later, this dataset includes 195 tracked projects and vendor relationships across ten segments – compute (third-party), compute (self-designed), networking, transmission, power equipment, energy suppliers, cooling, storage/memory, terrestrial and subsea fiber, and land/buildings/cabling. We separate sovereignty-driven projects, pursued for national policy reasons, from general-commercial hyperscaler and enterprise capacity.

Key findings

$933.6B in tracked AI infrastructure spending (all ownership types - enterprise, hyperscaler, government, telco, sovereign cloud operator) across 38 of the 42 countries in scope. Of the 195 tracked projects making up this total, 73 clear our sovereignty-qualifying bar: government or state-linked ownership, explicit sovereignty branding, or policy-driven funding, rather than ordinary hyperscaler or enterprise buildout. These 73 sovereignty-qualifying projects amount to about $367.4 billion of the $933.6B total - roughly 39% of every tracked AI infrastructure dollar. We believe this share will continue to rise: sovereign AI interest is high and will likely persist even if the broader AI market faces a major correction.

Qualifying deal-making is accelerating: 36% of all deals (29 of 80) announced in 1H2026 were sovereignty-qualifying, up from 32% (20 of 62) in 1H2025. Regional qualifying rates vary: Europe is highest (34 of 79 tracked deals, 43%), MEA close behind (12 of 28, 43%), Americas in between (7 of 17, 41%), Asia lowest (20 of 71, 28%) despite being second-largest market by volume.

Vendor concentration is high: out of the 73 qualifying Sovereign AI projects, NVIDIA is named as the primary vendor in 39 cases, far ahead of AMD’s 6 references. But the AI buildout is benefiting dozens of vendors across many product categories, and there is upside in the Sovereign AI piece.

Underlying technology selection and facility design is not fundamentally different across sovereign AI vs. commercial builds.

Waste-heat reuse for district heating is common in Europe: Germany’s Schwarz Digits AI Data Center (Lubbenau) and Deutsche Telekom/NVIDIA Industrial AI Cloud (Munich) both pair direct liquid cooling with heat piped into municipal district-heating networks. Market-wide, 8 of the 10 tracked heat-reuse commitments sit in Germany, Denmark, Finland, Norway, or Sweden.

Outlook

The current rush to build AI infrastructure has raised well-founded concerns about an AI spending bubble, and not every announced gigawatt will be built as planned. But demand for sovereign AI infrastructure specifically is more durable than the broader buildout. Shifts in the US political environment have sharpened interest elsewhere in reducing dependence on American companies and technology ecosystems. That pushes governments and regional operators toward AI capacity they control directly. We expect this sovereignty-driven segment to keep growing even if the broader hyperscaler spending cycle cools.

Table of Contents

  • Highlights - slide 2
  • Report summary - slide 3
  • Market Opportunity - slide 5
  • Sovereign vs. commercial: where the vendor mix differs - slide 13
  • Outlook for Sovereign AI - slide 25
  • About MTN Consulting - slide 26

Coverage

Organizations mentioned:

  • Adani Group
  • ADQ
  • Aker ASA
  • Alibaba
  • Amazon / AWS
  • AMD
  • Arista
  • Backblaze
  • Baidu
  • Bell Canada
  • Bpifrance
  • Broadcom
  • BUZZ HPC (HIVE Digital)
  • Centuria Capital Group
  • Ciena
  • Cisco
  • Cohere
  • Colt
  • CoreWeave
  • Corning
  • Dell
  • Deutsche Telekom
  • Digital Realty
  • EDF
  • Eni
  • Equinix
  • EuroHPC JU
  • EXA Infrastructure
  • Firmus Technologies
  • Fujitsu
  • G42
  • Google
  • GS Group
  • HUMAIN
  • Hypertec
  • Infinera
  • Ireland CRU
  • IT4LIA / CINECA
  • Italy Ministry of University and Research (MUR)
  • Keppel
  • Khazna
  • Kioxia
  • Lightstorm
  • Marvell
  • MERLIN Edged
  • Meta
  • Micron
  • Microsoft
  • Mistral AI
  • Mubadala
  • New York (Governor’s office)
  • Nokia
  • Nscale
  • NVIDIA
  • OpenAI
  • Oracle
  • PJM Interconnection
  • ResetData
  • SanDisk
  • Scality
  • Schneider Electric
  • Schwarz Digits
  • Singtel
  • SK Group
  • SK Hynix
  • SoftBank
  • Start Campus
  • Tata Communications
  • TikTok / ByteDance
  • Vantage Data Centers
  • VAST Data
  • Vertiv
  • Wall Street Journal
  • WEKA

Figure & Charts

  • Figure 1 (Slide 5): Sovereign AI Deal Volume by Quarter, 1Q2025-2Q2026 - All Tracked vs. Sovereignty-Qualifying
  • Figure 2 (Slide 7): Sovereignty-Qualifying (Tier A) Share of Tracked AI Infrastructure Spend, by Country
  • Figure 3 (Slide 11): Ownership Split of Tracked AI Infrastructure Spend, 2025-1H2026