查看購物車
  • English
  • Japanese
  • Korean
封面
市場調查報告書
商品編碼
2108421

中國人工智慧產品正向新興市場擴張

China's AI Bundle is Heading for Emerging Markets
出版日期: | 出版商: MTN Consulting, LLC | 英文 12 Pages | 訂單完成後即時交付

價格

中國無需贏得最先進模型的競爭來改變人工智慧基礎設施的供應商。中國真正需要的是成為150多個無法自主研發尖端模型的國家的人工智慧基礎設施預設供應商。這場競爭的勝負並非取決於基準測試得分，而是資金籌措、基礎建設以及與各國的關係。十多年來，中國一直透過「一帶一路」計劃及其數位化版本「數位絲綢之路」來建構這些關係。

對於基礎設施供應商而言，關鍵在於這並非一種全新的策略。通訊業早已採用類似的策略。華為憑藉捆綁式、價格主導的條款以及融資方案進入新興市場，並在該市場站穩腳跟長達15年之久。如今，華為在非洲、亞洲和拉丁美洲的大片區域佔主導地位，但在北美市場卻幾乎沒有立足之地。西方供應商在這些市場失敗的原因並非技術實力不足，而是資金籌措條款、捆綁式銷售和長期承諾方面存在劣勢。而現在，在這些國家，人工智慧基礎設施的建設正在如火如荼地進行中。 MTN顧問公司在其「供應商佔有率」系列報告中追蹤了各供應商的市場地位變化。

目前，四個層面正在同時發生變化，且變化速度差異巨大。在模型層面，結構性變革已經發生。阿里巴巴的Qwen系列模型在Hugging Face平台上的累計下載量已突破10億次，佔全球開源模型下載量的一半以上。在OpenRouter平台，美國模型處理的代幣比例將從2025年6月的約70%下降到2026年中的約30%。在計算層面，中國國內的變化顯著，但海外擴張幾乎尚未開始。華為預計其Ascend業務的銷售額將從2025年的75億美元成長到2026年的約120億美元，但這全部來自中國國內市場。同時，輝達目前尚未在中國累計任何資料中心銷售額。 Ascend目前幾乎沒有已確認的出口合約。在記憶層面，全球嚴重的供不應求為CXMT和YMTC等中國供應商創造了新的機遇，而這些機會在18個月前並不存在。在中國最具優勢的電力領域，預計2025年，中國企業將從60多個國家獲得366吉瓦時的新增海外儲能訂單，年增144%。這一成長不會受到出口限制的影響。

對競爭的影響因行業而異。輝達對中國市場的依賴眾所周知，但其在新興市場的佈局卻鮮為人知。美光、三星和SK海力士正將產能轉向用於人工智慧加速器的高頻寬記憶體（HBM），導致中端DRAM市場供不應求，而中國製造商正進軍這一領域。 AWS、微軟Azure和Google雲端將面臨競爭對手的挑戰，這些競爭對手在這個市場中已與政府建立了長達十年的關係。儘管該市場在全球整體規模較小，但卻至關重要。戴爾、HPE、思科和Arista將在新興市場資料中心建置領域與華為競爭，因為華為能夠提供網路設備、伺服器和儲存的打包解決方案，並附帶競爭對手難以匹敵的資金籌措方案。特斯拉能源、Fluence、施耐德電機和Vertiv也在電力和儲能領域與中國企業競爭，中國供應商實力雄厚，目前不受出口限制。

總部位於歐洲和北美的供應商不應將來自中國企業的競爭視為遙遠的未來風險。最相關的先例是通訊業。儘管華為已基本從北美消失，但它在世界許多地區仍佔主導地位。這是因為美國能夠在其自身及其部分盟友境內將華為排除在外，卻無法在其他地方實施類似措施。許多參與「數位絲綢之路」的國家都處於美國的勢力範圍之外。在這些國家運營的供應商需要預見競爭，包括資金籌措方面的競爭，並在人工智慧基礎設施在這些市場全面發展之前製定應對措施。

本報告檢視了中國企業在新興市場人工智慧基礎設施領域的存在情況，總結了中國企業在人工智慧模型、運算、記憶體和電力/能源儲存方面的競爭力，華為等公司與西方供應商之間的競爭結構，以及資金籌措和捆綁產品對市場佔有率的影響。

目錄

  • 概要
  • 介紹
  • 通訊業已經發生的事情
  • 以層次為基礎的分析：新興公司與受影響公司
  • 從基礎設施融資到人工智慧驅動的合約
  • 中國具有競爭力的領域和不具競爭力的領域。
  • 對位於歐洲和美國的供應商的影響
  • 結論
  • 關於 MTN 顧問公司

調查範圍

發布刊物的機構：

  • Alibaba
  • Amazon
  • AMD
  • Anthropic
  • Arista
  • ByteDance
  • Broadcom
  • BYD
  • CATL
  • Cisco
  • CXMT
  • Dell
  • Ericsson
  • Fluence
  • Google
  • H3C
  • Hithium
  • HPE
  • Huawei
  • HyperStrong
  • Indosat Ooredoo Hutchison
  • Inspur
  • Juniper
  • Knowledge Atlas Technology
  • Meta
  • Micron
  • Microsoft
  • Moonshot
  • Nokia
  • Nvidia
  • OpenAI
  • Samsung
  • Schneider Electric
  • SK Hynix
  • Supermicro
  • Sungrow
  • Telkom
  • Tencent
  • Tesla Energy
  • Vertiv
  • YMTC
  • Zhipu
  • ZTE
Product Code: GNI-10082026-1

China does not need to win the frontier model race to change who supplies AI infrastructure. It needs to become the default vendor to the 150-plus countries that will never train a frontier model of their own. That race will be settled by financing, physical infrastructure and relationships, not by benchmark scores. China has been building those relationships for over a decade under the Belt and Road Initiative and its digital arm, the Digital Silk Road.

For infrastructure vendors the important point is that this is not a new approach. It is the telecom approach. Huawei entered emerging markets on financed, bundled, price-led terms and stayed for fifteen years, and it now holds a leading position across much of Africa, Asia and Latin America while remaining effectively absent from North America. Western vendors did not lose that business on technology. They lost it on financing, bundling and patience, in the same countries now being wired for AI. MTN Consulting tracks the resulting vendor positions in its Vendor Share report series.

Four layers are moving at once and at very different speeds. At the model layer the shift has already happened. Alibaba’s Qwen family has passed one billion cumulative downloads on Hugging Face and accounts for more than half of global open-weight downloads. On OpenRouter, US-origin models fell from about 70% of tokens processed in June 2025 to about 30% by mid-2026. At the compute layer the shift is real at home, but barely started abroad. Huawei has guided to around $12bn of Ascend revenue in 2026, up from $7.5bn, all domestic, while Nvidia now books no data center revenue from China at all. Confirmed Ascend export contracts remain close to none, for now. At the memory layer, a severe global shortage has opened a door for Chinese suppliers CXMT and YMTC that did not exist eighteen months ago. At the power layer, where China is strongest, Chinese firms booked 366 GWh of new overseas energy storage orders in 2025, up 144%, across more than 60 countries, with no export-control exposure at all.

The competitive impact is uneven. Nvidia’s China exposure is common knowledge, but its emerging-market exposure is not. Micron, Samsung and SK Hynix have shifted capacity towards high-bandwidth-memory for AI accelerators, which leaves them short in the mid-tier DRAM market. Mid-tier DRAM is where Chinese suppliers are entering. AWS, Microsoft Azure and Google Cloud face a rival that is small worldwide but has a decade of government relationships in the markets that matter here. Dell, HPE, Cisco and Arista face Huawei in emerging-market data center builds, where it can bundle networking, servers and storage with financing options difficult to match. Tesla Energy, Fluence, Schneider Electric and Vertiv compete in power and energy storage, where Chinese suppliers are strong and where no export controls currently apply.

Western-anchored vendors should stop treating Chinese competition in these markets as a remote risk. Telecom is the closest precedent. Huawei ended up absent from North America but leading across much of the rest of the world. That’s because the US could enforce exclusion at home, and with a handful of allies, but nowhere else. Most Digital Silk Road countries sit outside that reach. Vendors selling into them should expect to compete on financing terms, and should work that out before the AI buildout starts in these markets.

Table of Contents

  • Summary
  • Introduction
  • We have seen this before in telecom
  • Layer by layer: who is coming, and who is exposed
  • From infrastructure loans to AI contracts
  • Where China is competitive, and where it is not
  • What this means for Western-anchored vendors
  • Conclusion
  • About MTN Consulting

Coverage

Organizations mentioned:

  • Alibaba
  • Amazon
  • AMD
  • Anthropic
  • Arista
  • ByteDance
  • Broadcom
  • BYD
  • CATL
  • Cisco
  • CXMT
  • Dell
  • Ericsson
  • Fluence
  • Google
  • H3C
  • Hithium
  • HPE
  • Huawei
  • HyperStrong
  • Indosat Ooredoo Hutchison
  • Inspur
  • Juniper
  • Knowledge Atlas Technology
  • Meta
  • Micron
  • Microsoft
  • Moonshot
  • Nokia
  • Nvidia
  • OpenAI
  • Samsung
  • Schneider Electric
  • SK Hynix
  • Supermicro
  • Sungrow
  • Telkom
  • Tencent
  • Tesla Energy
  • Vertiv
  • YMTC
  • Zhipu
  • ZTE

Figure & Charts

  • Figure 1: China’s overseas development finance to the ICT sector, 2008-2024
  • Figure 2: Huawei’s annualized sales to telcos comfortably ahead of all rivals (U$B)
  • Figure 3: Weekly token consumption on OpenRouter, Chinese-origin vs US-origin models
  • Figure 4: Huawei Ascend revenue against Nvidia’s data center revenue from China
  • Figure 5: Top recipients of China ICT-sector development finance, cumulative 2008-2024
  • Table 1: Chinese AI challengers by layer