
羅馬尼亞IT公司Expert IT計劃興建AI資料中心,首期投資逾800萬歐元,預計今年底前啟用;同時國內多地也正推動多項AI與資料中心計畫。
羅馬尼亞正積極擴充AI基礎設施以提升東南歐產業地位。IT公司Expert IT計畫在雷金興建AI資料中心,首期投資逾800萬歐元;此外,國內亦有多起大型資料中心與高效能運算設施建設計畫推動中,但穩定供電與電網承載力成重要挑戰。
AI-generated summary
羅馬尼亞近年積極擴充AI基礎設施,盼提升在東南歐區域的AI產業地位。
(中央社記者劉郁葶布拉格11日專電)近年來,羅馬尼亞積極建設AI相關基礎設施。羅馬尼亞IT公司Expert IT計劃興建AI資料中心,首期投資逾800萬歐元(約新台幣2.86億元),預計今年底前啟用,主要用於AI模型開發。由於AI模型訓練與運算耗電量大,如何確保穩定供電,成為推動相關計畫的重要挑戰。
羅馬尼亞近年積極擴充AI基礎設施,盼提升在東南歐區域的AI產業地位。根據「每日新聞匈牙利」(Daily News Hungary)的引述,羅馬尼亞IT公司Expert IT表示,計畫在羅馬尼亞特蘭西瓦尼亞地區(Transilvania)雷金(Reghin)興建AI資料中心。
Expert IT執行長法爾卡什(Mihai Farcaș Dan)說,此次建立的AI資料中心首期投資設備為7台伺服器,配置56顆高效能NVIDIA B300圖形處理器(GPU),為今年同類型採購案中規模最大的一項,另外涵蓋供電與冷卻等資料中心基礎設施。公司已討論擴充計畫,未來可能再增加512顆GPU,相關伺服器預計今年投入運作,將用於開發大型AI模型。
此計畫是Expert IT在雷金推動的整體開發案一環。Expert IT在12年前開始規劃投資,近3至4年則逐步將AI列為發展重點。整體開發案規模約1000萬歐元,基地面積約1.7公頃。
Expert IT表示,醫療是AI資料中心未來可能應用的主要領域之一,公司正考慮與羅馬尼亞診所及大學合作,開發可協助醫病溝通、支援醫療人員,及分析醫學影像的AI系統。相關技術未來可望應用於腫瘤及心臟疾病等領域,透過分析大量資料與醫學影像,協助醫師辨識疾病特徵及加速診斷。
除了雷金,羅馬尼亞其他地區也有多項AI及資料中心計畫正在推動。美國科技公司DriverAI與羅馬尼亞企業集團Leviatan Group合作在克盧日縣(Cluj County)建設「量子人工智慧資料中心」,第一階段投資額約10億美元,主要功能為AI模型訓練與運算。不過,該計畫目前仍處於準備階段,因鄰近土地業者提出異議而涉及法律爭議。
穆列什縣(Mureș County)議會與工業園區另規劃一座資料中心,預計投入約4500萬歐元,興建約1700平方公尺的設施,提供資料儲存、處理、伺服器託管及AI運算服務。據報導,相關計畫已取得所需資金。
此外,位於特爾古穆列什(Târgu Mureș)的一所醫藥與科技大學也正推動高效能運算基礎設施,總經費約1120萬歐元,將用於AI、機器學習、大規模資料分析、通訊安全及網路攻擊偵測等研究。
據報導,羅馬尼亞正在規劃「黑海AI超級工廠」(Black Sea AI Gigafactory),預計在切爾納沃德(Cernavodă)及多伊切什蒂(Doicești)興建相關設施,最終可配置超過10萬顆AI處理器。
不過,隨著AI資料中心規模擴大,電力供應與電網承載能力也成為羅馬尼亞發展相關基礎設施的重要課題。AI資料中心除了需要大量電力進行大型模型訓練與推論,也必須長時間維持伺服器運作,並仰賴冷卻系統控制設備溫度,因此對供電穩定性與能源成本的要求都相當高,也成為資料中心能否順利擴張的關鍵。
AI outlook — possibilities, not facts
Expert IT AI資料中心於今年底前啟用
Likely · Within months

Billionaire Musk admitted at the Tesla Owners Conference that he did not understand semiconductor manufacturing, and revealed that he had suffered from "artificial intelligence psychosis" many times this year after watching music videos produced by AI. He had previously predicted that China would overcome chip manufacturing constraints within two to three years.
On October 8, Huang Kunming, Secretary of the Guangdong Provincial Party Committee, went to Shenzhen Tencent headquarters for investigation. Ma Huateng, Chairman and CEO of Tencent, and Yao Shunyu, 28-year-old Chief AI Scientist of Tencent, attended the symposium together, demonstrating the determination of Guangdong and Tencent to fully develop artificial intelligence and empower advanced manufacturing.

The data center of Russian technology giant Yandex was attacked by drones again on the 11th, and its facilities in Vladimir Oblast were shut down. This is the third large-scale Yandex data center to be shut down due to attacks in the past few days.

On Thursday, Trump warned on Truth Social that anyone who refuses to use "superintelligence" and continues to use "artificial intelligence" will be regarded as an "enemy" of the White House. Experts analyzed that the move was aimed at boosting the economy and maintaining its leading position in technology, but the new name also raised public concerns about threats to security and employment.

US President Donald Trump seeks to ban the word 'artificial' from 'artificial intelligence,' arguing it makes the technology sound 'fake,' while the White House claims the new terminology better reflects AI's advancing capabilities and limitless opportunities. In contrast, China's term for AI emphasizes a human-centered approach.
Artificial intelligence has transformed from being able to talk to being able to do things. The autonomous actions of intelligent agents have triggered frequent security incidents around the world. OpenAI has suspended model training. Countries such as the United Kingdom and South Korea have reported similar incidents. Chaos such as AI face-changing fraud has occurred in China. Cheng Ying of the China Academy of Information and Communications Technology pointed out that the risk has escalated from saying the wrong thing to doing the wrong thing. my country is building a full-chain AI security management system through institutional supply, technical protection and international cooperation. As of the end of August, a total of 1,112 generative AI services have been registered, and more than 14,000 illegal products have been dealt with.