
外媒披露輝達傳出有意採用英特爾的Foveros先進封裝技術,最快可能在2028年下半年展開相關合作。
外媒披露輝達傳出有意在部分AI晶片後段製程中,採用英特爾的Foveros先進封裝技術,最快可能於2028年下半年展開合作,有望緩解台積電產能壓力。
AI-generated summary
台積電目前在全球先進封裝市場穩居龍頭,而英特爾正積極發展Foveros等先進封裝技術。
台積電目前在全球先進封裝市場,穩居龍頭霸主地位。不過,外媒披露,輝達傳出有意採用英特爾的Foveros先進封裝技術,最快可能在2028年下半年展開相關合作。
科技媒體《Wccftech》報導,根據GF Securities分析師Jeff Pu的最新看法,輝達未來可能在部分AI晶片後段製程中採用英特爾Foveros技術。隨著英特爾14A先進製程預計2028年下半年進入量產階段,輝達也可能在相近時間點導入英特爾的先進封裝服務。
事實上,這並非輝達首次傳出考慮採用英特爾技術。今年1月已有相關報導指出,輝達可能在下一代Feynman架構GPU產品中,採用英特爾18A或14A製程處理部分非核心晶片元件,但主要GPU運算晶片仍將依賴台積電先進製程。
此次消息的焦點則轉向Foveros封裝技術。相較於傳統將多個晶片並排連接的方式,Foveros採用3D堆疊設計,能將不同晶片垂直整合,有助於縮小封裝面積,並改善晶片間的資料傳輸效率與能源使用效率。
隨著全球AI運算需求快速成長,先進封裝已成為晶片製造的重要競爭領域。台積電CoWoS技術目前是高效能AI晶片的重要封裝方案,但相關產能持續受到市場高度關注。若輝達未來增加英特爾作為封裝合作夥伴,除了讓英特爾取得更多商機,也可能分散部分封裝需求,緩解台積電的產能壓力。
值得注意的是,英特爾近年積極發展先進封裝業務,包括EMIB、EMIB-T及Foveros等技術,先前也曾傳出Google TPU可能採用其封裝方案。若再成功切入輝達AI晶片供應鏈,將有助於提升英特爾在先進封裝市場的競爭地位。
AI outlook — possibilities, not facts
輝達最快於2028年下半年展開英特爾先進封裝合作
Possible · Within months

Bitdefender, an information security company, discovered that malware called Midnight Mimosa is hidden in the Android mobile phone systems of some Chinese niche brands. It has high permissions and can secretly install apps, modify permissions and remotely execute malicious programs. Thousands of devices in more than 150 countries around the world are affected. It is recommended to abandon the infected mobile phones immediately.

Google Cloud released the Gemini enterprise agent at the Gemini at Work event, which allows employees to delegate tasks through natural language on Gmail, Docs and other platforms. AI will plan its own steps, call the enterprise system and send the results back to the original working environment. It supports cross-device continuous execution and multi-model selection. It is currently in the early testing stage.

OpenAI announced the launch of Ultrafast ultra-high-speed mode for GPT-6.1 Sol. The token generation speed can be up to 8 times that of the standard mode, but the API usage cost has also increased to 6 times that of the standard version, even higher than the flagship Astra model. It is mainly aimed at professional developers and enterprise users. It has been gradually launched in API, Codex and ChatGPT Work.

A study by the Carnegie Endowment for International Peace found that only one in every 30 Chinese researchers engaged in AI research in the United States will return to China in 2025, but the overall number of Chinese researchers in the United States has increased by four percentage points. The study concluded by analyzing the author flow of NeurIPS papers that China has made progress in retaining local talent, while the United States relies on foreign researchers to maintain its lead. Chinese universities have risen significantly in the rankings of top institutions, with Peking University and Tsinghua University surpassing prestigious American schools such as MIT. At the same time, the Chinese government has intensified its restrictions on AI talents leaving the country, as shown by incidents such as the founder of Manus being blocked from leaving the country.

The U.S. government has suspended the acceptance of new and pending permanent labor certification applications submitted by eight technology companies including Microsoft and Adobe for foreign employees. Vice President Vance claimed that the visa system was abusing the visa system to prioritize hiring foreign employees. Labor Secretary Sandlin confirmed that the suspension covers all applications. Microsoft responded that its H-1B applications are mostly used to extend or change the status of existing employees rather than recruit new employees.

GlobalFoundries signed a five-year agreement worth US$2 billion with TSMC to produce silicon interposers for TSMC's CoWoS advanced packaging at its Malta, New York, factory. Production is expected to increase in the first half of 2028 to cope with the lack of advanced packaging production capacity caused by AI chip demand.