New AI Chips from Broadcom and Qualcomm
New AI Chips from Broadcom and Qualcomm
Broadcom 與 Qualcomm 推出新款 AI 晶片
Introduction
Broadcom and Qualcomm are making new computer parts for AI. They want to help data centers work better.
Broadcom 與 Qualcomm 正在開發新的 AI 電腦零件，旨在提升數據中心的運作效率。
Main Body
Broadcom and OpenAI made a new chip. It is called Jalapeño. This chip uses less power and costs less money. It will start working in 2026.
Broadcom 與 OpenAI 研發了一款名為 Jalapeño 的新晶片。這款晶片功耗較低且成本更低，將於 2026 年投入使用。
Qualcomm is making new chips for big computers. Meta will use these chips in 2028. Qualcomm also bought a company called Modular for nearly 4 billion dollars. This helps their software work on many different chips.
Qualcomm 正在開發用於大型電腦的新晶片，Meta 將於 2028 年採用。此外，Qualcomm 以近 40 億美元收購了名為 Modular 的公司，這有助於其軟體能兼容於多種不同的晶片。
Many big companies now make their own chips. They do not want to buy everything from one company. They need more power for AI to work fast.
目前許多大公司開始研發自有晶片，不希望過於依賴單一供應商。他們需要更強大的算力來提升 AI 的運行速度。
Conclusion
Companies are making special chips and software to compete in the AI market.
各公司正研發專用晶片與軟體，以在 AI 市場中競爭。
Vocabulary Learning
⚡ Quick Change: Now vs. Later
Look at how the story talks about time. We use different words to show if something is happening today or in the future.
1. Right Now (Present)
- "Companies are making..."
- "They do not want..."
- "They need..." → Use this for things that are true today.
2. Later (Future)
- "It will start working..."
- "Meta will use..."
💡 Simple Rule: If you see WILL, the action is a promise or a plan for the future (like 2026 or 2028).
Vocabulary for the Real World
- Less power Uses a small amount of electricity.
- Less money It is cheaper.
- Big computers Data centers.
Vocabulary Learning
New Strategies and Infrastructure Growth in the AI Chip Industry
AI 晶片產業的新策略與基礎設施成長
Introduction
Broadcom and Qualcomm have announced major improvements in their artificial intelligence hardware and software. These changes show that the industry is moving toward custom-made chips and a more diverse range of data center equipment.
Broadcom 與 Qualcomm 宣布在人工智慧硬體與軟體方面取得重大進展。這些改變顯示產業正趨向於定制晶片，以及更豐富的數據中心設備選擇。
Main Body
Broadcom and OpenAI have worked together to create 'Jalapeño,' a special chip designed for AI inference. This processor aims to improve energy efficiency and lower the cost of running large language models. Broadcom's CEO, Hock Tan, stated that the first chips will be used by late 2026, with full use expected by 2028. While this chip is meant to work alongside Nvidia's products rather than replace them, it allows OpenAI to rely less on a single supplier.
Broadcom 與 OpenAI 合作開發了名為 「Jalapeño」 的特殊晶片，專為 AI 推理而設計。該處理器旨在提高能源效率並降低運行大型語言模型的成本。Broadcom 執行長 Hock Tan 表示，首批晶片將於 2026 年底前投入使用，預計 2028 年全面普及。雖然此晶片旨在與 Nvidia 的產品協同工作而非取代，但這能讓 OpenAI 減少對單一供應商的依賴。
At the same time, Qualcomm is shifting its focus from smartphones toward the data center market. For example, Meta Platforms will start using Qualcomm's Dragonfly C1000 CPU in 2028. Furthermore, Qualcomm is buying a startup called Modular for about $3.92 billion. Modular creates software that allows AI models to work on different types of chips. This move is intended to give developers more flexibility and challenge Nvidia's dominant CUDA software.
與此同時，Qualcomm 正將重心從智慧型手機轉向數據中心市場。例如，Meta Platforms 將於 2028 年開始使用 Qualcomm 的 Dragonfly C1000 CPU。此外，Qualcomm 正以約 39.2 億美元收購一家名為 Modular 的初創公司。Modular 開發的軟體可讓 AI 模型在不同類型的晶片上運行。此舉旨在為開發者提供更多靈活性，並挑戰 Nvidia 佔主導地位的 CUDA 軟體。
These developments show a general trend where companies want more control over their own technology and supply chains. Other giants like Amazon and Alphabet are also creating their own chips to ensure they have a steady supply. Because the demand for computing power is so high, the industry is moving toward specialized accelerators and open software to make generative AI more affordable and efficient.
這些發展顯示出一個整體趨勢，即公司希望對自身的技術和供應鏈擁有更多控制權。其他巨頭如 Amazon 和 Alphabet 也在開發自有晶片以確保供應穩定。由於對運算能力的需求極高，產業正向專用加速器和開放軟體發展，以使生成式 AI 更加經濟且高效。
Conclusion
The AI chip market is now characterized by the growth of custom chips and the purchase of flexible software to challenge the current market leaders.
目前的 AI 晶片市場特點在於定制晶片的成長，以及透過收購靈活的軟體來挑戰目前的市場領導者。
Vocabulary Learning
🚀 The 'B2 Shift': From Simple Verbs to Strategic Power-Verbs
At the A2 level, you use words like make, change, or get. To reach B2, you need to use precise verbs that describe how something is happening. The text provides a perfect map for this transition.
⚡ Precision Upgrade
Look at how the text replaces basic ideas with professional-grade verbs:
-
Instead of "change" Use "shift"
- A2: Qualcomm is changing its focus.
- B2: Qualcomm is shifting its focus.
- Why? "Shift" implies a strategic move from one position to another.
-
Instead of "make" Use "create" or "design"
- A2: They made a new chip.
- B2: They designed a special chip / created software.
- Why? "Design" shows there was a plan and engineering involved.
-
Instead of "help" Use "allow" or "ensure"
- A2: This helps OpenAI use other companies.
- B2: This allows OpenAI to rely less on one supplier.
- Why? "Allow" describes giving permission or creating a possibility.
🧩 The " Connector Logic"
B2 speakers don't just list facts; they link them to show cause and effect. Notice these three structures from the article:
- "Rather than..." Used to contrast two options (e.g., Work alongside Nvidia rather than replace them).
- "Furthermore..." A sophisticated way to say "also" when adding a new, important point.
- "Because..." Used at the start of a long sentence to explain the reason before the result.
🛠 Quick Guide: Power-Phrases for your Vocabulary
| A2 Phrase | B2 Professional Equivalent | Context |
|---|---|---|
| Big changes | Major improvements | Tech/Business |
| Use it | Rely on | Dependency |
| Main one | Dominant | Market Power |
| Fast growth | Infrastructure growth | Industry |
Vocabulary Learning
Strategic Diversification and Infrastructure Expansion within the AI Semiconductor Sector
AI 半導體領域的戰略多元化與基礎設施擴張
Introduction
Broadcom and Qualcomm have announced significant advancements in their artificial intelligence hardware and software capabilities, signaling a broader industry shift toward custom silicon and diversified data center infrastructure.
Broadcom 與 Qualcomm 宣布在人工智慧硬體與軟體能力方面取得重大進展，標誌著整個產業正向定制晶片與多元化數據中心基礎設施轉型。
Main Body
The collaboration between Broadcom and OpenAI has culminated in the unveiling of 'Jalapeño,' a specialized Application-Specific Integrated Circuit (ASIC) optimized for AI inference. This 'Intelligence Processor' is designed to enhance performance-per-watt and reduce the operational cost per token, thereby addressing the substantial energy requirements of large language models. Broadcom CEO Hock Tan indicated that initial deployment is anticipated by the end of 2026, with full-scale implementation projected for 2028. While this custom silicon is intended to complement rather than replace Nvidia's GPU-accelerated solutions, it represents a strategic effort by OpenAI to mitigate reliance on a single supplier and optimize its full-stack infrastructure.
Broadcom 與 OpenAI 的合作最終推出了 ''Jalapeño''，這是一款專為 AI 推理而優化的專用集成電路 (ASIC)。這款 ''智能處理器'' 旨在提升每瓦性能並降低每個 token 的運營成本，從而解決大型語言模型對能源的巨大需求。Broadcom 執行長 Hock Tan 指出，預計 2026 年底將開始初步部署，而全面實施則預計於 2028 年。雖然這款定制晶片是用來補充而非取代 Nvidia 的 GPU 加速方案，但這代表了 OpenAI 企圖減少對單一供應商的依賴，並優化其全棧基礎設施的戰略努力。
Concurrently, Qualcomm is executing a pivot from its primary smartphone-centric revenue model toward the data center market. This transition is evidenced by the introduction of the Dragonfly C1000 CPU, which Meta Platforms is slated to utilize beginning in 2028. Qualcomm has further expanded its portfolio by securing custom chip agreements with two unnamed hyperscalers. To enhance its software ecosystem, Qualcomm has entered into an all-stock agreement to acquire Modular, a startup specializing in a horizontal software platform that enables AI models to operate across diverse chip architectures. This acquisition, valued at approximately $3.92 billion, is positioned as a strategic countermeasure to Nvidia's proprietary CUDA software, aiming to provide developers with greater flexibility in compute environment deployment.
與此同時，Qualcomm 正將其主要以智慧型手機為中心的收入模式轉向數據中心市場。Meta Platforms 計劃於 2028 年開始使用 Dragonfly C1000 CPU，便證明了這次轉型。Qualcomm 進一步透過與兩家未披露名稱的超大規模雲端業者簽署定制晶片協議來擴展其產品組合。為了強化其軟體生態系統，Qualcomm 達成了一項全股票交易協議，收購 Modular，這是一家專攻橫向軟體平台的初創公司，使 AI 模型能夠在不同的晶片架構上運行。這次收購價值約 39.2 億美元，被視為針對 Nvidia 專有 CUDA 軟體的戰略對策，旨在為開發者在計算環境部署上提供更大的靈活性。
These institutional movements reflect a systemic trend toward vertical integration and supply chain diversification. The emergence of custom silicon from entities such as Amazon and Alphabet, alongside the strategic overtures of ByteDance toward Qualcomm and MediaTek, suggests a collective industry objective to enhance supply chain security. The prevailing market dynamic is characterized by an 'insatiable' demand for computing power, driving a transition where specialized inference accelerators and open software layers are increasingly utilized to optimize the economic and technical efficiency of generative AI deployment.
這些機構的動作反映了一種向垂直整合與供應鏈多元化發展的系統性趨勢。Amazon 與 Alphabet 等實體推出定制晶片，加上 ByteDance 向 Qualcomm 與 MediaTek 採取的戰略接洽，顯示出整個產業有共同目標要提升供應鏈安全。目前的市場動態特點是對計算能力有 ''不飽之欲'' 的需求，驅動了一種轉型，即越來越多地使用專門的推理加速器與開放軟體層，以優化生成式 AI 部署的經濟與技術效率。
Conclusion
The AI semiconductor landscape is currently defined by the proliferation of custom ASICs and the strategic acquisition of cross-platform software to challenge existing market hegemonies.
AI 半導體版圖目前的特點是定制 ASIC 的激增，以及透過戰略收購跨平台軟體來挑戰現有的市場壟斷。
Vocabulary Learning
The Architecture of 'Nominalization' as a C2 Power-Tool
To bridge the gap from B2 to C2, one must transition from describing actions to conceptualizing systems. The provided text is a masterclass in Heavy Nominalization—the linguistic process of turning verbs (actions) and adjectives (qualities) into nouns. This is the hallmark of high-level academic and strategic English, as it allows the writer to pack dense information into the subject position of a sentence.
1. Deconstructing the 'Abstract Subject'
Observe the shift in density here:
- B2 Level: Qualcomm is changing its business model because it wants to move from smartphones to data centers. (Linear, action-oriented).
- C2 Level (from text): "This transition is evidenced by the introduction of the Dragonfly C1000 CPU..."
In the C2 version, the action of "changing" is crystallized into the noun "transition." This noun then becomes the subject, allowing the writer to attribute properties to it (e.g., that it is "evidenced"). This creates a more objective, authoritative tone by removing the human agent and focusing on the phenomenon.
2. Precision through Complex Noun Phrases
C2 mastery requires the ability to string nouns together to create a singular, highly specific technical concept. Analyze these clusters from the article:
"...strategic countermeasure to Nvidia's proprietary CUDA software" "...systemic trend toward vertical integration and supply chain diversification"
The Linguistic Mechanism:
- Strategic countermeasure: [Adjective] [Noun]. It doesn't just say "a plan to fight," but defines the category of the action.
- Supply chain diversification: [Noun] [Noun] [Noun]. This is a 'noun chain.' It compresses a complex process (the act of making a supply chain more diverse) into a single conceptual object.
3. The 'Semantic Weight' Shift
Notice how the text handles the concept of demand:
- *"The prevailing market dynamic is characterized by an 'insatiable' demand..."
Instead of saying "People want more computing power than ever before," the author uses "prevailing market dynamic" and "insatiable demand."
C2 Strategy Tip: To elevate your writing, stop using verbs to describe trends. Instead, identify the noun that represents that trend and make it the anchor of your sentence. Replace "they are integrating vertically" with "the trend toward vertical integration." This shifts the focus from the doer to the strategic reality.