BMW 2025 Earnings Review: When a Luxury Automaker Starts Redefining Itself Through Software

BMW has long been one of the most important bellwethers in the global premium automotive industry. Its financial reports are not only highly informative, but also unusually polished in the way they connect brand strategy, technology roadmaps, product plans, and capital allocation. For me, that is exactly what makes BMW one of the most enjoyable automakers to follow: its reports rarely read like a collection of dry numbers. Instead, they often feel like a carefully constructed statement about how the company sees the next decade unfolding.

That is especially true for the investor presentation BMW released in January 2026 under the theme “Rethinking Premium Individual Mobility for the Next 100 Years.” Together with the company’s March 2026 disclosure of its 2025 annual results, the message becomes quite clear. BMW is trying to do two things at once: preserve the operating discipline and brand strength that made it one of the world’s premier luxury carmakers, while simultaneously rebuilding itself around software-defined vehicles, centralized computing, electrification, and digital services. Read together, these two documents are not just a review of how much BMW earned in 2025. They are a window into how the company is trying to balance its legacy with its future.

(This article is for OT's self-learning purposes, but feel free to share and repost with a link to the original source.)

Qualcomm FY2025 Overview: A Pivotal Year for AI and Multi-Platform Transformation

In February 2026, Qualcomm released its Fiscal Year 2026 first-quarter results (Q1 FY26), covering the period from September 29, 2025, to December 28, 2025. Since Qualcomm does not publish a standard calendar-year financial report, the following analysis is based on OT’s consolidation of the four quarterly reports from Fiscal Year 2025 (September 30, 2024, to September 28, 2025) .

Viewed through this consolidated perspective, Qualcomm’s 2025 performance may initially appear contradictory—showing both growth and decline. However, a closer examination reveals that it is, in fact, a year marked by accelerated transformation. Total revenue reached $44,284 million (+14% YoY), while gross profit rose to $24,546 million (+12.07% YoY), maintaining a strong gross margin of 55.43%. At the same time, GAAP net income declined sharply by 45% to $5,541 million. While this divergence may suggest weakening fundamentals at first glance, it is better understood as a classic case of accounting distortion rather than operational deterioration.

(This article is for OT's self-learning purposes, but feel free to share and repost with a link to the original source.)

NXP Q4 2025 Earnings: A Cyclical Recovery Story—and a Strategic Shift Toward the Edge

In early 2026, NXP Semiconductors released its financial results for the fourth quarter and full-year 2025, alongside a comprehensive investor presentation. NXP delivered a scorecard that signals a cyclical bottoming and recovery. More importantly, hidden beneath the financial metrics is a clear message: the company is undergoing a profound structural evolution.

The fourth-quarter performance has already revealed several positive signals. Revenue returned to over $3.3 billion, with both quarter-over-quarter and year-over-year growth turning positive, indicating that demand is at least no longer deteriorating. Even more noteworthy is the cash flow. Non-GAAP free cash flow approached $800 million, accounting for more than 20% of total revenue—a level rarely seen before a full cyclical recovery in the semiconductor industry. To some extent, this demonstrates that NXP is not merely "waiting for the cycle to return," but has found a way to maintain stable output across different economic phases.

This article will guide you through NXP's true strategic layout in the face of the massive AI wave, dissecting it across three dimensions: financial performance, business structure, and long-term strategy.

(This article is for OT's self-learning purposes, but feel free to share and repost with a link to the original source.)

Beyond the Cycle: Deciphering Infineon’s Zonal Architecture and AI Power Strategy in 2026

On February 4, Infineon released its financial results for the fourth quarter of fiscal year 2025, along with its full-year performance. The results show that, while Infineon maintained relatively stable revenue throughout 2025, profitability remained under pressure due to cyclical softness in key end markets and continued high investment intensity.

Infineon’s business is structured across four main segments—Automotive (ATV), Power & Sensor Systems (PSS), Green Industrial Power (GIP), and Connected Secure Systems (CSS). Among these, Automotive continues to be the primary revenue contributor. However, Power & Sensor Systems (PSS) stands out as a key growth driver, supported by strong demand from AI-related applications and power solutions for data centers.

Beyond the financial overview, this report also provides a deeper look into Infineon’s strategic direction. The company outlines its future vision across several key domains, including Automotive, Electromobility, Software-Defined Vehicles, Green Industrial Power, Power & Sensor Systems, and Connected Secure Systems—highlighting how it is positioning itself for long-term growth at the intersection of electrification and digitalization.

(This article is for OT's self-learning purposes, but feel free to share and repost with a link to the original source.)

快讀2024年諾貝爾物理學獎 - 「使用物理學訓練人工神經網路」

2024年諾貝爾物理學獎頒發給約翰·J·霍普菲爾德及傑弗瑞·E·辛頓，以表彰他們在推動人工神經網路機器學習領域中的基礎性發現與發明。

對於今年的頒獎結果，OT確實感到有些驚訝，因為諾貝爾物理學獎歷來多數是表彰基礎物理的重大發現，而類神經網路則偏向於工程應用。將物理學應用於類神經網路的過程，對OT來說這樣的關聯有些牽強，彷彿是因應近年來人工智慧風潮而頒發的獎項。

首先，我們必須探討「人工智慧」是否應該被視為一門「科學」。科學的核心在於相同條件下進行相同的步驟能夠得到一致的結果，即具備可驗證性。然而，目前的人工智慧並不完全符合這一標準，甚至許多AI系統的結果生成過程無法完全被重複驗證。雖然隨著大數據的發展，AI在圖像識別和語音識別領域取得了巨大進展，但從科學方法的角度來看，AI尤其是基於深度學習的系統仍然面臨兩大挑戰：

1. 可重現性問題：人工智慧，尤其是深度神經網路，依賴於大量數據的訓練，結果往往受到初始條件、隨機因素及數據集特徵的影響。即使使用相同的神經網路架構，不同的訓練過程可能導致不同的結果。這與科學實驗中所要求的重複性並不一致，因此許多批評者認為，這樣的技術難以符合傳統意義上的「科學」。
2. 不可解釋性問題：當前的深度學習模型大多是「黑箱」過程，難以清楚解釋其內部如何產生特定結果。這導致我們無法輕易追蹤模型的決策邏輯，進而影響結果的可信度。這一點在醫療診斷或司法系統等應用中尤為引發關注，因為結果的可解釋性對這些領域至關重要。

儘管如此，人工智慧領域仍然結合了許多科學知識，如概率統計、優化算法、信息理論等，這些無疑都是基於嚴格的數學基礎。尤其在影像識別和語音識別等應用中，隨著大數據的擴展，AI模型已能在大規模測試中產生穩定的一致結果。但OT認為這些應用更多依賴數學模型和大數據的輔助，而非基於物理學的核心發現。

OT的想法是將類神經網路與物理學聯繫在一起可能有些牽強。從今年的諾貝爾物理學獎背景資料來看，獎項確實試圖將人工神經網路與物理學中的概念（如自旋模型和能量景觀）進行對比和聯繫。這種聯繫在科學上並非毫無根據，因為神經網路的數學模型與統計物理中的某些模型（如玻爾茲曼機和自旋理論）確實有相似之處。像約翰·霍普菲爾德這樣的物理學家，對這些領域也做出了重要貢獻。然而，這樣的關聯在當前AI熱潮的背景下，可能更多是對於AI技術廣泛應用的認可，也可能在未來有助於解決上述提到的「可重現性問題」以及「不可解釋性問題」。諾貝爾委員會或許希望通過頒發這個獎項，承認AI技術對現代社會的巨大貢獻。

如果真是如此，那麼這個獎項一方面反映了神經網路在物理學中的根源和應用，另一方面也象徵著對當前AI技術的認可。畢竟，這些技術已經深刻改變了我們的生活與科學發展。讓我們一起來仔細閱讀諾貝爾委員會如何詮釋今年的物理學獎。

本圖來自：諾貝爾獎官方網站

本篇文章不僅供OT自我學習使用，也歡迎各位朋友轉載並註明原文網址。

AI改變學習的方式：個人化與多主題深度學習的實踐

人工智慧（AI）在現代學習中的應用日益廣泛，並且正在重塑我們學習知識、解決問題以及掌握新技能的方式。AI技術可以自動化大量的學習過程，提供個性化的學習體驗，並幫助學生和專業人士更有效地理解和應用知識。本文分享AI在現代學習中的一些主要應用，特別是像ChatGPT這樣的語言模型的作用，以及OT在這一年多來（重度）使用後的心得。

從盤古開天到拜訪太陽系，用神話串接STEAM的故事志工之旅

不知不覺女兒進入校園的第一個學年已經結束，暑假也過了一半，時間還過得真快！第一個學年的故事志工服務結束，即將迎接第二個學年的到來，也趁這個空檔整理一下對於故事志工服務的理念。

本篇承接《說故事志工 - 為孩子點亮心中的閱讀燈塔》一文，分享OT對於說故事志工故事安排的策略改變，如何讓傳統神話與想像力、科學素養串接起來。

結構性失業的新挑戰：AI和未來的勞動力市場

經濟學中通常將失業分為三種主要的類型：

• 摩擦性失業：是指因為工人在找尋新的工作崗位或轉換職業時所產生；這種失業通常是短期的，因為工人只需要一段時間來找到新的工作。
• 結構性失業：是指由於產業變動或技術進步導致某些工作崗位消失所引起，例如，當新的技術取代了舊的製造方式，工人可能就會失去工作；這種失業可能需要更長的時間來解決，因為工人需要重新培訓以適應新的工作環境。
• 週期性失業：是指由於經濟周期的波動引起；在經濟衰退期間，需求下降可能會導致企業裁員，從而導致失業率上升。但是在經濟復甦時，企業又會開始招聘員工，失業率就會下降。

(Figure by Midjourney)