(#105) 🍏 Apple has reached peak smartphone; How TikTok works
Google is back in the XR space
Knowledge Partner: ENGIE Romania
Doreşti predictibilitate, pe termen lung, a costurilor cu energia electrică şi un business sustenabil?
Descarcă gratuit ebook-ul despre Green PPA, creat de ENGIE, lider mondial în livrarea de energie verde prin contracte de tip PPA*.
Află informații utile privind modul în care poți obține energie pe termen lung, din surse regenerabile. Ebook-ul te va ghida, pas cu pas, prezentând, în detaliu, beneficiile unui astfel de acord şi recomandările ENGIE pentru o contractare de tip Green PPA avantajoasă, cu zero investiții inițiale de capital.
Descoperă care sunt avantajele oferite de ENGIE în colaborarea prin contracte de tip Green PPA şi principalele strategii care au ca scop atenuarea riscurilor financiare, operaționale şi de piață asociate încheierii unui parteneriat Green PPA.
Descarcă gratuit ebook-ul aici!
*Conform BloombergNEF: Corporate Clean Power Buying Grew 12% to New Record in 2023, According to BloombergNEF | BloombergNEF (bnef.com)
Google is back in the XR space
1/ Google’s return to the VR and AR market with Android XR marks an interesting shift, especially after stepping back from this space a few years ago. The new platform combines VR and AR under a unified ecosystem for developers, bolstered by strategic partnerships with Qualcomm and Samsung. This collaboration ensures a solid technological foundation and hints at Google’s intent to leverage existing hardware innovation to accelerate adoption. With the first devices expected to ship next year, Android XR positions itself as a challenger to Apple’s VisionOS, though its success will hinge on whether it can deliver practical utility beyond what current offerings provide.
2/ What sets Google apart here is its focus on integrating AI-powered features into the XR experience. Unlike Apple’s hardware-centric approach, Google is emphasizing software capabilities like live translation and real-time object recognition. These features align with Google’s strength in AI and could make XR devices more useful in day-to-day life. However, while these applications sound compelling on paper, they rely on hardware that is still too bulky, expensive, or limited to achieve mass-market appeal.
3/ The timing of this announcement is notable, as the entire XR space remains in what can be called a “winter” period. Consumer traction for both VR and AR has been lukewarm, with Apple and Meta struggling to move beyond niche markets like gaming, fitness, and industrial use cases. Despite significant progress, the hardware still hasn’t reached the necessary combination of display quality, weight, and affordability to appeal to a broader audience. For Google, offering a platform to OEMs is a logical move, but without clear consumer demand, the ecosystem might face a slow start.
4/ This move appears to be more of a strategic marker for the future rather than an immediate game-changer. Google understands that XR’s long-term potential, particularly augmented reality with AI overlays, could transform how people interact with the real world. However, until hardware catches up and use cases become more compelling, XR will likely remain in its experimental phase. For now, Android XR represents a bet on the technology’s eventual maturity, while serving as a signal to the industry that Google intends to play a major role in shaping that future. LINK
🚗 General Motors Calls quits its robotaxi dream (ie. Cruise)
1/ General Motors is ending its ambitious robotaxi project, a $50 billion bet led by CEO Mary Barra, citing high costs and limited returns. The decision marks a major strategic shift as GM folds Cruise, its autonomous driving subsidiary, back into its core operations. Cruise once rivaled Google’s Waymo in the race for self-driving taxis but struggled with scaling challenges and regulatory backlash, particularly following a pedestrian accident that drew intense scrutiny.
2/ Instead of pursuing robotaxis, GM will focus on enhancing SuperCruise, its advanced assisted-driving feature, and integrating Cruise’s expertise to eventually develop fully autonomous features for consumer vehicles. The company sees this path as a more realistic return on investment, abandoning dreams of transforming itself into a multi-platform mobility giant.
3/ GM’s retreat highlights the broader struggle in the self-driving car industry, where even significant players like Tesla and Waymo face hurdles. The decision will save GM over $1 billion annually, but it raises questions about missed opportunities and whether Cruise could have achieved profitability with more time and investment.
4/ Ultimately, GM’s move signals a return to its roots as a carmaker, abandoning the ambitious but costly pursuit of mobility-as-a-service. While Waymo and Tesla forge ahead, GM’s pullback proves once again the difficulties in achieving reliable and scalable autonomous driving technology.
I am skeptical in general about GM’s future. LINK
🚀 Anduril is the future of war (together with Palantir)
1/ Anduril’s release of the Lattice SDK marks a major step in its mission to become the “operating system” of modern defense systems. The SDK allows third-party developers to build applications that operate on Anduril’s Lattice platform, which integrates data from various sensors and systems. By providing open APIs and tools, Anduril aims to decentralize military technology development, enabling a wider range of solutions while improving situational awareness and operational efficiency.
2/ Lattice is Anduril’s answer to the traditional, fragmented approach of defense systems, where software is tied rigidly to specific hardware. By unifying these capabilities, Lattice functions as a scalable network for decision-making, sense-making, and task automation, even in environments with disrupted communications. This positions Anduril as a platform company, much like Windows or iOS, but for the defense sector, enabling both Anduril’s systems and third-party tools to interoperate seamlessly.
3/ Finally, the recent collaboration between Anduril and Palantir highlights complementary roles: Anduril’s Lattice provides real-time tactical data for frontline operations, while Palantir processes that data for strategic decision-making. This partnership reflects a broader trend toward open, interoperable platforms in defense, moving away from centralized mega-projects. As the Pentagon embraces these modern approaches, Anduril’s platform-driven model could reshape military operations and unlock new opportunities for tech startups in the defense sector. LINK
Chat GPT goes o1 PRO. What does it mean?
1/ OpenAI’s ChatGPT Pro is a new $200/month subscription tier targeting power users who require advanced capabilities and extensive usage of OpenAI’s top models. It includes unlimited access to o1, OpenAI’s powerful reasoning model, as well as additional compute-intensive features that improve problem-solving accuracy. The launch reflects OpenAI’s efforts to balance computing costs while capturing value from users who rely on AI for complex tasks. Despite competition from rivals like Claude, OpenAI’s superior product experience, particularly its native Mac app, has helped it maintain a lead in consumer AI mindshare.
2/ The o1 model represents a significant leap in inference capabilities, as it improves answer accuracy by using more computational resources and reasoning iterations during output generation. This shift marks a new paradigm where AI performance depends not just on model size but also on dynamic inference computing. While o1’s capabilities are impressive, they come at higher costs and slower speeds, making them ideal for specialized use cases rather than general tasks. The premium pricing reflects its differentiation, offering users a glimpse into AI that functions as an agent capable of performing autonomous, complex reasoning tasks.
3/ OpenAI also introduced Sora, its video-generation model, which is now available to Pro and Plus users. Sora can generate realistic videos with improved speed and quality, though issues like object permanence remain a challenge. The launch highlights OpenAI’s strategic push to justify higher-tier pricing by offering cutting-edge tools like Sora Turbo, which delivers higher resolution and greater monthly usage quotas. Overall, these announcements underscore OpenAI’s move toward premium AI offerings, targeting users who can extract significant value from its advanced capabilities and computational power.
I remain bullish on OpenAI to deliver exactly what consumers want. LINK
Algorithms are the new nukes. Case study: TikTok
Having been involved so far in four events (Gaza, Romania’s elections, and ongoing trouble and misinformation in Georgia andthe Republic of Moldova). So, it starts with their paper published two years ago where the company presented the power of its new algorithm.
Summary of the paper:
1/ The paper introduces Monolith, a real-time recommendation system designed to address challenges in large-scale, industrial recommendation settings, such as data sparsity, non-stationary user behavior, and the need for immediate feedback. Unlike traditional deep learning frameworks like TensorFlow, which separate batch training from real-time inference, Monolith combines online training and serving to allow models to adapt instantaneously to user interactions. This system is particularly suited for businesses reliant on time-sensitive user feedback, such as short-video platforms or ad rankings.
2/ At the core of Monolith’s innovation is its collisionless embedding table, built using a Cuckoo Hash Map to ensure unique, non-colliding embeddings for features, unlike common hashing methods that cause quality deterioration over time. The table also incorporates optimizations like feature frequency filtering and expiration mechanisms to limit memory usage without sacrificing model accuracy. These approaches help avoid the memory bloat typically caused by sparsely used or outdated features, ensuring efficient and scalable parameter management.
2/ The system also emphasizes online training by continuously integrating user feedback into the model in near real-time. This is achieved through a streaming engine that processes user actions and feature data in tandem, enabling frequent updates to the model’s parameters. Monolith employs an incremental parameter synchronization mechanism to keep the online serving model updated efficiently, balancing computational overhead and latency. Sparse parameters are updated frequently, while dense parameters are synchronized less often, reducing strain on resources.
4/ For fault tolerance, Monolith incorporates a snapshot-based recovery system, which stores model states periodically. The system finds an optimal trade-off between reliability and computation costs by allowing longer snapshot intervals (e.g., daily) while ensuring performance degradation remains negligible. Experiments demonstrated that models using Monolith consistently outperform systems with hash-based collisions and batch-only updates, achieving higher AUC (Area Under Curve) scores, even in production environments.
5/ In conclusion, Monolith addresses key limitations of traditional recommendation systems by combining a collisionless embedding table, real-time online training, and efficient parameter management. It demonstrates significant improvements in model performance and scalability, making it a robust solution for large-scale, real-time recommendation systems. LINK
🍏 Apple has reached peak smartphone
1/ Apple’s plan to introduce thinner iPhones and foldable devices highlights the company’s attempt to reignite growth amid stagnating smartphone sales. For years, Apple has relied on incremental upgrades, like faster chips and better cameras, but this strategy no longer drives significant user upgrades. The new designs include a sleeker iPhone to appeal to budget-conscious users and foldable devices, including a compact foldable iPhone and a larger model resembling a laptop with a nearly 19-inch screen. These innovations, though evolutionary rather than revolutionary, aim to rejuvenate Apple’s hardware lineup in a market where smartphones and laptops have reached maturity.
2/ The move comes as the industry hits the "peak smartphone" and laptop plateau, with limited room for growth. Users are holding onto devices longer as hardware advancements feel less compelling, and the market for foldable devices remains niche, accounting for just 1.5% of total smartphone sales in 2024. Apple’s delay in entering the foldable segment reflects its high standards for weight, durability, and energy efficiency, challenges that competitors like Samsung and Huawei have struggled to overcome. Despite these hurdles, Apple’s focus on foldables signals that it sees hardware design changes as essential to driving the next wave of upgrades.
3/ Beyond smartphones, Apple is also exploring new growth areas like the Vision Pro, its mixed-reality headset. However, Vision Pro sales have been sluggish due to its steep price tag and limited ecosystem, prompting Apple to consider integrating iPhones to power future iterations. This broader strategy shows Apple’s need to diversify and innovate as the traditional smartphone market plateaus. While foldable devices and thinner designs may not revolutionize the industry, they are a necessary evolution to maintain Apple’s leadership in a mature tech landscape. LINK
The feedback from the readers for my newest book has been amazing…so far 🙃
Through 28 chapters I covered three parts: (1) Strategy, (2) Innovation & Growth, and (3) Generative AI.
See a full sample - the chapter on Network Effects. - click HERE
More about the book:
Buy a hardcopy on EMAG
Soon to come to Amazon and Gumroad.
Thank you for being an onStrategy reader!
Interesting newsletters 💡
